support for fp8

* test batch sampler w varying batch lens

* wip

* multipack batchsampler wip

* wip

* fix for prepare data loader to get correct # of steps based on gpues

* lint and clean up

* calculate len estimate

* fix total num steps calc

* add options for dataloader_num_workers and dataloader_pin_memory

* remove gitbook

* support prefetch_factor for dataloader optimization

* fix the kwarg

.github/ISSUE_TEMPLATE/bug-report.yaml

@@ -53,6 +53,13 @@ body:
     validations:
       required: true
 
+  - type: textarea
+    id: config
+    attributes:
+      label: Config yaml
+      description: |
+        Please attach the config yaml!
+
   - type: textarea
     id: possible-solution
     attributes:

.github/workflows/base.yml

@@ -25,6 +25,11 @@ jobs:
             python_version: "3.10"
             pytorch: 2.0.1
             torch_cuda_arch_list: "7.0 7.5 8.0 8.6 9.0+PTX"
+          - cuda: "118"
+            cuda_version: 11.8.0
+            python_version: "3.10"
+            pytorch: 2.1.0
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 9.0+PTX"
     steps:
       - name: Checkout
         uses: actions/checkout@v3

.github/workflows/main.yml

@@ -23,6 +23,12 @@ jobs:
             python_version: "3.10"
             pytorch: 2.0.1
             axolotl_extras:
+            is_latest: true
+          - cuda: 118
+            cuda_version: 11.8.0
+            python_version: "3.10"
+            pytorch: 2.1.0
+            axolotl_extras:
     runs-on: [self-hosted, gpu, docker]
     steps:
       - name: Checkout
@@ -46,9 +52,12 @@ jobs:
           build-args: |
             BASE_TAG=${{ github.ref_name }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}
             CUDA=${{ matrix.cuda }}
+            PYTORCH_VERSION=${{ matrix.pytorch }}
           file: ./docker/Dockerfile
           push: ${{ github.event_name != 'pull_request' }}
-          tags: ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+          tags: |
+            ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+            ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
           labels: ${{ steps.metadata.outputs.labels }}
   build-axolotl-runpod:
     needs: build-axolotl
@@ -68,6 +77,11 @@ jobs:
             pytorch: 2.0.1
             axolotl_extras:
             is_latest: true
+          - cuda: 118
+            cuda_version: 11.8.0
+            python_version: "3.10"
+            pytorch: 2.1.0
+            axolotl_extras:
     runs-on: [self-hosted, gpu, docker]
     steps:
       - name: Checkout

.github/workflows/tests.yml

@@ -4,7 +4,13 @@ on:
   push:
     branches:
       - "main"
+    paths:
+      - '**.py'
+      - 'requirements.txt'
   pull_request:
+      paths:
+       - '**.py'
+       - 'requirements.txt'
   workflow_dispatch:
 
 jobs:
@@ -40,7 +46,7 @@ jobs:
 
       - name: Install dependencies
         run: |
-          pip3 install -e .
+          pip3 install -U -e .
           pip3 install -r requirements-tests.txt
 
       - name: Run tests
@@ -65,8 +71,8 @@ jobs:
 
       - name: Install dependencies
         run: |
-          pip3 install -e .
-          pip3 install flash-attn
+          pip3 uninstall -y transformers accelerate
+          pip3 install -U -e .[flash-attn]
           pip3 install -r requirements-tests.txt
 
       - name: Run e2e tests

.gitignore

@@ -161,3 +161,7 @@ cython_debug/
 #  and can be added to the global gitignore or merged into this file.  For a more nuclear
 #  option (not recommended) you can uncomment the following to ignore the entire idea folder.
 .idea/
+
+# WandB
+# wandb creates a folder to store logs for training runs
+wandb

.isort.cfg

@@ -1,2 +1,3 @@
 [settings]
 profile=black
+known_third_party=wandb

README.md

@@ -23,9 +23,10 @@ Features:
 - [Supported Features](#axolotl-supports)
 - [Quickstart](#quickstart-)
 - [Installation](#installation)
-  - [Docker Installation](#environment)
-  - [Conda/Pip venv Installation](#condapip-venv)
-  - [LambdaLabs Installation](#lambdalabs)
+  - [Docker](#docker)
+  - [Conda/Pip venv](#condapip-venv)
+  - [LambdaLabs](#lambdalabs)
+  - [Windows](#windows)
 - [Dataset](#dataset)
   - [How to Add Custom Prompts](#how-to-add-custom-prompts)
   - [How to Use Custom Pretokenized Dataset](#how-to-use-your-custom-pretokenized-dataset)
@@ -49,7 +50,7 @@ Features:
       <b>Axolotl provides a unified repository for fine-tuning <br />a variety of AI models with ease</b>
     </p>
     <p>
-      Go ahead and axolotl questions!!
+      Go ahead and Axolotl questions!!
     </p>
     <img src="https://github.com/OpenAccess-AI-Collective/axolotl/actions/workflows/pre-commit.yml/badge.svg?branch=main" alt="pre-commit">
     <img alt="PyTest Status" src="https://github.com/OpenAccess-AI-Collective/axolotl/actions/workflows/tests.yml/badge.svg?branch=main">
@@ -73,6 +74,7 @@ Features:
 | gpt-j    | ✅         | ✅    | ✅     | ❌             | ❌                 | ❓          | ❓            |
 | XGen     | ✅         | ❓    | ✅     | ❓             | ❓                 | ❓          | ✅            |
 | phi      | ✅         | ✅    | ✅     | ❓             | ❓                 | ❓          | ❓            |
+| RWKV     | ✅         | ❓    | ❓     | ❓             | ❓                 | ❓          | ❓            |
 
 
 ## Quickstart ⚡
@@ -86,7 +88,7 @@ git clone https://github.com/OpenAccess-AI-Collective/axolotl
 cd axolotl
 
 pip3 install packaging
-pip3 install -e .[flash-attn]
+pip3 install -e '.[flash-attn,deepspeed]'
 pip3 install -U git+https://github.com/huggingface/peft.git
 
 # finetune lora
@@ -95,13 +97,17 @@ accelerate launch -m axolotl.cli.train examples/openllama-3b/lora.yml
 # inference
 accelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \
     --lora_model_dir="./lora-out"
+
+# gradio
+accelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \
+    --lora_model_dir="./lora-out" --gradio
 ```
 
 ## Installation
 
 ### Environment
 
-- Docker
+#### Docker
   ```bash
   docker run --gpus '"all"' --rm -it winglian/axolotl:main-py3.10-cu118-2.0.1
   ```
@@ -113,18 +119,42 @@ accelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \
   docker compose up -d
   ```
 
-- Conda/Pip venv
+  <details>
+
+  <summary>Docker advanced</summary>
+
+  A more powerful Docker command to run would be this:
+
+  ```bash
+  docker run --gpus '"all"' --rm -it --name axolotl --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 --mount type=volume,src=axolotl,target=/workspace/axolotl -v ${HOME}/.cache/huggingface:/root/.cache/huggingface winglian/axolotl:main-py3.10-cu118-2.0.1
+  ```
+
+  It additionally:
+  * Prevents memory issues when running e.g. deepspeed (e.g. you could hit SIGBUS/signal 7 error) through `--ipc` and `--ulimit` args.
+  * Persists the downloaded HF data (models etc.) and your modifications to axolotl code through `--mount`/`-v` args.
+  * The `--name` argument simply makes it easier to refer to the container in vscode (`Dev Containers: Attach to Running Container...`) or in your terminal.
+
+  [More information on nvidia website](https://docs.nvidia.com/deeplearning/frameworks/user-guide/index.html#setincshmem)
+
+  </details>
+
+#### Conda/Pip venv
   1. Install python >=**3.9**
 
   2. Install pytorch stable https://pytorch.org/get-started/locally/
 
-  3. Install axolotl along with python dependencies
+  3. Install Axolotl along with python dependencies
         ```bash
         pip3 install packaging
-        pip3 install -e .[flash-attn]
+        pip3 install -e '.[flash-attn,deepspeed]'
         ```
+  4. (Optional) Login to Huggingface to use gated models/datasets.
+        ```bash
+        huggingface-cli login
+        ```
+        Get the token at huggingface.co/settings/tokens
 
-- LambdaLabs
+#### LambdaLabs
   <details>
 
   <summary>Click to Expand</summary>
@@ -157,7 +187,7 @@ accelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \
   cd axolotl
 
   pip3 install packaging
-  pip3 install -e .[flash-attn]
+  pip3 install -e '.[flash-attn,deepspeed]'
   pip3 install protobuf==3.20.3
   pip3 install -U --ignore-installed requests Pillow psutil scipy
   ```
@@ -168,7 +198,8 @@ accelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \
   ```
   </details>
 
-- Windows: Please use WSL or Docker!
+#### Windows
+Please use WSL or Docker!
 
 ### Dataset
 
@@ -179,7 +210,7 @@ Have dataset(s) in one of the following format (JSONL recommended):
   ```json
   {"instruction": "...", "input": "...", "output": "..."}
   ```
-- `sharegpt:chat`: conversations where `from` is `human`/`gpt`
+- `sharegpt`: conversations where `from` is `human`/`gpt`
   ```json
   {"conversations": [{"from": "...", "value": "..."}]}
   ```
@@ -244,6 +275,10 @@ Have dataset(s) in one of the following format (JSONL recommended):
   ```json
   {"article": "...", "question": "...", "answer": "..."}
   ```
+- `context_qa.load_v2`: in context question answering (alternate)
+  ```json
+  {"context": "...", "question": "...", "answer": "..."}
+  ```
 - `context_qa.load_404`: in context question answering from an article, with default response for no answer from context
   ```json
   {"article": "...", "unanswerable_question": "..."}
@@ -268,11 +303,11 @@ Have dataset(s) in one of the following format (JSONL recommended):
   ```json
   {"prompt": "...", "generation": "..."}
   ```
-- `sharegpt_simple.load_role`: conversations where `role` is used instead of `from`
+- `sharegpt.load_role`: conversations where `role` is used instead of `from`
   ```json
   {"conversations": [{"role": "...", "value": "..."}]}
   ```
-- `sharegpt_simple.load_guanaco`: conversations where `from` is `prompter`/`assistant` instead of default sharegpt
+- `sharegpt.load_guanaco`: conversations where `from` is `prompter`/`assistant` instead of default sharegpt
   ```json
   {"conversations": [{"from": "...", "value": "..."}]}
   ```
@@ -285,29 +320,28 @@ Have dataset(s) in one of the following format (JSONL recommended):
 
 #### How to add custom prompts
 
-Using yaml. Example:
+For a dataset that is preprocessed for instruction purposes:
+
+```json
+{"instruction": "...", "output": "..."}
+```
+
+You can use this example in your YAML config:
+
 ```yaml
 datasets:
   - path: repo
     type:
       system_prompt: ""
-      no_input_format: |-
-        User: {instruction}<|end_of_turn|>
-        Assistant:
-      format: |-
-        User: {instruction}
-        {input}<|end_of_turn|>
-        Assistant:
+      field_system: system
+      format: "[INST] {instruction} [/INST]"
+      no_input_format: "[INST] {instruction} [/INST]"
 ```
 
-Using file:
-1. Add your method to a file in [prompt_strategies](src/axolotl/prompt_strategies). Please see other files as example.
-2. Use your custom file name as the dataset type `<prompt_strategies_file>.load_<load_fn>`.
-
 #### How to use your custom pretokenized dataset
 
 - Do not pass a `type:`
-- Dataset must contain `input_ids`, `attention_mask`, `labels` in columns
+- Columns in Dataset must be exactly `input_ids`, `attention_mask`, `labels`
 
 
 ### Config
@@ -345,11 +379,24 @@ See [examples](examples) for quick start. It is recommended to duplicate and mod
         - typescript
       type: ... # unimplemented custom format
 
+  # fastchat conversation
+  # See 'conversation' options: https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py
+  datasets:
+    - path: ...
+      type: sharegpt
+      conversation: chatml
+
   # local
   datasets:
     - path: data.jsonl # or json
       ds_type: json # see other options below
       type: alpaca
+
+  # dataset with splits, but no train split
+  dataset:
+    - path: knowrohit07/know_sql
+      type: context_qa.load_v2
+      train_on_split: validation
   ```
 
 - loading
@@ -377,18 +424,18 @@ See [examples](examples) for quick start. It is recommended to duplicate and mod
 
 <details>
 
-<summary>All yaml options</summary>
+<summary>All yaml options (click me)</summary>
 
 ```yaml
-# this is the huggingface model that contains *.pt, *.safetensors, or *.bin files
-# this can also be a relative path to a model on disk
+# This is the huggingface model that contains *.pt, *.safetensors, or *.bin files
+# This can also be a relative path to a model on disk
 base_model: ./llama-7b-hf
-# you can specify an ignore pattern if the model repo contains more than 1 model type (*.pt, etc)
+# You can specify an ignore pattern if the model repo contains more than 1 model type (*.pt, etc)
 base_model_ignore_patterns:
-# if the base_model repo on hf hub doesn't include configuration .json files,
-# you can set that here, or leave this empty to default to base_model
+# If the base_model repo on hf hub doesn't include configuration .json files,
+# You can set that here, or leave this empty to default to base_model
 base_model_config: ./llama-7b-hf
-# you can specify to choose a specific model revision from huggingface hub
+# You can specify to choose a specific model revision from huggingface hub
 model_revision:
 # Optional tokenizer configuration override in case you want to use a different tokenizer
 # than the one defined in the base model
@@ -403,18 +450,24 @@ trust_remote_code:
 tokenizer_use_fast:
 # Whether to use the legacy tokenizer setting, defaults to True
 tokenizer_legacy:
-# resize the model embeddings when new tokens are added to multiples of 32
-# this is reported to improve training speed on some models
+# Resize the model embeddings when new tokens are added to multiples of 32
+# This is reported to improve training speed on some models
 resize_token_embeddings_to_32x:
 
-# whether you are training a 4-bit GPTQ quantized model
+# Used to identify which the model is based on
+is_falcon_derived_model:
+is_llama_derived_model:
+# Please note that if you set this to true, `padding_side` will be set to "left" by default
+is_mistral_derived_model:
+
+# Whether you are training a 4-bit GPTQ quantized model
 gptq: true
 gptq_groupsize: 128 # group size
 gptq_model_v1: false # v1 or v2
 
-# this will attempt to quantize the model down to 8 bits and use adam 8 bit optimizer
+# This will attempt to quantize the model down to 8 bits and use adam 8 bit optimizer
 load_in_8bit: true
-# use bitsandbytes 4 bit
+# Use bitsandbytes 4 bit
 load_in_4bit:
 
 # Use CUDA bf16
@@ -428,9 +481,9 @@ tf32: true # require >=ampere
 bfloat16: true # require >=ampere
 float16: true
 
-# a list of one or more datasets to finetune the model with
+# A list of one or more datasets to finetune the model with
 datasets:
-  # hf dataset repo | "json" for local dataset, make sure to fill data_files
+  # HuggingFace dataset repo | "json" for local dataset, make sure to fill data_files
   - path: vicgalle/alpaca-gpt4
   # The type of prompt to use for training. [alpaca, sharegpt, gpteacher, oasst, reflection]
     type: alpaca # format | format:<prompt_style> (chat/instruct) | <prompt_strategies>.load_<load_fn>
@@ -439,17 +492,21 @@ datasets:
     shards: # Optional[int] number of shards to split data into
     name: # Optional[str] name of dataset configuration to load
 
-  # custom user prompt
+    # Optional[str] fastchat conversation type, only used with type: sharegpt
+    conversation:  # Options (see Conversation 'name'): https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py
+
+  # Custom user prompt
   - path: repo
     type:
-      # the below are defaults. only set what's needed.
+      # The below are defaults. only set what's needed.
       system_prompt: ""
+      system_format: "{system}"
       field_system: system
       field_instruction: instruction
-      field_output: input
+      field_input: input
+      field_output: output
 
-      # customizable to be single line or multi-line
-      system_format: "{system}"
+      # Customizable to be single line or multi-line
       # 'format' can include {input}
       format: |-
         User: {instruction} {input}
@@ -457,21 +514,24 @@ datasets:
       # 'no_input_format' cannot include {input}
       no_input_format: "{instruction} "
 
-      # for completions datsets, uses the provided field if not `text`
+      # For `completion` datsets only, uses the provided field instead of `text` column
       field:
 
-# axolotl attempts to save the dataset as an arrow after packing the data together so
+# Axolotl attempts to save the dataset as an arrow after packing the data together so
 # subsequent training attempts load faster, relative path
 dataset_prepared_path: data/last_run_prepared
-# push prepared dataset to hub
+# Push prepared dataset to hub
 push_dataset_to_hub: # repo path
+# The maximum number of processes to use while preprocessing your input dataset. This defaults to `os.cpu_count()`
+# if not set.
+dataset_processes: # defaults to os.cpu_count() if not set
 # push checkpoints to hub
 hub_model_id: # repo path to push finetuned model
 # how to push checkpoints to hub
 # https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy
 hub_strategy:
-# whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets
-# required to be true when used in combination with `push_dataset_to_hub`
+# Whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets
+# Required to be true when used in combination with `push_dataset_to_hub`
 hf_use_auth_token: # boolean
 # How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc. 0 for no eval.
 val_set_size: 0.04
@@ -480,28 +540,34 @@ dataset_shard_num:
 # Index of shard to use for whole dataset
 dataset_shard_idx:
 
-# the maximum length of an input to train with, this should typically be less than 2048
+# The maximum length of an input to train with, this should typically be less than 2048
 # as most models have a token/context limit of 2048
 sequence_len: 2048
-# pad inputs so each step uses constant sized buffers
-# this will reduce memory fragmentation and may prevent OOMs, by re-using memory more efficiently
+# Pad inputs so each step uses constant sized buffers
+# This will reduce memory fragmentation and may prevent OOMs, by re-using memory more efficiently
 pad_to_sequence_len:
-# max sequence length to concatenate training samples together up to
-# inspired by StackLLaMA. see https://huggingface.co/blog/stackllama#supervised-fine-tuning
+# Max sequence length to concatenate training samples together up to
+# Inspired by StackLLaMA. see https://huggingface.co/blog/stackllama#supervised-fine-tuning
 # FutureWarning: This will soon be DEPRECATED
 max_packed_sequence_len: 1024
-# use efficient multi-packing with block diagonal attention and per sequence position_ids. Recommend set to 'true'
+# Use efficient multi-packing with block diagonal attention and per sequence position_ids. Recommend set to 'true'
 sample_packing:
-# you can set these packing optimizations AFTER starting a training at least once.
+# Set to 'false' if getting errors during eval with sample_packing on.
+eval_sample_packing:
+# You can set these packing optimizations AFTER starting a training at least once.
 # The trainer will provide recommended values for these values.
 sample_packing_eff_est:
 total_num_tokens:
 
-# if you want to use 'lora' or 'qlora' or leave blank to train all parameters in original model
+# If you want to use 'lora' or 'qlora' or leave blank to train all parameters in original model
 adapter: lora
-# if you already have a lora model trained that you want to load, put that here
-# lora hyperparameters
+# If you already have a lora model trained that you want to load, put that here.
+# This means after training, if you want to test the model, you should set this to the value of `lora_out_dir`.
 lora_model_dir:
+
+# LoRA hyperparameters
+# For more details about the following options, see:
+# https://www.anyscale.com/blog/fine-tuning-llms-lora-or-full-parameter-an-in-depth-analysis-with-llama-2
 lora_r: 8
 lora_alpha: 16
 lora_dropout: 0.05
@@ -513,81 +579,96 @@ lora_target_modules:
 #  - gate_proj
 #  - down_proj
 #  - up_proj
-lora_target_linear: # if true, will target all linear layers
+lora_target_linear: # If true, will target all linear layers
+
+# If you added new tokens to the tokenizer, you may need to save some LoRA modules because they need to know the new tokens.
+# For LLaMA and Mistral, you need to save `embed_tokens` and `lm_head`. It may vary for other models.
+# `embed_tokens` converts tokens to embeddings, and `lm_head` converts embeddings to token probabilities.
+# https://github.com/huggingface/peft/issues/334#issuecomment-1561727994
 lora_modules_to_save:
 #  - embed_tokens
 #  - lm_head
+
+# Once you complete training, the model will be saved to the following directory.
+# If you merge the adapter to the base model, a subdirectory `merged` will be created under this directory.
+# Make sure `lora_model_dir` points to this directory if you want to use the trained model.
 lora_out_dir:
 lora_fan_in_fan_out: false
 
 # ReLoRA configuration
-# must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed
-relora_steps: # number of steps per ReLoRA restart
-relora_warmup_steps: # number of per-restart warmup steps
-relora_cpu_offload: # true to perform lora weight merges on cpu during restarts, for modest gpu memory savings
+# Must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed
+relora_steps: # Number of steps per ReLoRA restart
+relora_warmup_steps: # Number of per-restart warmup steps
+relora_cpu_offload: # True to perform lora weight merges on cpu during restarts, for modest gpu memory savings
 
 # wandb configuration if you're using it
 wandb_mode: # "offline" to save run metadata locally and not sync to the server, "disabled" to turn off wandb
-wandb_project: # your wandb project name
-wandb_entity: # a wandb Team name if using a Team
+wandb_project: # Your wandb project name
+wandb_entity: # A wandb Team name if using a Team
 wandb_watch:
-wandb_run_id: # set the name of your wandb run
+wandb_run_id: # Set the name of your wandb run
 wandb_log_model: # "checkpoint" to log model to wandb Artifacts every `save_steps` or "end" to log only at the end of training
 
-# where to save the finished model to
+# Where to save the full-finetuned model to
 output_dir: ./completed-model
 
-# whether to use torch.compile and which backend to use
+# Whether to use torch.compile and which backend to use
 torch_compile:  # bool
 torch_compile_backend:  # Optional[str]
 
-# training hyperparameters
+# Training hyperparameters
+
+# If greater than 1, backpropagation will be skipped and the gradients will be accumulated for the given number of steps.
 gradient_accumulation_steps: 1
+# The number of samples to include in each batch. This is the number of samples sent to each GPU.
 micro_batch_size: 2
-eval_batch_size: 2
-num_epochs: 3
+eval_batch_size:
+num_epochs: 4
 warmup_steps: 100
 learning_rate: 0.00003
 lr_quadratic_warmup:
 logging_steps:
-save_strategy: # set to `no` to skip checkpoint saves
-save_steps: # leave empty to save at each epoch
-eval_steps: # leave empty to eval at each epoch
-save_total_limit: # checkpoints saved at a time
+save_strategy: # Set to `no` to skip checkpoint saves
+save_steps: # Leave empty to save at each epoch
+eval_steps: # Leave empty to eval at each epoch, integers for every N steps. decimal for fraction of total steps
+save_total_limit: # Checkpoints saved at a time
+# Maximum number of iterations to train for. It precedes num_epochs which means that
+# if both are set, num_epochs will not be guaranteed.
+# e.g., when 1 epoch is 1000 steps => `num_epochs: 2` and `max_steps: 100` will train for 100 steps
 max_steps:
 
-eval_table_size: # approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0
-eval_table_max_new_tokens: # total number of tokens generated for predictions sent to wandb. Default is 128
+eval_table_size: # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0
+eval_table_max_new_tokens: # Total number of tokens generated for predictions sent to wandb. Default is 128
 
-# save model as safetensors (require safetensors package)
+# Save model as safetensors (require safetensors package)
 save_safetensors:
 
-# whether to mask out or include the human's prompt from the training labels
+# Whether to mask out or include the human's prompt from the training labels
 train_on_inputs: false
-# group similarly sized data to minimize padding
-# may be slower to start, as it must download and sort the entire dataset
-# note that training loss may have an oscillating pattern with this enabled
+# Group similarly sized data to minimize padding.
+# May be slower to start, as it must download and sort the entire dataset.
+# Note that training loss may have an oscillating pattern with this enabled.
 group_by_length: false
 
 # Whether to use gradient checkpointing https://huggingface.co/docs/transformers/v4.18.0/en/performance#gradient-checkpointing
 gradient_checkpointing: false
 
-# stop training after this many evaluation losses have increased in a row
+# Stop training after this many evaluation losses have increased in a row
 # https://huggingface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppingCallback
 early_stopping_patience: 3
 
-# specify a scheduler and kwargs to use with the optimizer
+# Specify a scheduler and kwargs to use with the optimizer
 lr_scheduler: # 'one_cycle' | 'log_sweep' | empty for cosine
 lr_scheduler_kwargs:
 
-# for one_cycle optim
-lr_div_factor: # learning rate div factor
+# For one_cycle optim
+lr_div_factor: # Learning rate div factor
 
-# for log_sweep optim
+# For log_sweep optim
 log_sweep_min_lr:
 log_sweep_max_lr:
 
-# specify optimizer
+# Specify optimizer
 # Valid values are driven by the Transformers OptimizerNames class, see:
 # https://github.com/huggingface/transformers/blob/95b374952dc27d8511541d6f5a4e22c9ec11fb24/src/transformers/training_args.py#L134
 #
@@ -613,7 +694,7 @@ log_sweep_max_lr:
 # - paged_lion_32bit
 # - paged_lion_8bit
 optimizer:
-# specify weight decay
+# Specify weight decay
 weight_decay:
 # adamw hyperparams
 adam_beta1:
@@ -622,47 +703,58 @@ adam_epsilon:
 # Gradient clipping max norm
 max_grad_norm:
 
-# whether to bettertransformers
+# Augmentation techniques
+# NEFT https://arxiv.org/abs/2310.05914, set this to a number (paper default is 5) to add noise to embeddings
+# currently only supported on Llama and Mistral
+noisy_embedding_alpha:
+
+# Whether to bettertransformers
 flash_optimum:
-# whether to use xformers attention patch https://github.com/facebookresearch/xformers:
+# Whether to use xformers attention patch https://github.com/facebookresearch/xformers:
 xformers_attention:
-# whether to use flash attention patch https://github.com/Dao-AILab/flash-attention:
+# Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention:
 flash_attention:
-# whether to use scaled-dot-product attention
+flash_attn_cross_entropy:  # Whether to use flash-attention cross entropy implementation - advanced use only
+flash_attn_rms_norm:  # Whether to use flash-attention rms norm implementation - advanced use only
+flash_attn_fuse_qkv: # Whether to fuse QKV into a single operation
+flash_attn_fuse_mlp: # Whether to fuse part of the MLP into a single operation
+# Whether to use scaled-dot-product attention
 # https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html
 sdp_attention:
 # Landmark attention (only llama)
 landmark_attention:
 # xpos RoPE see https://github.com/kaiokendev/cutoff-len-is-context-len/blob/main/util/xpos_rope_llama_monkey_patch.py
-# llama only
+# LLaMA only
 xpos_rope:
 # RoPE Scaling https://github.com/huggingface/transformers/pull/24653
 rope_scaling:
   type: # linear | dynamic
   factor: # float
 
-# resume from a specific checkpoint dir
+# Resume from a specific checkpoint dir
 resume_from_checkpoint:
-# if resume_from_checkpoint isn't set and you simply want it to start where it left off
-# be careful with this being turned on between different models
+# If resume_from_checkpoint isn't set and you simply want it to start where it left off.
+# Be careful with this being turned on between different models.
 auto_resume_from_checkpoints: false
 
-# don't mess with this, it's here for accelerate and torchrun
+# Don't mess with this, it's here for accelerate and torchrun
 local_rank:
 
-# add or change special tokens
+# Add or change special tokens.
+# If you add tokens here, you don't need to add them to the `tokens` list.
 special_tokens:
   # bos_token: "<s>"
   # eos_token: "</s>"
   # unk_token: "<unk>"
-# add extra tokens
+
+# Add extra tokens.
 tokens:
 
 # FSDP
 fsdp:
 fsdp_config:
 
-# Deepspeed config path
+# Deepspeed config path. e.g., deepspeed/zero3.json
 deepspeed:
 
 # Advanced DDP Arguments
@@ -688,6 +780,66 @@ strict:
 
 </details>
 
+<details>
+<summary> Understanding of batch size and gradient accumulation steps </summary>
+<br/>
+Gradient accumulation means accumulating gradients over several mini-batches and updating the model weights afterward. When the samples in each batch are diverse, this technique doesn't significantly impact learning.
+
+This method allows for effective training with larger effective batch sizes without needing proportionally larger memory. Here's why:
+
+1. **Memory Consumption with Batch Size**: The primary reason increasing the batch size impacts memory is due to the storage requirements for intermediate activations. When you forward propagate a batch through a network, you have to store the activations at each layer for each sample in the batch, because these activations are used during backpropagation to compute gradients. Therefore, larger batches mean more activations, leading to greater GPU memory consumption.
+
+2. **Gradient Accumulation**: With gradient accumulation, you're effectively simulating a larger batch size by accumulating gradients over several smaller batches (or micro-batches). However, at any given time, you're only forward and backward propagating a micro-batch. This means you only store activations for the micro-batch, not the full accumulated batch. As a result, you can simulate the effect of a larger batch size without the memory cost of storing activations for a large batch.
+
+**Example 1:**
+Micro batch size: 3
+Gradient accumulation steps: 2
+Number of GPUs: 3
+Total batch size = 3 * 2 * 3 = 18
+
+```
+| GPU 1          | GPU 2          | GPU 3          |
+|----------------|----------------|----------------|
+| S1, S2, S3     | S4, S5, S6     | S7, S8, S9     |
+| e1, e2, e3     | e4, e5, e6     | e7, e8, e9     |
+|----------------|----------------|----------------|
+| → (accumulate) | → (accumulate) | → (accumulate) |
+|----------------|----------------|----------------|
+| S10, S11, S12  | S13, S14, S15  | S16, S17, S18  |
+| e10, e11, e12  | e13, e14, e15  | e16, e17, e18  |
+|----------------|----------------|----------------|
+| → (apply)      | → (apply)      | → (apply)      |
+
+Accumulated gradient for the weight w1 after the second iteration (considering all GPUs):
+Total gradient for w1 = e1 + e2 + e3 + e4 + e5 + e6 + e7 + e8 + e9 + e10 + e11 + e12 + e13 + e14 + e15 + e16 + e17 + e18
+
+Weight update for w1:
+w1_new = w1_old - learning rate x (Total gradient for w1 / 18)
+```
+
+**Example 2:**
+Micro batch size: 2
+Gradient accumulation steps: 1
+Number of GPUs: 3
+Total batch size = 2 * 1 * 3 = 6
+
+```
+| GPU 1     | GPU 2     | GPU 3     |
+|-----------|-----------|-----------|
+| S1, S2    | S3, S4    | S5, S6    |
+| e1, e2    | e3, e4    | e5, e6    |
+|-----------|-----------|-----------|
+| → (apply) | → (apply) | → (apply) |
+
+Accumulated gradient for the weight w1 (considering all GPUs):
+Total gradient for w1 = e1 + e2 + e3 + e4 + e5 + e6
+
+Weight update for w1:
+w1_new = w1_old - learning rate × (Total gradient for w1 / 6)
+```
+
+</details>
+
 ### Train
 
 Run
@@ -695,14 +847,41 @@ Run
 accelerate launch -m axolotl.cli.train your_config.yml
 ```
 
-#### Multi-GPU
+#### Preprocess dataset
+
+You can optionally pre-tokenize dataset with the following before finetuning.
+This is recommended for large datasets.
+
+- Set `push_dataset_to_hub: hf_user/repo` to push it to Huggingface.
+- Use `--debug` to see preprocessed examples.
 
-You can optionally pre-tokenize dataset with the following before finetuning:
 ```bash
-CUDA_VISIBLE_DEVICES="" accelerate launch -m axolotl.cli.train your_config.yml --prepare_ds_only
+python -m axolotl.cli.preprocess your_config.yml
 ```
 
-##### Config
+#### Multi-GPU
+
+Below are the options available in axolotl for training with multiple GPUs. Note that DeepSpeed
+is the recommended multi-GPU option currently because FSDP may experience
+[loss instability](https://github.com/huggingface/transformers/issues/26498).
+
+##### DeepSpeed
+
+Deepspeed is an optimization suite for multi-gpu systems allowing you to train much larger models than you
+might typically be able to fit into your GPU's VRAM. More information about the various optimization types
+for deepspeed is available at https://huggingface.co/docs/accelerate/main/en/usage_guides/deepspeed#what-is-integrated
+
+We provide several default deepspeed JSON configurations for ZeRO stage 1, 2, and 3.
+
+```yaml
+deepspeed: deepspeed/zero1.json
+```
+
+```shell
+accelerate launch -m axolotl.cli.train examples/llama-2/config.py --deepspeed deepspeed/zero1.json
+```
+
+##### FSDP
 
 - llama FSDP
 ```yaml
@@ -715,11 +894,6 @@ fsdp_config:
   fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
 ```
 
-- llama Deepspeed
-```yaml
-deepspeed: deepspeed/zero3.json
-```
-
 ##### Weights & Biases Logging
 
 - wandb options
@@ -749,6 +923,14 @@ Pass the appropriate flag to the train command:
   cat /tmp/prompt.txt | python -m axolotl.cli.inference examples/your_config.yml \
     --base_model="./completed-model" --prompter=None --load_in_8bit=True
   ```
+-- With gradio hosting
+  ```bash
+  python -m axolotl.cli.inference examples/your_config.yml --gradio
+  ```
+
+Please use `--sample_packing False` if you have it on and receive the error similar to below:
+
+> RuntimeError: stack expects each tensor to be equal size, but got [1, 32, 1, 128] at entry 0 and [1, 32, 8, 128] at entry 1
 
 ### Merge LORA to base
 
@@ -766,6 +948,8 @@ CUDA_VISIBLE_DEVICES="" python3 -m axolotl.cli.merge_lora ...
 
 ## Common Errors 🧰
 
+See also the [FAQ's](./docs/faq.md).
+
 > If you encounter a 'Cuda out of memory' error, it means your GPU ran out of memory during the training process. Here's how to resolve it:
 
 Please reduce any below

deepspeed/zero1.json

@@ -1,39 +1,41 @@
 {
-    "zero_optimization": {
-      "stage": 1,
-      "overlap_comm": true
-    },
-    "bf16": {
-      "enabled": "auto"
-    },
-    "fp16": {
-      "enabled": "auto",
-      "auto_cast": false,
-      "loss_scale": 0,
-      "initial_scale_power": 32,
-      "loss_scale_window": 1000,
-      "hysteresis": 2,
-      "min_loss_scale": 1
-    },
-    "optimizer": {
-      "type": "AdamW",
-      "params": {
-        "lr": "auto",
-        "betas": "auto",
-        "eps": "auto",
-        "weight_decay": "auto"
-      }
-    },
-    "scheduler": {
-      "type": "WarmupDecayLR",
-      "params": {
-        "warmup_min_lr": "auto",
-        "warmup_max_lr": "auto",
-        "warmup_num_steps": "auto",
-        "total_num_steps": "auto"
-      }
-    },
-    "train_batch_size": "auto",
-    "train_micro_batch_size_per_gpu": "auto",
-    "wall_clock_breakdown": false
+  "zero_optimization": {
+    "stage": 1,
+    "overlap_comm": true
+  },
+  "bf16": {
+    "enabled": "auto"
+  },
+  "fp16": {
+    "enabled": "auto",
+    "auto_cast": false,
+    "loss_scale": 0,
+    "initial_scale_power": 32,
+    "loss_scale_window": 1000,
+    "hysteresis": 2,
+    "min_loss_scale": 1
+  },
+  "optimizer": {
+    "type": "AdamW",
+    "params": {
+      "lr": "auto",
+      "betas": "auto",
+      "eps": "auto",
+      "weight_decay": "auto"
+    }
+  },
+  "scheduler": {
+    "type": "WarmupDecayLR",
+    "params": {
+      "warmup_min_lr": "auto",
+      "warmup_max_lr": "auto",
+      "warmup_num_steps": "auto",
+      "warmup_type": "linear",
+      "total_num_steps": "auto"
+    }
+  },
+  "gradient_accumulation_steps": "auto",
+  "train_batch_size": "auto",
+  "train_micro_batch_size_per_gpu": "auto",
+  "wall_clock_breakdown": false
 }

deepspeed/zero2.json

@@ -1,43 +1,45 @@
 {
-    "zero_optimization": {
-      "stage": 2,
-      "offload_optimizer": {
-        "device": "cpu"
-      },
-      "contiguous_gradients": true,
-      "overlap_comm": true
+  "zero_optimization": {
+    "stage": 2,
+    "offload_optimizer": {
+      "device": "cpu"
     },
-    "bf16": {
-      "enabled": "auto"
-    },
-    "fp16": {
-      "enabled": "auto",
-      "auto_cast": false,
-      "loss_scale": 0,
-      "initial_scale_power": 32,
-      "loss_scale_window": 1000,
-      "hysteresis": 2,
-      "min_loss_scale": 1
-    },
-    "optimizer": {
-      "type": "AdamW",
-      "params": {
-        "lr": "auto",
-        "betas": "auto",
-        "eps": "auto",
-        "weight_decay": "auto"
-      }
-    },
-    "scheduler": {
-      "type": "WarmupDecayLR",
-      "params": {
-        "warmup_min_lr": "auto",
-        "warmup_max_lr": "auto",
-        "warmup_num_steps": "auto",
-        "total_num_steps": "auto"
-      }
-    },
-    "train_batch_size": "auto",
-    "train_micro_batch_size_per_gpu": "auto",
-    "wall_clock_breakdown": false
+    "contiguous_gradients": true,
+    "overlap_comm": true
+  },
+  "bf16": {
+    "enabled": "auto"
+  },
+  "fp16": {
+    "enabled": "auto",
+    "auto_cast": false,
+    "loss_scale": 0,
+    "initial_scale_power": 32,
+    "loss_scale_window": 1000,
+    "hysteresis": 2,
+    "min_loss_scale": 1
+  },
+  "optimizer": {
+    "type": "AdamW",
+    "params": {
+      "lr": "auto",
+      "betas": "auto",
+      "eps": "auto",
+      "weight_decay": "auto"
+    }
+  },
+  "scheduler": {
+    "type": "WarmupDecayLR",
+    "params": {
+      "warmup_min_lr": "auto",
+      "warmup_max_lr": "auto",
+      "warmup_num_steps": "auto",
+      "warmup_type": "linear",
+      "total_num_steps": "auto"
+    }
+  },
+  "gradient_accumulation_steps": "auto",
+  "train_batch_size": "auto",
+  "train_micro_batch_size_per_gpu": "auto",
+  "wall_clock_breakdown": false
 }

deepspeed/zero3.json

@@ -1,14 +1,6 @@
 {
   "zero_optimization": {
     "stage": 3,
-    "offload_optimizer": {
-      "device": "cpu",
-      "pin_memory": true
-    },
-    "offload_param": {
-      "device": "cpu",
-      "pin_memory": true
-    },
     "overlap_comm": true,
     "contiguous_gradients": true,
     "sub_group_size": 0,
@@ -41,13 +33,16 @@
     }
   },
   "scheduler": {
-    "type": "WarmupLR",
+    "type": "WarmupDecayLR",
     "params": {
       "warmup_min_lr": "auto",
       "warmup_max_lr": "auto",
-      "warmup_num_steps": "auto"
+      "warmup_num_steps": "auto",
+      "warmup_type": "linear",
+      "total_num_steps": "auto"
     }
   },
+  "gradient_accumulation_steps": "auto",
   "train_batch_size": "auto",
   "train_micro_batch_size_per_gpu": "auto",
   "wall_clock_breakdown": false

docker/Dockerfile

@@ -5,6 +5,9 @@ ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
 ARG AXOLOTL_EXTRAS=""
 ARG CUDA="118"
 ENV BNB_CUDA_VERSION=$CUDA
+ARG PYTORCH_VERSION="2.0.1"
+
+ENV PYTORCH_VERSION=$PYTORCH_VERSION
 
 RUN apt-get update && \
     apt-get install -y vim curl
@@ -12,17 +15,19 @@ RUN apt-get update && \
 WORKDIR /workspace
 
 RUN git clone --depth=1 https://github.com/OpenAccess-AI-Collective/axolotl.git
+
+WORKDIR /workspace/axolotl
+
 # If AXOLOTL_EXTRAS is set, append it in brackets
-RUN cd axolotl && \
-    if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
-        pip install -e .[flash-attn,$AXOLOTL_EXTRAS]; \
+RUN sed -i "s/torch==.*/torch==$PYTORCH_VERSION/" requirements.txt
+RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
+        pip install -e .[deepspeed,flash-attn,$AXOLOTL_EXTRAS]; \
     else \
-        pip install -e .[flash-attn]; \
+        pip install -e .[deepspeed,flash-attn]; \
     fi
 
 # fix so that git fetch/pull from remote works
-RUN cd axolotl && \
-    git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
+RUN git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
     git config --get remote.origin.fetch
 
 # helper for huggingface-login cli

docker/Dockerfile-base

@@ -10,70 +10,28 @@ ENV PATH="/root/miniconda3/bin:${PATH}"
 ARG PYTHON_VERSION="3.9"
 ARG PYTORCH_VERSION="2.0.1"
 ARG CUDA="118"
+ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
 
 ENV PYTHON_VERSION=$PYTHON_VERSION
+ENV TORCH_CUDA_ARCH_LIST=$TORCH_CUDA_ARCH_LIST
 
-RUN apt-get update
-RUN apt-get install -y wget git build-essential ninja-build git-lfs libaio-dev && rm -rf /var/lib/apt/lists/*
-
-RUN wget \
+RUN apt-get update \
+    && apt-get install -y wget git build-essential ninja-build git-lfs libaio-dev && rm -rf /var/lib/apt/lists/* \
+    && wget \
     https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh \
     && mkdir /root/.conda \
     && bash Miniconda3-latest-Linux-x86_64.sh -b \
-    && rm -f Miniconda3-latest-Linux-x86_64.sh
-
-RUN conda create -n "py${PYTHON_VERSION}" python="${PYTHON_VERSION}"
+    && rm -f Miniconda3-latest-Linux-x86_64.sh \
+    && conda create -n "py${PYTHON_VERSION}" python="${PYTHON_VERSION}"
 
 ENV PATH="/root/miniconda3/envs/py${PYTHON_VERSION}/bin:${PATH}"
 
 WORKDIR /workspace
 
 RUN python3 -m pip install --upgrade pip && pip3 install packaging && \
-    python3 -m pip install --no-cache-dir -U torch==${PYTORCH_VERSION}+cu${CUDA} --extra-index-url https://download.pytorch.org/whl/cu$CUDA
+    python3 -m pip install --no-cache-dir -U torch==${PYTORCH_VERSION}+cu${CUDA} deepspeed-kernels --extra-index-url https://download.pytorch.org/whl/cu$CUDA
 
-FROM base-builder AS deepspeed-builder
-
-ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
-
-WORKDIR /workspace
-
-RUN git clone https://github.com/microsoft/DeepSpeed.git && \
-    cd DeepSpeed && \
-    MAX_CONCURRENCY=8 DS_BUILD_SPARSE_ATTN=0 DS_BUILD_OPS=1 DS_BUILD_EVOFORMER_ATTN=0 python3 setup.py bdist_wheel
-
-FROM base-builder AS bnb-builder
-
-WORKDIR /workspace
-ARG CUDA="118"
-ENV CUDA=$CUDA
-
-RUN git clone https://github.com/TimDettmers/bitsandbytes.git && \
-    cd bitsandbytes && \
-    CUDA_VERSION=$CUDA make cuda11x && \
-    python setup.py bdist_wheel
-
-FROM base-builder
-
-ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
-ENV TORCH_CUDA_ARCH_LIST=$TORCH_CUDA_ARCH_LIST
-
-# recompile apex
-RUN python3 -m pip uninstall -y apex
-RUN git clone https://github.com/NVIDIA/apex
-#  `MAX_JOBS=1` disables parallel building to avoid cpu memory OOM when building image on GitHub Action (standard) runners
-RUN cd apex && MAX_JOBS=1 python3 -m pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" ./
-
-RUN mkdir -p /workspace/builds
-COPY --from=bnb-builder /workspace/bitsandbytes /workspace/builds/bitsandbytes
-
-RUN mkdir -p /workspace/wheels/bitsandbytes
-COPY --from=deepspeed-builder /workspace/DeepSpeed/dist/deepspeed-*.whl wheels
-COPY --from=bnb-builder /workspace/bitsandbytes/dist/bitsandbytes-*.whl wheels
-COPY --from=bnb-builder /workspace/bitsandbytes/bitsandbytes/libbitsandbytes*.so wheels/bitsandbytes
-
-RUN pip3 install wheels/deepspeed-*.whl
-RUN cd /workspace/builds/bitsandbytes && python3 setup.py install
-RUN git lfs install --skip-repo
-RUN pip3 install awscli && \
+RUN git lfs install --skip-repo && \
+    pip3 install awscli && \
     # The base image ships with `pydantic==1.8.2` which is not working
     pip3 install -U --no-cache-dir pydantic==1.10.10

docs/faq.md

@@ -0,0 +1,18 @@
+# Axolotl FAQ's
+
+
+> The trainer stopped and hasn't progressed in several minutes.
+
+Usually an issue with the GPU's communicating with each other. See the [NCCL doc](../docs/nccl.md)
+
+> Exitcode -9
+
+This usually happens when you run out of system RAM.
+
+> Exitcode -7 while using deepspeed
+
+Try upgrading deepspeed w: `pip install -U deepspeed`
+
+> AttributeError: 'DummyOptim' object has no attribute 'step'
+
+You may be using deepspeed with single gpu. Please don't set `deepspeed:` in yaml or cli.

docs/multi-node.md

@@ -0,0 +1,45 @@
+# Multi Node
+
+You will need to create a configuration for accelerate, either by using `accelerate config` and follow the instructions or you can use one of the preset below:
+
+~/.cache/huggingface/accelerate/default_config.yaml
+```yaml
+compute_environment: LOCAL_MACHINE
+debug: false
+distributed_type: FSDP
+downcast_bf16: 'no'
+machine_rank: 0 # Set to 0 for the main machine, increment by one for other machines
+main_process_ip: 10.0.0.4 # Set to main machine's IP
+main_process_port: 5000
+main_training_function: main
+mixed_precision: bf16
+num_machines: 2 # Change to the number of machines
+num_processes: 4 # That's the total number of GPUs, (for example: if you have 2 machines with 4 GPU, put 8)
+rdzv_backend: static
+same_network: true
+tpu_env: []
+tpu_use_cluster: false
+tpu_use_sudo: false
+use_cpu: false
+```
+
+Configure your model to use FSDP with for example:
+```yaml
+fsdp:
+  - full_shard
+  - auto_wrap
+fsdp_config:
+  fsdp_offload_params: true
+  fsdp_state_dict_type: FULL_STATE_DICT
+  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
+```
+
+## Machine configuration
+
+On each machine you need a copy of Axolotl, we suggest using the same commit to ensure compatibility.
+
+You will also need to have the same configuration file for your model on each machine.
+
+On the main machine only, make sure the port you set as `main_process_port` is open in TCP and reachable by other machines.
+
+All you have to do now is launch using accelerate as you would usually do on each machine and voila, the processes will start once you have launched accelerate on every machine.

docs/multipack.md

@@ -0,0 +1,51 @@
+# Multipack
+
+4k context, bsz =4,
+each character represents 256 tokens
+X represents a padding token
+
+```
+   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
+[[ A A A A A A A A A A A ]
+   B B B B B B ]
+   C C C C C C C ]
+   D D D D ]]
+
+[[ E E E E E E E E ]
+ [ F F F F ]
+ [ G G G ]
+ [ H H H H ]]
+
+[[ I I I ]
+ [ J J J ]
+ [ K K K K K]
+ [ L L L ]]
+```
+
+after padding to longest input in each step
+```
+   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
+[[ A A A A A A A A A A A ]
+   B B B B B B X X X X X X ]
+   C C C C C C C X X X X ]
+   D D D D X X X X X X X ]]
+
+[[ E E E E E E E E ]
+ [ F F F F X X X X ]
+ [ G G G X X X X X ]
+ [ H H H H X X X X ]]
+
+[[ I I I X X ]
+ [ J J J X X ]
+ [ K K K K K ]
+ [ L L L X X ]]
+```
+
+w packing ( note it's the same effective number of tokens per step, but a true bsz of 1)
+```
+   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
+[[ A A A A A A A A A A A B B B B B
+   B C C C C C C C D D D D E E E E
+   E E E E F F F F F G G G H H H H
+   I I I J J J J K K K K K L L L X ]]
+```

examples/cerebras/btlm-ft.yml

@@ -1,5 +1,4 @@
 base_model: cerebras/btlm-3b-8k-base
-base_model_config: cerebras/btlm-3b-8k-base
 model_type: AutoModelForCausalLM
 tokenizer_type: GPT2Tokenizer
 trust_remote_code: true
@@ -15,7 +14,7 @@ datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
 dataset_prepared_path: last_prepared_run
-val_set_size: 0.01
+val_set_size: 0.05
 
 adapter:
 lora_model_dir:

examples/cerebras/qlora.yml

@@ -1,5 +1,4 @@
 base_model: cerebras/Cerebras-GPT-1.3B
-base_model_config: cerebras/Cerebras-GPT-1.3B
 load_in_8bit: false
 load_in_4bit: true
 strict: false
@@ -7,8 +6,8 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 adapter: qlora
 lora_model_dir:
 sequence_len: 2048
@@ -50,7 +49,7 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-eval_steps: 20
+eval_steps: 0.05
 save_steps:
 debug:
 deepspeed:

examples/code-llama/13b/lora.yml

@@ -1,5 +1,4 @@
 base_model: codellama/CodeLlama-13b-hf
-base_model_config: codellama/CodeLlama-13b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
 is_llama_derived_model: true
@@ -11,8 +10,8 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 output_dir: ./lora-out
 
 sequence_len: 4096
@@ -35,7 +34,7 @@ wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -55,7 +54,7 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 20
+eval_steps: 0.05
 save_steps:
 debug:
 deepspeed:

examples/code-llama/13b/qlora.yml

@@ -1,5 +1,4 @@
 base_model: codellama/CodeLlama-13b-hf
-base_model_config: codellama/CodeLlama-13b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
 is_llama_derived_model: true
@@ -11,8 +10,8 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 output_dir: ./qlora-out
 
 adapter: qlora
@@ -37,7 +36,7 @@ wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 3
+num_epochs: 4
 optimizer: paged_adamw_32bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -57,7 +56,7 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 20
+eval_steps: 0.05
 save_steps:
 debug:
 deepspeed:

examples/code-llama/34b/lora.yml

@@ -1,5 +1,4 @@
 base_model: codellama/CodeLlama-34b-hf
-base_model_config: codellama/CodeLlama-34b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
 is_llama_derived_model: true
@@ -11,8 +10,8 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 output_dir: ./lora-out
 
 sequence_len: 4096
@@ -35,7 +34,7 @@ wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -55,7 +54,7 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 20
+eval_steps: 0.05
 save_steps:
 debug:
 deepspeed:

examples/code-llama/34b/qlora.yml

@@ -1,5 +1,4 @@
 base_model: codellama/CodeLlama-34b-hf
-base_model_config: codellama/CodeLlama-34b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
 is_llama_derived_model: true
@@ -11,8 +10,8 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 output_dir: ./qlora-out
 
 adapter: qlora
@@ -37,7 +36,7 @@ wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 3
+num_epochs: 4
 optimizer: paged_adamw_32bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -57,7 +56,7 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 20
+eval_steps: 0.05
 save_steps:
 debug:
 deepspeed:

examples/code-llama/7b/lora.yml

@@ -1,5 +1,4 @@
 base_model: codellama/CodeLlama-7b-hf
-base_model_config: codellama/CodeLlama-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
 is_llama_derived_model: true
@@ -11,8 +10,8 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 output_dir: ./lora-out
 
 sequence_len: 4096
@@ -35,7 +34,7 @@ wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -55,7 +54,7 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 20
+eval_steps: 0.05
 save_steps:
 debug:
 deepspeed:

examples/code-llama/7b/qlora.yml

@@ -1,5 +1,4 @@
 base_model: codellama/CodeLlama-7b-hf
-base_model_config: codellama/CodeLlama-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
 is_llama_derived_model: true
@@ -11,8 +10,8 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 output_dir: ./qlora-out
 
 adapter: qlora
@@ -37,7 +36,7 @@ wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 3
+num_epochs: 4
 optimizer: paged_adamw_32bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -57,7 +56,7 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 20
+eval_steps: 0.05
 save_steps:
 debug:
 deepspeed:

examples/falcon/config-7b-lora.yml

@@ -1,8 +1,8 @@
 base_model: tiiuae/falcon-7b
-base_model_config: tiiuae/falcon-7b
 trust_remote_code: true
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
+is_falcon_derived_model: true
 load_in_8bit: true
 load_in_4bit: false
 gptq: false
@@ -11,8 +11,8 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca:chat
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 adapter: lora
 lora_model_dir:
 sequence_len: 2048

examples/falcon/config-7b-qlora.yml

@@ -1,11 +1,11 @@
 # 1b: tiiuae/falcon-rw-1b
 # 40b: tiiuae/falcon-40b
 base_model: tiiuae/falcon-7b
-base_model_config: tiiuae/falcon-7b
 # required by falcon custom model code: https://huggingface.co/tiiuae/falcon-7b/tree/main
 trust_remote_code: true
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
+is_falcon_derived_model: true
 load_in_8bit: false
 # enable 4bit for QLoRA
 load_in_4bit: true
@@ -17,8 +17,8 @@ datasets:
     data_files:
       - Chain-of-Thought/formatted_cot_data/gsm8k_train.json
     type: "alpaca:chat"
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 # enable QLoRA
 adapter: qlora
 lora_model_dir:
@@ -53,7 +53,7 @@ output_dir: ./qlora-out
 # decrease if OOM, increase for max VRAM utilization
 micro_batch_size: 1
 gradient_accumulation_steps: 2
-num_epochs: 3
+num_epochs: 4
 # Optimizer for QLoRA
 optimizer: paged_adamw_32bit
 torchdistx_path:

examples/falcon/config-7b.yml

@@ -1,8 +1,8 @@
 base_model: tiiuae/falcon-7b
-base_model_config: tiiuae/falcon-7b
 trust_remote_code: true
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
+is_falcon_derived_model: true
 load_in_8bit: false
 load_in_4bit: false
 gptq: false
@@ -11,8 +11,8 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca:chat
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 adapter:
 lora_model_dir:
 sequence_len: 2048

examples/gptj/qlora.yml

@@ -1,5 +1,4 @@
 base_model: EleutherAI/gpt-j-6b
-base_model_config: EleutherAI/gpt-j-6b
 load_in_8bit: false
 load_in_4bit: true
 strict: false
@@ -7,8 +6,8 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 adapter: qlora
 lora_model_dir:
 sequence_len: 2048
@@ -47,7 +46,7 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-eval_steps: 20
+eval_steps: 0.05
 save_steps:
 debug:
 deepspeed:

examples/jeopardy-bot/config.yml

@@ -1,12 +1,11 @@
 base_model: huggyllama/llama-7b
-base_model_config: huggyllama/llama-7b
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 load_in_8bit: false
 datasets:
   - path: openaccess-ai-collective/jeopardy
     type: jeopardy
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter:
 lora_model_dir:
@@ -25,7 +24,7 @@ wandb_log_model:
 output_dir: ./jeopardy-bot-7b
 gradient_accumulation_steps: 1
 micro_batch_size: 1
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine

examples/llama-2/README.md

@@ -9,12 +9,16 @@ gradient_accumulation_steps: 2
 micro_batch_size: 1
 
 ```shell
-accelerate launch scripts/finetune.py examples/llama-2/qlora.yml
-
+accelerate launch -m axolotl.cli.train examples/llama-2/qlora.yml
 ```
 or
 
 ```shell
-accelerate launch scripts/finetune.py examples/llama-2/lora.yml
-
+accelerate launch -m axolotl.cli.train examples/llama-2/lora.yml
+```
+
+To launch a full finetuning with 16-bit precision:
+
+```shell
+accelerate launch -m axolotl.cli.train examples/llama-2/fft_optimized.yml
 ```

examples/llama-2/fft_optimized.yml

@@ -0,0 +1,72 @@
+base_model: NousResearch/Llama-2-7b-hf
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: false
+load_in_4bit: false
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.05
+output_dir: ./out
+
+sequence_len: 4096
+sample_packing: true
+pad_to_sequence_len: true
+
+adapter:
+lora_model_dir:
+lora_r:
+lora_alpha:
+lora_dropout:
+lora_target_linear:
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+
+gradient_accumulation_steps: 1
+micro_batch_size: 1
+num_epochs: 1
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+flash_attn_cross_entropy: false
+flash_attn_rms_norm: true
+flash_attn_fuse_qkv: false
+flash_attn_fuse_mlp: true
+
+warmup_steps: 100
+eval_steps: 0.05
+eval_table_size:
+save_steps:
+debug:
+deepspeed: #deepspeed/zero2.json # multi-gpu only
+weight_decay: 0.1
+fsdp:
+fsdp_config:
+special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"

examples/llama-2/gptq-lora.yml

@@ -1,5 +1,4 @@
 base_model: TheBloke/Llama-2-7B-GPTQ
-base_model_config: TheBloke/Llama-2-7B-GPTQ
 is_llama_derived_model: false
 gptq: true
 gptq_disable_exllama: true
@@ -15,8 +14,8 @@ hf_use_auth_token: true
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 adapter: lora
 lora_model_dir:
 sequence_len: 4096
@@ -38,7 +37,7 @@ wandb_log_model:
 output_dir: ./model-out
 gradient_accumulation_steps: 1
 micro_batch_size: 1
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_torch
 adam_beta2: 0.95
 adam_eps: 0.00001

examples/llama-2/lora.yml

@@ -1,5 +1,4 @@
-base_model: meta-llama/Llama-2-7b-hf
-base_model_config: meta-llama/Llama-2-7b-hf
+base_model: NousResearch/Llama-2-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 is_llama_derived_model: true
@@ -11,8 +10,8 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 output_dir: ./lora-out
 
 sequence_len: 4096
@@ -35,7 +34,7 @@ wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -55,8 +54,8 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 20
-eval_table_size: 5
+eval_steps: 0.05
+eval_table_size:
 eval_table_max_new_tokens: 128
 save_steps:
 debug:

examples/llama-2/qlora.yml

@@ -1,5 +1,4 @@
-base_model: meta-llama/Llama-2-7b-hf
-base_model_config: meta-llama/Llama-2-7b-hf
+base_model: NousResearch/Llama-2-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 is_llama_derived_model: true
@@ -11,8 +10,8 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 output_dir: ./qlora-out
 
 adapter: qlora
@@ -37,7 +36,7 @@ wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 3
+num_epochs: 4
 optimizer: paged_adamw_32bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -57,8 +56,8 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 20
-eval_table_size: 5
+eval_steps: 0.05
+eval_table_size:
 save_steps:
 debug:
 deepspeed:

examples/llama-2/relora.yml

@@ -1,5 +1,4 @@
-base_model: meta-llama/Llama-2-7b-hf
-base_model_config: meta-llama/Llama-2-7b-hf
+base_model: NousResearch/Llama-2-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 is_llama_derived_model: true
@@ -11,8 +10,8 @@ strict: false
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 output_dir: ./relora-out
 
 adapter: qlora
@@ -41,7 +40,7 @@ wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 4
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -61,7 +60,7 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 20
+eval_steps: 0.05
 save_steps: 50
 debug:
 deepspeed:

examples/llama-2/tiny-llama.yml

@@ -1,5 +1,4 @@
 base_model: PY007/TinyLlama-1.1B-step-50K-105b
-base_model_config: PY007/TinyLlama-1.1B-step-50K-105b
 
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
@@ -12,8 +11,8 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 output_dir: ./lora-out
 
 sequence_len: 4096
@@ -35,7 +34,7 @@ wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -55,8 +54,8 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 20
-eval_table_size: 5
+eval_steps: 0.05
+eval_table_size:
 save_steps:
 debug:
 deepspeed:

examples/mistral/README.md

@@ -0,0 +1,12 @@
+**Mistral 7B** is a language model with a total of 7.3 billion parameters, showcasing a notable performance across a variety of benchmarks.
+
+Fine Tune:
+```shell
+accelerate launch -m axolotl.cli.train examples/mistral/config.yml
+
+```
+
+If you run into CUDA OOM, use deepspeed with config zero2.json:
+```shell
+accelerate launch -m axolotl.cli.train examples/mistral/config.yml --deepspeed deepspeed/zero2.json
+```

examples/mistral/config.yml

@@ -0,0 +1,61 @@
+base_model: mistralai/Mistral-7B-v0.1
+model_type: MistralForCausalLM
+tokenizer_type: LlamaTokenizer
+is_mistral_derived_model: true
+
+load_in_8bit: false
+load_in_4bit: false
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./out
+
+sequence_len: 8192
+sample_packing: true
+pad_to_sequence_len: true
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 4
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.000005
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+eval_steps: 0.05
+eval_table_size:
+eval_table_max_new_tokens: 128
+save_steps:
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"

examples/mistral/qlora.yml

@@ -0,0 +1,78 @@
+base_model: mistralai/Mistral-7B-v0.1
+model_type: MistralForCausalLM
+tokenizer_type: LlamaTokenizer
+is_mistral_derived_model: true
+
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.05
+output_dir: ./qlora-out
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 8192
+sample_packing: true
+pad_to_sequence_len: true
+
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_fan_in_fan_out:
+lora_target_modules:
+  - gate_proj
+  - down_proj
+  - up_proj
+  - q_proj
+  - v_proj
+  - k_proj
+  - o_proj
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 1
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+eval_steps: 0.05
+eval_table_size:
+eval_table_max_new_tokens: 128
+save_steps:
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"

examples/mpt-7b/config.yml

@@ -1,12 +1,11 @@
 base_model: mosaicml/mpt-7b
-base_model_config: mosaicml/mpt-7b
 tokenizer_type: AutoTokenizer
 trust_remote_code: true  # required for mpt as their model class is not merged into transformers yet
 load_in_8bit: false
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter:
 lora_model_dir:
@@ -27,7 +26,7 @@ wandb_log_model:
 output_dir: ./mpt-alpaca-7b
 gradient_accumulation_steps: 1
 micro_batch_size: 1
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine

examples/openllama-3b/config.yml

@@ -1,5 +1,4 @@
 base_model: openlm-research/open_llama_3b_v2
-base_model_config: openlm-research/open_llama_3b_v2
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 load_in_8bit: false
@@ -9,7 +8,7 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter:
 lora_model_dir:

examples/openllama-3b/lora.yml

@@ -1,5 +1,4 @@
 base_model: openlm-research/open_llama_3b_v2
-base_model_config: openlm-research/open_llama_3b_v2
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 load_in_8bit: true
@@ -9,7 +8,7 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter: lora
 lora_model_dir:

examples/openllama-3b/qlora.yml

@@ -1,5 +1,4 @@
 base_model: openlm-research/open_llama_3b_v2
-base_model_config: openlm-research/open_llama_3b_v2
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 load_in_8bit: false
@@ -9,8 +8,8 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 adapter: qlora
 lora_model_dir:
 sequence_len: 1024

examples/phi/phi-ft.yml

@@ -1,5 +1,4 @@
 base_model: microsoft/phi-1_5
-base_model_config: microsoft/phi-1_5
 model_type: MixFormerSequentialForCausalLM
 tokenizer_type: AutoTokenizer
 is_llama_derived_model: false
@@ -13,7 +12,7 @@ datasets:
   - path: garage-bAInd/Open-Platypus
     type: alpaca
 
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.05
 output_dir: ./phi-sft-out
 

examples/phi/phi-qlora.yml

@@ -1,5 +1,4 @@
 base_model: microsoft/phi-1_5
-base_model_config: microsoft/phi-1_5
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
 is_llama_derived_model: false
@@ -13,7 +12,7 @@ datasets:
   - path: garage-bAInd/Open-Platypus
     type: alpaca
 
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.05
 output_dir: ./phi-sft-out
 

examples/pythia-12b/config.yml

@@ -1,5 +1,4 @@
 base_model: EleutherAI/pythia-12b-deduped
-base_model_config: EleutherAI/pythia-12b-deduped
 base_model_ignore_patterns: pytorch*  # prefer safetensors
 model_type: GPTNeoXForCausalLM
 tokenizer_type: AutoTokenizer
@@ -10,7 +9,7 @@ device_map: auto
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.05
 adapter:
 lora_model_dir:

examples/pythia/lora.yml

@@ -1,10 +1,9 @@
 base_model: EleutherAI/pythia-1.4b-deduped
-base_model_config: EleutherAI/pythia-1.4b-deduped
 load_in_8bit: true
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.05
 adapter: lora
 lora_model_dir:
@@ -24,15 +23,15 @@ wandb_log_model:
 output_dir: ./lora-alpaca-pythia
 gradient_accumulation_steps: 1
 micro_batch_size: 4
-num_epochs: 3
+num_epochs: 4
 learning_rate: 0.00001
 train_on_inputs: false
 group_by_length: false
-bf16: True
-tf32: True
+bf16: true
+tf32: true
 early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
 weight_decay: 0.1
-eval_steps: 20
+eval_steps: 0.05
 logging_steps: 1

examples/redpajama/config-3b.yml

@@ -1,5 +1,4 @@
 base_model: togethercomputer/RedPajama-INCITE-Chat-3B-v1
-base_model_config: togethercomputer/RedPajama-INCITE-Chat-3B-v1
 model_type: GPTNeoXForCausalLM
 tokenizer_type: AutoTokenizer
 trust_remote_code:
@@ -7,7 +6,7 @@ load_in_8bit: false
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter:
 lora_model_dir:
@@ -28,7 +27,7 @@ wandb_log_model:
 output_dir: ./redpajama-alpaca-3b
 batch_size: 4
 micro_batch_size: 1
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine

examples/replit-3b/config-lora.yml

@@ -1,11 +1,10 @@
 base_model: replit/replit-code-v1-3b
-base_model_config: replit/replit-code-v1-3b
 trust_remote_code: true
 load_in_8bit: false
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.05
 adapter: lora
 lora_model_dir:
@@ -27,7 +26,7 @@ wandb_log_model:
 output_dir: ./lora-replit
 batch_size: 8
 micro_batch_size: 1
-num_epochs: 3
+num_epochs: 4
 optimizer:
 torchdistx_path:
 lr_scheduler:

examples/xgen-7b/xgen-7b-8k-qlora.yml

@@ -1,7 +1,6 @@
 # An example finetuning Saleforce's XGen-7b model with 8k context using qlora
 # on Tim Dettmer's Guanaco dataset.
 base_model: Salesforce/xgen-7b-8k-base
-base_model_config: Salesforce/xgen-7b-8k-base
 trust_remote_code: true
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
@@ -16,8 +15,8 @@ datasets:
     data_files:
       - openassistant_best_replies_train.jsonl
     type: "completion"
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 # enable QLoRA
 adapter: qlora
 lora_model_dir:
@@ -52,7 +51,7 @@ output_dir: ./qlora-out
 # decrease if OOM, increase for max VRAM utilization
 micro_batch_size: 1
 gradient_accumulation_steps: 1
-num_epochs: 3
+num_epochs: 4
 # Optimizer for QLoRA
 optimizer: paged_adamw_32bit
 torchdistx_path:

requirements.txt

@@ -1,23 +1,23 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/
 torch==2.0.1
-auto-gptq
+auto-gptq==0.4.2
 packaging
-peft @ git+https://github.com/huggingface/peft.git
-transformers @ git+https://github.com/huggingface/transformers.git
+peft==0.6.0
+transformers @ git+https://github.com/huggingface/transformers.git@acc394c4f5e1283c19783581790b3dc3105a3697
 bitsandbytes>=0.41.1
-accelerate @ git+https://github.com/huggingface/accelerate
+accelerate @ git+https://github.com/huggingface/accelerate@80da9cfb09bb3cc9f1b385cb55d6b90d025a5fd9
+deepspeed
 addict
-evaluate
 fire
 PyYAML>=6.0
 datasets
-flash-attn>=2.2.1
+flash-attn>=2.3.0
 sentencepiece
 wandb
 einops
-xformers
-optimum
+xformers>=0.0.22
+optimum==1.13.2
 hf_transfer
 colorama
 numba
@@ -30,3 +30,5 @@ scipy
 scikit-learn==1.2.2
 pynvml
 art
+fschat==0.2.29
+gradio

scripts/finetune.py

@@ -7,6 +7,7 @@ import transformers
 
 from axolotl.cli import (
     check_accelerate_default_config,
+    check_user_token,
     do_inference,
     do_merge_lora,
     load_cfg,
@@ -31,6 +32,7 @@ def do_cli(config: Path = Path("examples/"), **kwargs):
     )
     parsed_cfg = load_cfg(config, **kwargs)
     check_accelerate_default_config()
+    check_user_token()
     parser = transformers.HfArgumentParser((TrainerCliArgs))
     parsed_cli_args, _ = parser.parse_args_into_dataclasses(
         return_remaining_strings=True
@@ -43,8 +45,6 @@ def do_cli(config: Path = Path("examples/"), **kwargs):
         shard(cfg=parsed_cfg, cli_args=parsed_cli_args)
     else:
         dataset_meta = load_datasets(cfg=parsed_cfg, cli_args=parsed_cli_args)
-        if parsed_cli_args.prepare_ds_only:
-            return
         train(cfg=parsed_cfg, cli_args=parsed_cli_args, dataset_meta=dataset_meta)
 
 

setup.py

@@ -13,9 +13,22 @@ def parse_requirements():
                 # Handle custom index URLs
                 _, url = line.split()
                 _dependency_links.append(url)
-            elif "flash-attn" not in line and line and line[0] != "#":
+            elif (
+                "flash-attn" not in line
+                and "deepspeed" not in line
+                and line
+                and line[0] != "#"
+            ):
                 # Handle standard packages
                 _install_requires.append(line)
+
+    # TODO(wing) remove once xformers release supports torch 2.1.0
+    if "torch==2.1.0" in _install_requires:
+        _install_requires.pop(_install_requires.index("xformers>=0.0.22"))
+        _install_requires.append(
+            "xformers @ git+https://github.com/facebookresearch/xformers.git@main"
+        )
+
     return _install_requires, _dependency_links
 
 
@@ -33,9 +46,9 @@ setup(
     dependency_links=dependency_links,
     extras_require={
         "flash-attn": [
-            "flash-attn>=2.2.1",
+            "flash-attn>=2.3.0",
         ],
-        "extras": [
+        "deepspeed": [
             "deepspeed",
         ],
     },

src/axolotl/cli/__init__.py

@@ -6,15 +6,19 @@ import os
 import random
 import sys
 from pathlib import Path
+from threading import Thread
 from typing import Any, Dict, List, Optional, Union
 
+import gradio as gr
 import torch
 import yaml
 
 # add src to the pythonpath so we don't need to pip install this
 from accelerate.commands.config import config_args
 from art import text2art
-from transformers import GenerationConfig, TextStreamer
+from huggingface_hub import HfApi
+from huggingface_hub.utils import LocalTokenNotFoundError
+from transformers import GenerationConfig, TextIteratorStreamer, TextStreamer
 
 from axolotl.common.cli import TrainerCliArgs, load_model_and_tokenizer
 from axolotl.logging_config import configure_logging
@@ -49,7 +53,7 @@ def print_axolotl_text_art(suffix=None):
 
 
 def get_multi_line_input() -> Optional[str]:
-    print("Give me an instruction (Ctrl + D to finish): ")
+    print("Give me an instruction (Ctrl + D to submit): ")
     instruction = ""
     for line in sys.stdin:
         instruction += line  # pylint: disable=consider-using-join
@@ -151,6 +155,91 @@ def do_inference(
         print(tokenizer.decode(generated["sequences"].cpu().tolist()[0]))
 
 
+def do_inference_gradio(
+    *,
+    cfg: DictDefault,
+    cli_args: TrainerCliArgs,
+):
+    model, tokenizer = load_model_and_tokenizer(cfg=cfg, cli_args=cli_args)
+    prompter = cli_args.prompter
+    default_tokens = {"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>"}
+
+    for token, symbol in default_tokens.items():
+        # If the token isn't already specified in the config, add it
+        if not (cfg.special_tokens and token in cfg.special_tokens):
+            tokenizer.add_special_tokens({token: symbol})
+
+    prompter_module = None
+    if prompter:
+        prompter_module = getattr(
+            importlib.import_module("axolotl.prompters"), prompter
+        )
+
+    if cfg.landmark_attention:
+        from axolotl.monkeypatch.llama_landmark_attn import set_model_mem_id
+
+        set_model_mem_id(model, tokenizer)
+        model.set_mem_cache_args(
+            max_seq_len=255, mem_freq=50, top_k=5, max_cache_size=None
+        )
+
+    model = model.to(cfg.device)
+
+    def generate(instruction):
+        if not instruction:
+            return
+        if prompter_module:
+            # pylint: disable=stop-iteration-return
+            prompt: str = next(
+                prompter_module().build_prompt(instruction=instruction.strip("\n"))
+            )
+        else:
+            prompt = instruction.strip()
+        batch = tokenizer(prompt, return_tensors="pt", add_special_tokens=True)
+
+        model.eval()
+        with torch.no_grad():
+            generation_config = GenerationConfig(
+                repetition_penalty=1.1,
+                max_new_tokens=1024,
+                temperature=0.9,
+                top_p=0.95,
+                top_k=40,
+                bos_token_id=tokenizer.bos_token_id,
+                eos_token_id=tokenizer.eos_token_id,
+                pad_token_id=tokenizer.pad_token_id,
+                do_sample=True,
+                use_cache=True,
+                return_dict_in_generate=True,
+                output_attentions=False,
+                output_hidden_states=False,
+                output_scores=False,
+            )
+            streamer = TextIteratorStreamer(tokenizer)
+            generation_kwargs = {
+                "inputs": batch["input_ids"].to(cfg.device),
+                "generation_config": generation_config,
+                "streamer": streamer,
+            }
+
+            thread = Thread(target=model.generate, kwargs=generation_kwargs)
+            thread.start()
+
+            all_text = ""
+
+            for new_text in streamer:
+                all_text += new_text
+                yield all_text
+
+    demo = gr.Interface(
+        fn=generate,
+        inputs="textbox",
+        outputs="text",
+        title=cfg.get("gradio_title", "Axolotl Gradio Interface"),
+    )
+    demo.queue().launch(show_api=False, share=True)
+
+
 def choose_config(path: Path):
     yaml_files = list(path.glob("*.yml"))
 
@@ -192,6 +281,7 @@ def load_cfg(config: Path = Path("examples/"), **kwargs):
     # load the config from the yaml file
     with open(config, encoding="utf-8") as file:
         cfg: DictDefault = DictDefault(yaml.safe_load(file))
+    cfg.axolotl_config_path = config
     # if there are any options passed in the cli, if it is something that seems valid from the yaml,
     # then overwrite the value
     cfg_keys = cfg.keys()
@@ -219,7 +309,9 @@ def load_datasets(
 ) -> TrainDatasetMeta:
     tokenizer = load_tokenizer(cfg)
 
-    train_dataset, eval_dataset, total_num_steps = prepare_dataset(cfg, tokenizer)
+    train_dataset, eval_dataset, total_num_steps, prompters = prepare_dataset(
+        cfg, tokenizer
+    )
 
     if cli_args.debug or cfg.debug:
         LOG.info("check_dataset_labels...")
@@ -235,6 +327,10 @@ def load_datasets(
             text_only=cli_args.debug_text_only,
         )
 
+        LOG.info("printing prompters...")
+        for prompter in prompters:
+            LOG.info(prompter)
+
     return TrainDatasetMeta(
         train_dataset=train_dataset,
         eval_dataset=eval_dataset,
@@ -247,3 +343,16 @@ def check_accelerate_default_config():
         LOG.warning(
             f"accelerate config file found at {config_args.default_yaml_config_file}. This can lead to unexpected errors"
         )
+
+
+def check_user_token():
+    # Verify if token is valid
+    api = HfApi()
+    try:
+        user_info = api.whoami()
+        return bool(user_info)
+    except LocalTokenNotFoundError:
+        LOG.warning(
+            "Error verifying HuggingFace token. Remember to log in using `huggingface-cli login` and get your access token from https://huggingface.co/settings/tokens if you want to use gated models or datasets."
+        )
+        return False

src/axolotl/cli/inference.py

@@ -6,21 +6,30 @@ from pathlib import Path
 import fire
 import transformers
 
-from axolotl.cli import do_inference, load_cfg, print_axolotl_text_art
+from axolotl.cli import (
+    do_inference,
+    do_inference_gradio,
+    load_cfg,
+    print_axolotl_text_art,
+)
 from axolotl.common.cli import TrainerCliArgs
 
 
-def do_cli(config: Path = Path("examples/"), **kwargs):
+def do_cli(config: Path = Path("examples/"), gradio=False, **kwargs):
     # pylint: disable=duplicate-code
     print_axolotl_text_art()
     parsed_cfg = load_cfg(config, **kwargs)
+    parsed_cfg.sample_packing = False
     parser = transformers.HfArgumentParser((TrainerCliArgs))
     parsed_cli_args, _ = parser.parse_args_into_dataclasses(
         return_remaining_strings=True
     )
     parsed_cli_args.inference = True
 
-    do_inference(cfg=parsed_cfg, cli_args=parsed_cli_args)
+    if gradio:
+        do_inference_gradio(cfg=parsed_cfg, cli_args=parsed_cli_args)
+    else:
+        do_inference(cfg=parsed_cfg, cli_args=parsed_cli_args)
 
 
 if __name__ == "__main__":

src/axolotl/cli/preprocess.py

@@ -0,0 +1,53 @@
+"""
+CLI to run training on a model
+"""
+import logging
+from pathlib import Path
+
+import fire
+import transformers
+from colorama import Fore
+
+from axolotl.cli import (
+    check_accelerate_default_config,
+    check_user_token,
+    load_cfg,
+    load_datasets,
+    print_axolotl_text_art,
+)
+from axolotl.common.cli import PreprocessCliArgs
+from axolotl.common.const import DEFAULT_DATASET_PREPARED_PATH
+
+LOG = logging.getLogger("axolotl.cli.preprocess")
+
+
+def do_cli(config: Path = Path("examples/"), **kwargs):
+    # pylint: disable=duplicate-code
+    print_axolotl_text_art()
+    parsed_cfg = load_cfg(config, **kwargs)
+    check_accelerate_default_config()
+    check_user_token()
+    parser = transformers.HfArgumentParser((PreprocessCliArgs))
+    parsed_cli_args, _ = parser.parse_args_into_dataclasses(
+        return_remaining_strings=True
+    )
+    if not parsed_cfg.dataset_prepared_path:
+        msg = (
+            Fore.RED
+            + "preprocess CLI called without dataset_prepared_path set, "
+            + f"using default path: {DEFAULT_DATASET_PREPARED_PATH}"
+            + Fore.RESET
+        )
+        LOG.warning(msg)
+        parsed_cfg.dataset_prepared_path = DEFAULT_DATASET_PREPARED_PATH
+
+    _ = load_datasets(cfg=parsed_cfg, cli_args=parsed_cli_args)
+    LOG.info(
+        Fore.GREEN
+        + f"Success! Preprocessed data path: `dataset_prepared_path: {parsed_cfg.dataset_prepared_path}`"
+        + Fore.RESET
+    )
+
+
+if __name__ == "__main__":
+    fire.Fire(do_cli)

src/axolotl/cli/train.py

@@ -1,6 +1,7 @@
 """
 CLI to run training on a model
 """
+import logging
 from pathlib import Path
 
 import fire
@@ -8,6 +9,7 @@ import transformers
 
 from axolotl.cli import (
     check_accelerate_default_config,
+    check_user_token,
     load_cfg,
     load_datasets,
     print_axolotl_text_art,
@@ -15,20 +17,20 @@ from axolotl.cli import (
 from axolotl.common.cli import TrainerCliArgs
 from axolotl.train import train
 
+LOG = logging.getLogger("axolotl.cli.train")
+
 
 def do_cli(config: Path = Path("examples/"), **kwargs):
     # pylint: disable=duplicate-code
     print_axolotl_text_art()
     parsed_cfg = load_cfg(config, **kwargs)
     check_accelerate_default_config()
+    check_user_token()
     parser = transformers.HfArgumentParser((TrainerCliArgs))
     parsed_cli_args, _ = parser.parse_args_into_dataclasses(
         return_remaining_strings=True
     )
-
     dataset_meta = load_datasets(cfg=parsed_cfg, cli_args=parsed_cli_args)
-    if parsed_cli_args.prepare_ds_only:
-        return
     train(cfg=parsed_cfg, cli_args=parsed_cli_args, dataset_meta=dataset_meta)
 
 

src/axolotl/common/cli.py

@@ -25,11 +25,22 @@ class TrainerCliArgs:
     debug_num_examples: int = field(default=5)
     inference: bool = field(default=False)
     merge_lora: bool = field(default=False)
-    prepare_ds_only: bool = field(default=False)
     prompter: Optional[str] = field(default=None)
     shard: bool = field(default=False)
 
 
+@dataclass
+class PreprocessCliArgs:
+    """
+    dataclass representing arguments for preprocessing only
+    """
+
+    debug: bool = field(default=False)
+    debug_text_only: bool = field(default=False)
+    debug_num_examples: int = field(default=1)
+    prompter: Optional[str] = field(default=None)
+
+
 def load_model_and_tokenizer(
     *,
     cfg: DictDefault,

src/axolotl/common/const.py

@@ -0,0 +1,5 @@
+"""
+Various shared constants
+"""
+
+DEFAULT_DATASET_PREPARED_PATH = "last_run_prepared"

src/axolotl/core/trainer_builder.py

@@ -0,0 +1,755 @@
+"""
+Builder for the training args and trainer
+"""
+
+import abc
+import importlib
+import logging
+import math
+import sys
+from abc import abstractmethod
+from dataclasses import dataclass, field
+from functools import partial
+from pathlib import Path
+from typing import Optional, Union
+
+import torch
+import transformers
+from datasets import Dataset
+from torch.optim.lr_scheduler import OneCycleLR
+from torch.utils.data import BatchSampler, DataLoader, RandomSampler, SequentialSampler
+from transformers import EarlyStoppingCallback, Trainer, TrainingArguments
+from transformers.trainer_utils import seed_worker
+
+from axolotl.monkeypatch.relora import ReLoRACallback, ReLoRAScheduler
+from axolotl.utils.callbacks import (
+    EvalFirstStepCallback,
+    GPUStatsCallback,
+    SaveAxolotlConfigtoWandBCallback,
+    SaveBetterTransformerModelCallback,
+    bench_eval_callback_factory,
+    log_prediction_callback_factory,
+)
+from axolotl.utils.collators import BatchSamplerDataCollatorForSeq2Seq
+from axolotl.utils.dataloader import MultipackDistributedDataloader
+from axolotl.utils.samplers import MultipackBatchSampler
+from axolotl.utils.schedulers import get_cosine_schedule_with_quadratic_warmup
+
+try:
+    import torch._dynamo  # pylint: disable=ungrouped-imports
+except ImportError:
+    pass
+
+LOG = logging.getLogger("axolotl.core.trainer_builder")
+
+
+@dataclass
+class AxolotlTrainingArguments(TrainingArguments):
+    """
+    Extend the base TrainingArguments for axolotl helpers
+    """
+
+    lr_quadratic_warmup: bool = field(
+        default=False,
+        metadata={"help": "Use quadratic warmup for cosine scheduling."},
+    )
+    sample_packing: bool = field(
+        default=False,
+        metadata={"help": "Use sample packing for efficient training."},
+    )
+    eval_sample_packing: Optional[bool] = field(
+        default=None,
+        metadata={"help": "Use sample packing for efficient evals."},
+    )
+    sample_packing_efficiency: float = field(
+        default=1.0,
+        metadata={"help": "Sample packing efficiency for calculating batch length."},
+    )
+    max_seq_length: int = field(
+        default=2048,
+        metadata={"help": "The maximum sequence length the model can handle"},
+    )
+    sample_packing_seq_len_multiplier: int = field(
+        default=1,
+        metadata={"help": "the multiplier for the max len for packed sequences"},
+    )
+    relora_steps: Optional[int] = field(
+        default=None,
+        metadata={"help": "how often to reset for ReLoRA"},
+    )
+    relora_warmup_steps: Optional[int] = field(
+        default=None,
+        metadata={"help": "how many warmup steps to take after reset for ReLoRA"},
+    )
+    bench_split: Optional[str] = field(
+        default="eval", metadata={"help": "The benchmark split to run on"}
+    )
+    bench_dataset: Optional[str] = field(
+        default="pharaouk/dharma-1/dharma_1_mini.json",
+        metadata={
+            "help": "Benchmark dataset to use: options are `mmlu-zs`, `mmlu-fs`, or the full path to the dataset file"
+        },
+    )
+    do_bench_eval: Optional[bool] = field(
+        default=False, metadata={"help": "Whether to run the Benchmark evaluation."}
+    )
+    max_bench_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": "If set, only evaluates on `max_bench_samples` of the benchmark dataset."
+        },
+    )
+    bench_source_max_len: int = field(
+        default=2048, metadata={"help": "Maximum source sequence length for bench."}
+    )
+    dataloader_prefetch_factor: Optional[int] = field(
+        default=None,
+        metadata={"help": "prefetch_factor argument to the dataloader"},
+    )
+
+
+class AxolotlTrainer(Trainer):
+    """
+    Extend the base Trainer for axolotl helpers
+    """
+
+    args = None  # type: AxolotlTrainingArguments
+
+    def __init__(self, *args, num_epochs=1, bench_data_collator=None, **kwargs):
+        self.num_epochs = num_epochs
+        self.bench_data_collator = bench_data_collator
+        super().__init__(*args, **kwargs)
+
+    def create_scheduler(
+        self, num_training_steps: int, optimizer: torch.optim.Optimizer = None
+    ):
+        """
+        Setup the scheduler. The optimizer of the trainer must have been set up either before this method is called or
+        passed as an argument.
+
+        Args:
+            num_training_steps (int): The number of training steps to do.
+            optimizer (torch.optim.Optimizer): The training optimizer
+        """
+
+        # fmt: off
+        if self.lr_scheduler is None:  # type: ignore  # pylint: disable=access-member-before-definition
+            # fmt: on
+            if (
+                self.args.lr_scheduler_type == "cosine"
+                and self.args.lr_quadratic_warmup is True
+            ):
+                self.lr_scheduler = get_cosine_schedule_with_quadratic_warmup(  # pylint: disable=attribute-defined-outside-init
+                    optimizer,
+                    num_warmup_steps=self.args.get_warmup_steps(num_training_steps),
+                    num_training_steps=num_training_steps,
+                )
+            else:
+                return super().create_scheduler(num_training_steps, optimizer)
+        return self.lr_scheduler
+
+    def _get_train_sampler(self) -> Optional[torch.utils.data.Sampler]:
+        if self.args.sample_packing:
+            return MultipackBatchSampler(
+                RandomSampler(self.train_dataset),
+                self.args.train_batch_size,
+                drop_last=True,
+                batch_max_len=self._train_batch_size * self.args.max_seq_length,
+                lengths=(
+                    self.train_dataset.data.column("position_ids")
+                    .to_pandas()
+                    .apply(lambda x: x[-1] + 1)
+                    .values
+                ),
+                packing_efficiency_estimate=self.args.sample_packing_efficiency,
+            )
+        return super()._get_train_sampler()
+
+    def _get_eval_sampler(
+        self, eval_dataset: Dataset
+    ) -> Optional[torch.utils.data.Sampler]:
+        if self.args.sample_packing and self.args.eval_sample_packing is not False:
+            return MultipackBatchSampler(
+                SequentialSampler(eval_dataset),
+                self.args.per_device_eval_batch_size,
+                drop_last=True,
+                batch_max_len=self.args.eval_batch_size * self.args.max_seq_length,
+                lengths=(
+                    eval_dataset.data.column("position_ids")
+                    .to_pandas()
+                    .apply(lambda x: x[-1] + 1)
+                    .values
+                ),
+                packing_efficiency_estimate=self.args.sample_packing_efficiency,
+            )
+        return super()._get_eval_sampler(eval_dataset)
+
+    def get_train_dataloader(self) -> DataLoader:
+        if self.args.sample_packing:
+            train_dataset = self.train_dataset
+            train_dataset = train_dataset.remove_columns(["length"])
+            data_collator = self.data_collator
+            dataloader_params = {
+                "batch_size": self._train_batch_size,
+                "collate_fn": data_collator,
+                "num_workers": self.args.dataloader_num_workers,
+                "pin_memory": self.args.dataloader_pin_memory,
+            }
+            if self.args.dataloader_prefetch_factor:
+                dataloader_params[
+                    "prefetch_factor"
+                ] = self.args.dataloader_prefetch_factor
+
+            sampler = self._get_train_sampler()
+            if isinstance(sampler, BatchSampler):
+                dataloader_params["batch_sampler"] = sampler
+                del dataloader_params["batch_size"]
+            else:
+                dataloader_params["sampler"] = sampler
+                dataloader_params["drop_last"] = self.args.dataloader_drop_last
+            dataloader_params["worker_init_fn"] = seed_worker
+
+            self.accelerator.even_batches = False
+            return self.accelerator.prepare_data_loader(
+                DataLoader(train_dataset, **dataloader_params)
+            )
+        return super().get_train_dataloader()
+
+    def get_eval_dataloader(
+        self, eval_dataset: Optional[Dataset] = None
+    ) -> Union[DataLoader, MultipackDistributedDataloader]:
+        if self.args.sample_packing and self.args.eval_sample_packing is not False:
+            eval_dataset = (
+                eval_dataset if eval_dataset is not None else self.eval_dataset
+            )
+
+            eval_sampler = self._get_eval_sampler(eval_dataset)
+            eval_dataset = eval_dataset.remove_columns(["length"])
+            data_collator = self.data_collator
+            dataloader_params = {
+                "batch_size": self.args.eval_batch_size,
+                "collate_fn": data_collator,
+                "num_workers": self.args.dataloader_num_workers,
+                "pin_memory": self.args.dataloader_pin_memory,
+            }
+            if self.args.dataloader_prefetch_factor:
+                dataloader_params[
+                    "prefetch_factor"
+                ] = self.args.dataloader_prefetch_factor
+
+            if isinstance(eval_sampler, BatchSampler):
+                dataloader_params["batch_sampler"] = eval_sampler
+                del dataloader_params["batch_size"]
+            else:
+                dataloader_params["sampler"] = eval_sampler
+                dataloader_params["drop_last"] = self.args.dataloader_drop_last
+
+            self.accelerator.even_batches = False
+            return self.accelerator.prepare_data_loader(
+                DataLoader(eval_dataset, **dataloader_params)
+            )
+        return super().get_eval_dataloader(eval_dataset)
+
+    def _get_bench_sampler(
+        self, bench_dataset: Dataset
+    ) -> Optional[torch.utils.data.Sampler]:
+        if self.args.world_size <= 1:
+            return SequentialSampler(bench_dataset)
+        return None
+
+    def get_bench_dataloader(
+        self,
+        bench_dataset: Dataset,
+    ) -> Union[DataLoader, MultipackDistributedDataloader]:
+        dataloader_params = {
+            "batch_size": self.args.eval_batch_size,
+            "collate_fn": self.bench_data_collator,
+            "num_workers": self.args.dataloader_num_workers,
+            "pin_memory": self.args.dataloader_pin_memory,
+        }
+        if self.args.dataloader_prefetch_factor:
+            dataloader_params["prefetch_factor"] = self.args.dataloader_prefetch_factor
+
+        if not isinstance(bench_dataset, torch.utils.data.IterableDataset):
+            dataloader_params["sampler"] = self._get_bench_sampler(bench_dataset)
+            dataloader_params["drop_last"] = self.args.dataloader_drop_last
+
+        return DataLoader(bench_dataset, **dataloader_params)
+        # return self.accelerator.prepare(DataLoader(bench_dataset, **dataloader_params))
+
+    def compute_loss(self, model, inputs, return_outputs=False):
+        # use one's weighted cross entropy loss calc
+        # if self.args.sample_packing:
+        #     labels = inputs.pop("labels")
+        #     outputs = model(**inputs)
+        #     loss = trainer_weighted_loss(outputs, labels, shift_labels=True)
+        #     return (loss, outputs) if return_outputs else loss
+        return super().compute_loss(model, inputs, return_outputs=return_outputs)
+
+
+class OneCycleLRSchedulerTrainer(AxolotlTrainer):
+    """
+    Trainer subclass that uses the OneCycleLR scheduler
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.lr_scheduler = None
+
+    def create_scheduler(
+        self,
+        num_training_steps: int,
+        optimizer: Optional[torch.optim.Optimizer] = None,
+    ):
+        optimizer = self.optimizer if optimizer is None else optimizer
+        num_warmup_steps = self.args.get_warmup_steps(num_training_steps)
+        pct_start = num_warmup_steps / num_training_steps
+
+        self.lr_scheduler = OneCycleLR(
+            optimizer,
+            max_lr=self.args.learning_rate,
+            total_steps=num_training_steps,
+            pct_start=pct_start,
+            div_factor=6,
+        )
+
+        return self.lr_scheduler
+
+
+class ReLoRATrainer(AxolotlTrainer):
+    """
+    Trainer subclass that uses the OneCycleLR scheduler
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.lr_scheduler = None
+
+    def create_scheduler(
+        self,
+        num_training_steps: int,
+        optimizer: Optional[torch.optim.Optimizer] = None,
+    ):
+        optimizer = self.optimizer if optimizer is None else optimizer
+        lr_scheduler = super().create_scheduler(num_training_steps, optimizer)
+
+        if self.args.relora_steps:
+            warmup_steps = (
+                self.args.relora_warmup_steps if self.args.relora_warmup_steps else 10
+            )
+            self.lr_scheduler = ReLoRAScheduler(
+                optimizer,
+                lr_scheduler,
+                self.args.relora_steps,
+                warmup_steps,
+            )
+        else:
+            self.lr_scheduler = lr_scheduler
+
+        return self.lr_scheduler
+
+
+class TrainerBuilderBase(abc.ABC):
+    """
+    Base class for trainer builder
+    """
+
+    _train_dataset = None
+    _eval_dataset = None
+
+    def __init__(self, cfg, model, tokenizer):
+        self.cfg = cfg
+        self.model = model
+        self.tokenizer = tokenizer
+
+    @property
+    def train_dataset(self):
+        return self._train_dataset
+
+    @train_dataset.setter
+    def train_dataset(self, dataset):
+        self._train_dataset = dataset
+
+    @property
+    def eval_dataset(self):
+        return self._eval_dataset
+
+    @eval_dataset.setter
+    def eval_dataset(self, dataset):
+        self._eval_dataset = dataset
+
+    @abstractmethod
+    def build(self, total_num_steps):
+        pass
+
+    @abstractmethod
+    def get_callbacks(self):
+        pass
+
+    @abstractmethod
+    def get_post_trainer_create_callbacks(self, trainer):
+        """
+        Callbacks added after the trainer is created, usually b/c these need access to the trainer
+        """
+
+
+class HFCausalTrainerBuilder(TrainerBuilderBase):
+    """
+    Build the HuggingFace training args/trainer for Causal models
+    """
+
+    def hook_pre_create_training_args(self, training_arguments_kwargs):
+        # TODO
+        return training_arguments_kwargs
+
+    def hook_post_create_training_args(self, training_arguments):
+        # TODO
+        return training_arguments
+
+    def hook_pre_create_trainer(self, trainer_kwargs, trainer_cls):
+        # TODO
+        return trainer_kwargs, trainer_cls
+
+    def hook_post_create_trainer(self, trainer):
+        # TODO
+        return trainer
+
+    def get_callbacks(self):
+        callbacks = []
+        callbacks.append(GPUStatsCallback(self.cfg))
+        callbacks.append(EvalFirstStepCallback)
+
+        if self.cfg.relora_steps:
+            callbacks.append(ReLoRACallback(self.cfg))
+
+        if (
+            hasattr(self.model, "use_bettertransformer")
+            and self.model.use_bettertransformer is True
+        ):
+            callbacks.append(SaveBetterTransformerModelCallback)
+
+        if self.cfg.use_wandb:
+            callbacks.append(
+                SaveAxolotlConfigtoWandBCallback(self.cfg.axolotl_config_path)
+            )
+
+        return callbacks
+
+    def get_post_trainer_create_callbacks(self, trainer):
+        callbacks = []
+        if self.cfg.use_wandb and self.cfg.eval_table_size > 0:
+            LogPredictionCallback = log_prediction_callback_factory(
+                trainer, self.tokenizer
+            )
+            callbacks.append(LogPredictionCallback(self.cfg))
+
+        if self.cfg.do_bench_eval:
+            callbacks.append(bench_eval_callback_factory(trainer, self.tokenizer))
+
+        if self.cfg.early_stopping_patience:
+            early_stop_cb = EarlyStoppingCallback(
+                self.cfg.early_stopping_patience,
+            )
+            callbacks.append(early_stop_cb)
+
+        return callbacks
+
+    def _get_trainer_cls(self):
+        if self.cfg.lr_scheduler == "one_cycle" and (
+            self.cfg.fsdp or self.cfg.adapter == "qlora"
+        ):
+            return OneCycleLRSchedulerTrainer
+        if self.cfg.relora_steps:
+            return ReLoRATrainer
+        return AxolotlTrainer
+
+    def build(self, total_num_steps):
+        warmup_steps = (
+            self.cfg.warmup_steps
+            if self.cfg.warmup_steps is not None
+            else min(int(0.03 * total_num_steps), 100)
+        )
+        logging_steps = (
+            self.cfg.logging_steps
+            if self.cfg.logging_steps is not None
+            else max(min(int(0.005 * total_num_steps), 10), 1)
+        )
+
+        training_arguments_kwargs = {}
+        if self.cfg.bf16 == "full":
+            training_arguments_kwargs["bf16_full_eval"] = True
+        else:
+            training_arguments_kwargs["bf16"] = self.cfg.bf16
+        training_arguments_kwargs["fp16"] = (
+            self.cfg.fp16 and not self.cfg.bf16
+        ) or False
+        if self.cfg.fp8:
+            training_arguments_kwargs["fp16"] = False
+            training_arguments_kwargs["bf16"] = False
+
+        training_arguments_kwargs["tf32"] = self.cfg.tf32
+        training_arguments_kwargs["warmup_steps"] = warmup_steps
+        training_arguments_kwargs["logging_steps"] = logging_steps
+
+        if self.cfg.seed:
+            training_arguments_kwargs["seed"] = self.cfg.seed
+
+        if self.cfg.gradient_checkpointing:
+            training_arguments_kwargs[
+                "gradient_checkpointing"
+            ] = self.cfg.gradient_checkpointing
+        if self.cfg.fsdp:
+            training_arguments_kwargs["fsdp"] = self.cfg.fsdp
+            if self.cfg.fsdp_config:
+                training_arguments_kwargs["fsdp_config"] = dict(self.cfg.fsdp_config)
+
+        # deepspeed
+        if self.cfg.deepspeed:
+            training_arguments_kwargs["deepspeed"] = self.cfg.deepspeed
+
+        if self.cfg.lr_quadratic_warmup is not None:
+            training_arguments_kwargs[
+                "lr_quadratic_warmup"
+            ] = self.cfg.lr_quadratic_warmup
+
+        if self.cfg.adam_beta1:
+            training_arguments_kwargs["adam_beta1"] = self.cfg.adam_beta1
+        if self.cfg.adam_beta2:
+            training_arguments_kwargs["adam_beta2"] = self.cfg.adam_beta2
+        if self.cfg.adam_epsilon:
+            training_arguments_kwargs["adam_epsilon"] = self.cfg.adam_epsilon
+        if self.cfg.max_grad_norm:
+            training_arguments_kwargs["max_grad_norm"] = self.cfg.max_grad_norm
+
+        if self.cfg.hub_model_id:
+            training_arguments_kwargs["hub_model_id"] = self.cfg.hub_model_id
+            training_arguments_kwargs["push_to_hub"] = True
+            training_arguments_kwargs["hub_private_repo"] = True
+
+            if self.cfg.hub_strategy:
+                training_arguments_kwargs["hub_strategy"] = self.cfg.hub_strategy
+
+        if self.cfg.save_safetensors:
+            training_arguments_kwargs["save_safetensors"] = self.cfg.save_safetensors
+
+        if self.cfg.sample_packing_eff_est:
+            training_arguments_kwargs[
+                "sample_packing_efficiency"
+            ] = self.cfg.sample_packing_eff_est
+
+        if self.cfg.dataloader_pin_memory is not None:
+            training_arguments_kwargs[
+                "dataloader_pin_memory"
+            ] = self.cfg.dataloader_pin_memory
+        if self.cfg.dataloader_num_workers is not None:
+            training_arguments_kwargs[
+                "dataloader_num_workers"
+            ] = self.cfg.dataloader_num_workers
+        if self.cfg.dataloader_prefetch_factor is not None:
+            training_arguments_kwargs[
+                "dataloader_prefetch_factor"
+            ] = self.cfg.dataloader_prefetch_factor
+
+        if self.cfg.eval_steps:
+            training_arguments_kwargs["evaluation_strategy"] = "steps"
+            training_arguments_kwargs["eval_steps"] = self.cfg.eval_steps
+        elif self.cfg.evaluation_strategy:
+            training_arguments_kwargs[
+                "evaluation_strategy"
+            ] = self.cfg.evaluation_strategy
+        elif self.cfg.val_set_size == 0:
+            # no eval set, so don't eval
+            training_arguments_kwargs["evaluation_strategy"] = "no"
+        else:
+            # we have an eval set, but no steps defined, default to use epoch
+            training_arguments_kwargs["evaluation_strategy"] = "epoch"
+
+        if self.cfg.save_steps:
+            training_arguments_kwargs["save_strategy"] = "steps"
+            training_arguments_kwargs["save_steps"] = self.cfg.save_steps
+        elif self.cfg.save_strategy:
+            training_arguments_kwargs["save_strategy"] = self.cfg.save_strategy
+        else:
+            # default to saving each epoch if not defined
+            training_arguments_kwargs["save_strategy"] = "epoch"
+
+        if self.cfg.do_bench_eval:
+            training_arguments_kwargs["do_bench_eval"] = self.cfg.do_bench_eval
+            if self.cfg.bench_dataset:
+                training_arguments_kwargs["bench_dataset"] = self.cfg.bench_dataset
+        if self.cfg.metric_for_best_model:
+            training_arguments_kwargs[
+                "metric_for_best_model"
+            ] = self.cfg.metric_for_best_model
+        if self.cfg.greater_is_better:
+            training_arguments_kwargs["greater_is_better"] = self.cfg.greater_is_better
+
+        if self.cfg.torch_compile:
+            if torch.__version__ < "2.1.0":  # pylint: disable=protected-access
+                LOG.warning("torch>=2.1.0 required for torch_compile to work properly")
+            elif torch._dynamo:  # pylint: disable=protected-access
+                torch._dynamo.config.suppress_errors = (  # pylint: disable=protected-access
+                    True
+                )
+                training_arguments_kwargs["torch_compile"] = self.cfg.torch_compile
+                if self.cfg.torch_compile_backend:
+                    training_arguments_kwargs[
+                        "torch_compile_backend"
+                    ] = self.cfg.torch_compile_backend
+
+        # DDP Config
+        if self.cfg.ddp_timeout:
+            training_arguments_kwargs["ddp_timeout"] = self.cfg.ddp_timeout
+        # see https://pytorch.org/docs/stable/generated/torch.nn.parallel.DistributedDataParallel.html
+        if self.cfg.ddp_bucket_cap_mb:
+            training_arguments_kwargs["ddp_bucket_cap_mb"] = self.cfg.ddp_bucket_cap_mb
+        if self.cfg.ddp_broadcast_buffers is not None:
+            training_arguments_kwargs[
+                "ddp_broadcast_buffers"
+            ] = self.cfg.ddp_broadcast_buffers
+
+        # these are all the "standard" kwargs that are def used
+        training_arguments_kwargs["max_steps"] = (
+            total_num_steps if self.cfg.max_steps else -1
+        )
+        training_arguments_kwargs["max_seq_length"] = self.cfg.sequence_len
+        training_arguments_kwargs[
+            "per_device_train_batch_size"
+        ] = self.cfg.micro_batch_size
+        training_arguments_kwargs[
+            "per_device_eval_batch_size"
+        ] = self.cfg.eval_batch_size
+        training_arguments_kwargs[
+            "gradient_accumulation_steps"
+        ] = self.cfg.gradient_accumulation_steps
+        training_arguments_kwargs[
+            "eval_accumulation_steps"
+        ] = self.cfg.gradient_accumulation_steps
+        training_arguments_kwargs["num_train_epochs"] = self.cfg.num_epochs
+        training_arguments_kwargs["learning_rate"] = self.cfg.learning_rate
+        training_arguments_kwargs["output_dir"] = self.cfg.output_dir
+        training_arguments_kwargs["save_total_limit"] = (
+            self.cfg.save_total_limit if self.cfg.save_total_limit else 4
+        )
+        training_arguments_kwargs["load_best_model_at_end"] = (
+            (
+                self.cfg.load_best_model_at_end is not False
+                or self.cfg.early_stopping_patience
+            )
+            and self.cfg.val_set_size > 0
+            and self.cfg.save_steps
+            and self.cfg.eval_steps
+            and self.cfg.save_steps % self.cfg.eval_steps == 0
+        ) or False
+        training_arguments_kwargs["ddp_find_unused_parameters"] = (
+            False if self.cfg.ddp else None
+        )
+        training_arguments_kwargs["group_by_length"] = self.cfg.group_by_length
+        training_arguments_kwargs["report_to"] = "wandb" if self.cfg.use_wandb else None
+        training_arguments_kwargs["run_name"] = (
+            self.cfg.wandb_run_id if self.cfg.use_wandb else None
+        )
+        training_arguments_kwargs["optim"] = (
+            self.cfg.optimizer if self.cfg.optimizer else "adamw_hf"
+        )
+        training_arguments_kwargs["lr_scheduler_type"] = (
+            self.cfg.lr_scheduler
+            if self.cfg.lr_scheduler
+            and self.cfg.lr_scheduler not in ("one_cycle", "log_sweep")
+            else "cosine"
+        )
+        training_arguments_kwargs["weight_decay"] = (
+            self.cfg.weight_decay if self.cfg.weight_decay is not None else 0.0
+        )
+        training_arguments_kwargs["sample_packing"] = (
+            self.cfg.sample_packing if self.cfg.sample_packing else False
+        )
+        training_arguments_kwargs["eval_sample_packing"] = (
+            self.cfg.sample_packing if self.cfg.sample_packing else False
+        )
+        training_arguments_kwargs[
+            "sample_packing_seq_len_multiplier"
+        ] = self.cfg.micro_batch_size
+        training_arguments_kwargs["relora_steps"] = self.cfg.relora_steps
+        training_arguments_kwargs["relora_warmup_steps"] = self.cfg.relora_warmup_steps
+        training_arguments_kwargs = self.hook_pre_create_training_args(
+            training_arguments_kwargs
+        )
+        training_args = (
+            AxolotlTrainingArguments(  # pylint: disable=unexpected-keyword-arg
+                **training_arguments_kwargs,
+            )
+        )
+        training_args = self.hook_post_create_training_args(training_args)
+        trainer_kwargs = {}
+
+        if self.cfg.optimizer == "adamw_anyprecision":
+            if Path(self.cfg.torchdistx_path).exists():
+                sys.path.append(self.cfg.torchdistx_path)
+                importlib.import_module("torchdistx")
+
+        data_collator_kwargs = {
+            "padding": True,  # True/"longest" is the default
+        }
+        if self.cfg.pad_to_sequence_len:
+            data_collator_kwargs["pad_to_multiple_of"] = 64 * math.ceil(
+                self.cfg.sequence_len / 64
+            )
+        else:
+            # A100 is best at 64, while others at 8. Let's use the larger so we don't have to check
+            # https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html
+            data_collator_kwargs["pad_to_multiple_of"] = 64
+
+        if self.cfg.is_llama_derived_model and self.cfg.landmark_attention:
+            from axolotl.monkeypatch.llama_landmark_attn import (
+                add_mem_tokens,
+                get_mem_id,
+                set_model_mem_id,
+            )
+
+            set_model_mem_id(self.model, self.tokenizer)
+
+            LOG.info("Adding landmark attention tokens to dataset")
+
+            for dataset in [self.train_dataset, self.eval_dataset]:
+                dataset = dataset.map(
+                    partial(
+                        add_mem_tokens, mem_freq=50, mem_id=get_mem_id(self.tokenizer)
+                    ),
+                    batched=False,
+                    num_proc=32,
+                )
+
+        trainer_cls = self._get_trainer_cls()
+        trainer_kwargs, trainer_cls = self.hook_pre_create_trainer(
+            trainer_kwargs, trainer_cls
+        )
+        trainer = trainer_cls(
+            model=self.model,
+            train_dataset=self.train_dataset,
+            eval_dataset=self.eval_dataset,
+            args=training_args,
+            data_collator=BatchSamplerDataCollatorForSeq2Seq(
+                self.tokenizer,
+                return_tensors="pt",
+                **data_collator_kwargs,
+            ),
+            bench_data_collator=transformers.DataCollatorForSeq2Seq(
+                self.tokenizer,
+                return_tensors="pt",
+                **data_collator_kwargs,
+            ),
+            callbacks=self.get_callbacks(),
+            num_epochs=self.cfg.num_epochs,
+            **trainer_kwargs,
+        )
+        trainer = self.hook_post_create_trainer(trainer)
+        for callback in self.get_post_trainer_create_callbacks(trainer):
+            trainer.add_callback(callback)
+
+        if self.cfg.deepspeed and self.cfg.sample_packing:
+            trainer.accelerator.state.deepspeed_plugin.deepspeed_config[
+                "train_micro_batch_size_per_gpu"
+            ] = self.cfg.micro_batch_size
+
+        return trainer

src/axolotl/datasets.py

@@ -2,7 +2,7 @@
 
 import logging
 import os
-from typing import List
+from typing import List, Optional
 
 import torch
 from datasets import Dataset, IterableDataset
@@ -22,7 +22,7 @@ class TokenizedPromptDataset(Dataset):
     """
     Dataset that returns tokenized prompts from a stream of text files.
         Args:
-            prompt_tokenizer (PromptTokenizingStrategy): The prompt tokenizing method for proccessing the data.
+            prompt_tokenizer (PromptTokenizingStrategy): The prompt tokenizing method for processing the data.
             dataset (dataset.Dataset): Dataset with text files.
     """
 
@@ -30,18 +30,29 @@ class TokenizedPromptDataset(Dataset):
         self,
         prompt_tokenizer: PromptTokenizingStrategy,
         dataset: IterableDataset,
+        process_count: Optional[int] = None,
         **kwargs,
     ):
         self.prompt_tokenizer = prompt_tokenizer
+        self.process_count = process_count
         super().__init__(self.process(dataset).data, **kwargs)
 
     def process(self, dataset):
         features = dataset.features.keys()
-        num_proc = min(64, os.cpu_count())
+        num_proc = (
+            min(64, self.process_count)
+            if self.process_count
+            else min(64, os.cpu_count())
+        )
+        map_kwargs = {}
+        if self.prompt_tokenizer.supports_batched:
+            map_kwargs["batched"] = True
+            map_kwargs["batch_size"] = 100
         return dataset.map(
             self.prompt_tokenizer.tokenize_prompt,
             num_proc=num_proc,
             remove_columns=features,
+            **map_kwargs,
         )
 
 
@@ -50,7 +61,7 @@ class ConstantLengthDataset(IterableDataset):
     """
     Iterable dataset that returns constant length chunks of tokens from stream of text files.
         Args:
-            tokenizer (Tokenizer): The processor used for proccessing the data.
+            tokenizer (Tokenizer): The processor used for processing the data.
             dataset (dataset.Dataset): Dataset with text files.
             seq_length (int): Length of token sequences to return.
     """

src/axolotl/models/phi/modeling_mixformer_sequential.py

@@ -711,12 +711,8 @@ class ParallelBlock(nn.Module):
         self.resid_dropout = nn.Dropout(config.resid_pdrop)
         self.block_idx = block_idx
 
-        self.mixer = MHA(config=config, **mixer, layer_idx=block_idx)
-        mlp_cls = mlp.pop("mlp_cls")
-        if mlp_cls == "fused_mlp":
-            self.mlp = FusedMLP(config=config, **mlp)
-        else:
-            self.mlp = MLP(config=config, **mlp)
+        self.mixer = MHA(config, layer_idx=block_idx)
+        self.mlp = MLP(config)
 
     def forward(
         self,

src/axolotl/monkeypatch/btlm_attn_hijack_flash.py

@@ -6,8 +6,8 @@ import importlib
 import logging
 from typing import Optional, Tuple
 
-import accelerate
 import torch
+from accelerate import init_empty_weights
 from flash_attn.flash_attn_interface import flash_attn_func
 from transformers import AutoConfig, AutoModelForCausalLM
 
@@ -18,8 +18,8 @@ def replace_btlm_attn_with_flash_attn(model_name="cerebras/btlm-3b-8k-base"):
     # this is a wonky hack to get the remotely loaded module
     model_config = AutoConfig.from_pretrained(model_name, trust_remote_code=True)
     # we need to load the model here in order for modeling_btlm to be available
-    with accelerate.init_empty_weights():
-        AutoModelForCausalLM(model_config)
+    with init_empty_weights():
+        AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True)
     module_name = model_config.__class__.__module__.replace(
         ".configuration_btlm", ".modeling_btlm"
     )

src/axolotl/monkeypatch/falcon_attn_hijack_flash.py

@@ -1,101 +0,0 @@
-"""
-Flash Attention monkey patch for Falcon
-
-copied from https://github.com/pacman100/DHS-LLM-Workshop/blob/main/chat_assistant/training/falcon_flash_attn_monkey_patch.py
-"""
-
-from typing import Optional, Tuple
-
-import torch
-import transformers
-from flash_attn import flash_attn_func
-
-
-def forward(
-    self,
-    hidden_states: torch.Tensor,
-    alibi: Optional[torch.Tensor],
-    attention_mask: torch.Tensor,  # pylint: disable=unused-argument
-    layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
-    head_mask: Optional[torch.Tensor] = None,  # pylint: disable=unused-argument
-    use_cache: bool = False,
-    output_attentions: bool = False,  # pylint: disable=unused-argument
-):
-    fused_qkv = self.query_key_value(
-        hidden_states
-    )  # [batch_size, seq_length, 3 x hidden_size]
-    num_kv_heads = (
-        self.num_heads if self.new_decoder_architecture else self.num_kv_heads
-    )
-    # 3 x [batch_size, seq_length, num_heads, head_dim]
-    (
-        query_layer,
-        key_layer,
-        value_layer,
-    ) = self._split_heads(  # pylint: disable=protected-access
-        fused_qkv
-    )
-
-    batch_size, query_length, _, _ = query_layer.shape
-
-    query_layer = query_layer.transpose(1, 2).reshape(
-        batch_size * self.num_heads, query_length, self.head_dim
-    )
-    key_layer = key_layer.transpose(1, 2).reshape(
-        batch_size * num_kv_heads,
-        query_length,
-        self.head_dim,
-    )
-    value_layer = value_layer.transpose(1, 2).reshape(
-        batch_size * num_kv_heads, query_length, self.head_dim
-    )
-
-    past_kv_length = 0 if layer_past is None else layer_past[0].shape[1]
-    query_layer, key_layer = self.maybe_rotary(query_layer, key_layer, past_kv_length)
-
-    if layer_past is not None:
-        past_key, past_value = layer_past
-        # concatenate along seq_length dimension:
-        #  - key: [batch_size * self.num_heads, kv_length, head_dim]
-        #  - value: [batch_size * self.num_heads, kv_length, head_dim]
-        key_layer = torch.cat((past_key, key_layer), dim=1)
-        value_layer = torch.cat((past_value, value_layer), dim=1)
-
-    # unused
-    # _, kv_length, _ = key_layer.shape
-    if use_cache:
-        present = (key_layer, value_layer)
-    else:
-        present = None
-    # unused
-    # attention_mask_float = (attention_mask * 1.0).masked_fill(attention_mask, float("-1e9")).to(query_layer.dtype)
-    query_layer_ = (
-        query_layer.reshape(batch_size, self.num_heads, -1, self.head_dim)
-        .transpose(1, 2)
-        .to(torch.bfloat16)
-    )
-    key_layer_ = (
-        key_layer.reshape(batch_size, num_kv_heads, -1, self.head_dim)
-        .transpose(1, 2)
-        .to(torch.bfloat16)
-    )
-    value_layer_ = (
-        value_layer.reshape(batch_size, num_kv_heads, -1, self.head_dim)
-        .transpose(1, 2)
-        .to(torch.bfloat16)
-    )
-
-    if alibi is not None:
-        raise ValueError("`alibi` is not supported when `use_flash_attn` is True")
-
-    # below output will have shape (batch_size, seqlen, nheads, headdim)
-    attn_output = flash_attn_func(query_layer_, key_layer_, value_layer_, causal=True)
-    attn_output = attn_output.reshape(
-        batch_size, query_length, self.num_heads * self.head_dim
-    )
-    output_tensor = self.dense(attn_output)
-    return output_tensor, present
-
-
-def replace_falcon_attn_with_flash_attn():
-    transformers.models.falcon.modeling_falcon.FalconAttention.forward = forward

src/axolotl/monkeypatch/fastchat_conversation_turns.py

@@ -0,0 +1,174 @@
+"""
+monkeypatch to add a get_turns method
+"""
+
+import logging
+from typing import Generator, Tuple
+
+from fastchat.conversation import SeparatorStyle
+
+LOG = logging.getLogger("axolotl.monkeypatch.fastchat_conversation_turns")
+
+
+def get_prompt(self) -> str:
+    ret = ""
+    for role, msg in self.get_turns():
+        ret += role + msg
+    return ret
+
+
+def get_turns(  # pylint: disable=too-many-return-statements
+    self,
+) -> Generator[Tuple[str, str], None, None]:
+    """Get the prompt for generation."""
+    system_prompt = self.system_template.format(system_message=self.system_message)
+    if self.sep_style == SeparatorStyle.ADD_COLON_SINGLE:
+        yield "", system_prompt + self.sep
+        for role, message in self.messages:
+            if message:
+                yield role + ": ", message + self.sep
+            else:
+                yield role + ":", ""
+        return
+    if self.sep_style == SeparatorStyle.ADD_COLON_TWO:
+        seps = [self.sep, self.sep2]
+        yield "", system_prompt + seps[0]
+        for i, (role, message) in enumerate(self.messages):
+            if message:
+                yield role + ": ", message + seps[i % 2]
+            else:
+                yield role + ":", ""
+        return
+    if self.sep_style == SeparatorStyle.ADD_COLON_SPACE_SINGLE:
+        yield "", system_prompt + self.sep
+        for role, message in self.messages:
+            if message:
+                yield role + ": ", message + self.sep
+            else:
+                yield role + ": ", ""  # must be end with a space
+        return
+    if self.sep_style == SeparatorStyle.ADD_NEW_LINE_SINGLE:
+        yield "", "" if system_prompt == "" else system_prompt + self.sep
+        for role, message in self.messages:
+            if message:
+                yield role + "\n", message + self.sep
+            else:
+                yield role + "\n", ""
+        return
+    if self.sep_style == SeparatorStyle.NO_COLON_SINGLE:
+        yield "", system_prompt
+        for role, message in self.messages:
+            if message:
+                yield role, message + self.sep
+            else:
+                yield role, ""
+        return
+    if self.sep_style == SeparatorStyle.NO_COLON_TWO:
+        seps = [self.sep, self.sep2]
+        yield "", system_prompt
+        for i, (role, message) in enumerate(self.messages):
+            if message:
+                yield role, message + seps[i % 2]
+            else:
+                yield role, ""
+        return
+    if self.sep_style == SeparatorStyle.RWKV:
+        yield "", system_prompt
+        for i, (role, message) in enumerate(self.messages):
+            if message:
+                yield role + ": ", message.replace("\r\n", "\n").replace(
+                    "\n\n", "\n"
+                ) + "\n\n"
+            else:
+                yield role + ":", ""
+        return
+    if self.sep_style == SeparatorStyle.LLAMA2:
+        seps = [self.sep, self.sep2]
+        if self.system_message:
+            yield "", system_prompt
+        else:
+            yield "", "[INST] "
+        for i, (role, message) in enumerate(self.messages[1:]):
+            if message:
+                yield role + " ", message + seps[i % 2]
+            else:
+                yield role, ""
+        return
+    if self.sep_style == SeparatorStyle.CHATGLM:
+        # source: https://huggingface.co/THUDM/chatglm-6b/blob/1d240ba371910e9282298d4592532d7f0f3e9f3e/modeling_chatglm.py#L1302-L1308
+        # source2: https://huggingface.co/THUDM/chatglm2-6b/blob/e186c891cf64310ac66ef10a87e6635fa6c2a579/modeling_chatglm.py#L926
+        round_add_n = 1 if self.name == "chatglm2" else 0
+        if system_prompt:
+            yield "", system_prompt + self.sep
+
+        for i, (role, message) in enumerate(self.messages):
+            if i % 2 == 0:
+                yield "", f"[Round {i//2 + round_add_n}]{self.sep}"
+
+            if message:
+                yield f"{role}：", f"{message}{self.sep}"
+            else:
+                yield f"{role}：", ""
+        return
+    if self.sep_style == SeparatorStyle.CHATML:
+        yield "", "" if system_prompt == "" else system_prompt + self.sep + "\n"
+        for role, message in self.messages:
+            if message:
+                yield role + "\n", message + self.sep + "\n"
+            else:
+                yield role + "\n", ""
+        return
+    if self.sep_style == SeparatorStyle.CHATINTERN:
+        # source: https://huggingface.co/internlm/internlm-chat-7b-8k/blob/bd546fa984b4b0b86958f56bf37f94aa75ab8831/modeling_internlm.py#L771
+        seps = [self.sep, self.sep2]
+        yield "", system_prompt
+        for i, (role, message) in enumerate(self.messages):
+            prefix = "<s>" if i % 2 == 0 else ""
+            if message:
+                yield prefix + role + ":", message + seps[i % 2] + "\n"
+            else:
+                yield role + ":", ""
+        return
+    if self.sep_style == SeparatorStyle.DOLLY:
+        seps = [self.sep, self.sep2]
+        yield "", system_prompt
+        for i, (role, message) in enumerate(self.messages):
+            if message:
+                suffix = "\n\n" if i % 2 == 1 else ""
+                yield role + ":\n", message + seps[i % 2] + suffix
+            else:
+                yield role + ":\n", ""
+        return
+    if self.sep_style == SeparatorStyle.PHOENIX:
+        yield "", system_prompt
+        for role, message in self.messages:
+            if message:
+                yield role + ": ", "<s>" + message + "</s>"
+            else:
+                yield role + ": " + "<s>", ""
+        return
+    if self.sep_style == SeparatorStyle.ROBIN:
+        yield "", system_prompt + self.sep
+        for role, message in self.messages:
+            if message:
+                yield role + ":\n", message + self.sep
+            else:
+                yield role + ":\n", ""
+        return
+    if self.sep_style == SeparatorStyle.FALCON_CHAT:
+        if self.system_message:
+            yield "", system_prompt + self.sep
+        for role, message in self.messages:
+            if message:
+                yield role + ": ", message + self.sep
+            else:
+                yield role + ":", ""
+    else:
+        raise ValueError(f"Invalid style: {self.sep_style}")
+
+
+def add_get_turns_to_conversation():
+    import fastchat.conversation
+
+    fastchat.conversation.Conversation.get_turns = get_turns
+    fastchat.conversation.Conversation.get_prompt = get_prompt

src/axolotl/monkeypatch/llama_attn_hijack_flash.py

@@ -13,12 +13,18 @@ import transformers
 from einops import rearrange
 from flash_attn.bert_padding import pad_input, unpad_input
 from transformers.modeling_outputs import BaseModelOutputWithPast
+from transformers.models.llama.modeling_llama import LlamaAttention
 from transformers.models.llama.modeling_llama import (
     LlamaDecoderLayer as OriginalLlamaDecoderLayer,
 )
-from transformers.models.llama.modeling_llama import apply_rotary_pos_emb, repeat_kv
+from transformers.models.llama.modeling_llama import (
+    LlamaMLP,
+    apply_rotary_pos_emb,
+    repeat_kv,
+)
+from xformers.ops import SwiGLU
 
-from axolotl.monkeypatch.utils import get_cu_seqlens_from_pos_ids
+from axolotl.monkeypatch.utils import get_cu_seqlens_from_pos_ids, set_module_name
 
 try:
     from flash_attn.flash_attn_interface import (  # pylint: disable=ungrouped-imports
@@ -38,7 +44,33 @@ except ImportError:
 LOG = logging.getLogger("axolotl")
 
 
-def replace_llama_attn_with_flash_attn(packed: Optional[bool] = False):
+def replace_llama_mlp_with_swiglu(model):
+    for name, module in model.named_modules():
+        if isinstance(module, LlamaMLP):
+            mlp = FusedMLP(
+                module.config, module.gate_proj, module.up_proj, module.down_proj
+            )
+            set_module_name(model, name, mlp)
+
+
+def replace_llama_qkv_with_fused(model):
+    for name, module in model.named_modules():
+        if isinstance(module, LlamaAttention):
+            qkv = FusedAttention(
+                module.config,
+                module.q_proj,
+                module.k_proj,
+                module.v_proj,
+                module.o_proj,
+            )
+            set_module_name(model, name, qkv)
+
+
+def replace_llama_attn_with_flash_attn(
+    packed: Optional[bool] = False,
+    cross_entropy: Optional[bool] = False,
+    rms_norm: Optional[bool] = False,
+):
     transformers.models.llama.modeling_llama.LlamaModel._prepare_decoder_attention_mask = (  # pylint: disable=protected-access
         _prepare_decoder_attention_mask
     )
@@ -49,34 +81,124 @@ def replace_llama_attn_with_flash_attn(packed: Optional[bool] = False):
             llama_model_forward
         )
 
-    try:
-        from flash_attn.losses.cross_entropy import CrossEntropyLoss
+    # skip only if explicitly disabled
+    if cross_entropy:
+        try:
+            from flash_attn.losses.cross_entropy import CrossEntropyLoss
 
-        LOG.info("patching with flash_attn.losses.cross_entropy")
-        transformers.models.llama.modeling_llama.CrossEntropyLoss = partial(
-            CrossEntropyLoss, inplace_backward=True
+            LOG.info("patching with flash_attn.losses.cross_entropy")
+            transformers.models.llama.modeling_llama.CrossEntropyLoss = partial(
+                CrossEntropyLoss, inplace_backward=True
+            )
+        except ImportError:
+            LOG.info(
+                "optimized flash-attention CrossEntropyLoss not found (run `pip install 'git+https://github.com/Dao-AILab/flash-attention.git#egg=xentropy_cuda_lib&subdirectory=csrc/xentropy'`)"
+            )
+
+    # skip only if explicitly disabled
+    if rms_norm:
+        try:
+            from flash_attn.ops.rms_norm import RMSNorm
+
+            class LlamaRMSNorm(RMSNorm):
+                """Patched LLamaRMSNorm"""
+
+                def __init__(self, hidden_size, eps=1e-6):
+                    super().__init__(hidden_size, eps=eps)
+
+            LOG.info("patching with flash_attn.ops.rms_norm")
+            transformers.models.llama.modeling_llama.LlamaRMSNorm = LlamaRMSNorm
+        except ImportError:
+            LOG.info(
+                "optimized flash-attention RMSNorm not found (run `pip install 'git+https://github.com/Dao-AILab/flash-attention.git#egg=dropout_layer_norm&subdirectory=csrc/layer_norm'`)"
+            )
+
+
+class FusedAttention(LlamaAttention):
+    """
+    Fused QKV Attention layer for incrementally improved training efficiency
+    """
+
+    def __init__(
+        self,
+        config,
+        q: torch.nn.Linear,  # pylint: disable=invalid-name
+        k: torch.nn.Linear,  # pylint: disable=invalid-name
+        v: torch.nn.Linear,  # pylint: disable=invalid-name
+        o: torch.nn.Linear,  # pylint: disable=invalid-name
+    ):
+        super().__init__(config)
+        self.config = config
+        self.init_device = next(iter(q.state_dict().values())).device
+
+        # define equivalent fused qkv projection
+        self.out_features: List[int] = [q.out_features, k.out_features, v.out_features]
+        self.qkv_proj = torch.nn.Linear(
+            q.in_features, sum(self.out_features), device=self.init_device, bias=False
         )
-    except ImportError:
-        LOG.info(
-            "optimized flash-attention CrossEntropyLoss not found (run `pip install 'git+https://github.com/Dao-AILab/flash-attention.git#egg=xentropy_cuda_lib&subdirectory=csrc/xentropy'`)"
+        self.o_proj = o
+
+        # overwrite initialized weights with pretrained weights
+        self.qkv_proj.weight.data = torch.cat(
+            (q.weight.data, k.weight.data, v.weight.data), dim=0
         )
 
-    try:
-        from flash_attn.ops.rms_norm import RMSNorm
-
-        class LlamaRMSNorm(RMSNorm):
-            """Patched LLamaRMSNorm"""
-
-            def __init__(self, hidden_size, eps=1e-6):
-                super().__init__(hidden_size, eps=eps)
-
-        LOG.info("patching with flash_attn.ops.rms_norm")
-        transformers.models.llama.modeling_llama.LlamaRMSNorm = LlamaRMSNorm
-    except ImportError:
-        LOG.info(
-            "optimized flash-attention RMSNorm not found (run `pip install 'git+https://github.com/Dao-AILab/flash-attention.git#egg=dropout_layer_norm&subdirectory=csrc/layer_norm'`)"
+    def _post_training(self, model, name):
+        q_proj, k_proj, v_proj = torch.split(
+            self.qkv_proj.weight.data, self.out_features, dim=0
         )
 
+        new_attn = LlamaAttention(self.config)
+        new_attn.q_proj.weight.data = q_proj
+        new_attn.k_proj.weight.data = k_proj
+        new_attn.v_proj.weight.data = v_proj
+        new_attn.o_proj.weight.data = self.o_proj.weight.data
+
+        set_module_name(model, name, new_attn)
+
+
+class FusedMLP(torch.nn.Module):
+    """
+    Fused MLP layer for incrementally improved training efficiency
+    """
+
+    def __init__(
+        self,
+        config,
+        gate_proj: torch.nn.Linear,
+        up_proj: torch.nn.Linear,
+        down_proj: torch.nn.Linear,
+    ):
+        super().__init__()
+        self.config = config
+        self.swiglu = SwiGLU(
+            in_features=config.hidden_size,
+            hidden_features=config.intermediate_size,
+            bias=False,
+            _pack_weights=True,
+        )
+        # overwrite initialized weights with pretrained weights
+        self.swiglu.w12.weight.data = torch.cat(
+            (gate_proj.weight.data, up_proj.weight.data), dim=0
+        )
+        self.swiglu.w3.weight.data = down_proj.weight.data
+
+    def _post_training(self, model, name):
+        w1, w2 = torch.split(  # pylint: disable=invalid-name
+            self.swiglu.w12.weight.data, self.config.intermediate_size, dim=0
+        )
+
+        # Assign the split weights back to the original layers
+        new_mlp = LlamaMLP(self.config)
+        new_mlp.gate_proj.weight.data = w1
+        new_mlp.up_proj.weight.data = w2
+        new_mlp.down_proj.weight.data = self.swiglu.w3.weight.data
+
+        set_module_name(model, name, new_mlp)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:  # pylint: disable=invalid-name
+        return self.swiglu(x)
+
 
 # Disable the transformation of the attention mask in LlamaModel as the flash attention
 # requires the attention mask to be the same as the key_padding_mask
@@ -99,6 +221,7 @@ def flashattn_forward(
     past_key_value: Optional[Tuple[torch.Tensor]] = None,
     output_attentions: bool = False,
     use_cache: bool = False,
+    padding_mask: Optional[torch.LongTensor] = None,  # pylint: disable=unused-argument
     cu_seqlens: Optional[torch.Tensor] = None,
     max_seqlen: Optional[torch.Tensor] = None,
 ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
@@ -138,9 +261,14 @@ def flashattn_forward(
         value_states = torch.cat(value_states, dim=-1)
 
     else:
-        query_states = self.q_proj(hidden_states)
-        key_states = self.k_proj(hidden_states)
-        value_states = self.v_proj(hidden_states)
+        if isinstance(self, FusedAttention):
+            query_states, key_states, value_states = self.qkv_proj(hidden_states).split(
+                self.out_features, dim=-1
+            )
+        else:
+            query_states = self.q_proj(hidden_states)
+            key_states = self.k_proj(hidden_states)
+            value_states = self.v_proj(hidden_states)
 
     query_states = query_states.view(
         bsz, q_len, self.num_heads, self.head_dim
@@ -476,6 +604,13 @@ def llama_model_forward(
             dtype=torch.bool,
             device=inputs_embeds.device,
         )
+        padding_mask = None
+    else:
+        if 0 in attention_mask:
+            padding_mask = attention_mask
+        else:
+            padding_mask = None
+
     attention_mask = (
         self._prepare_decoder_attention_mask(  # pylint: disable=protected-access
             attention_mask,
@@ -510,7 +645,9 @@ def llama_model_forward(
             def create_custom_forward(module):
                 def custom_forward(*inputs):
                     # None for past_key_value
-                    return module(*inputs)
+                    return module(
+                        *inputs,
+                    )
 
                 return custom_forward
 
@@ -519,9 +656,10 @@ def llama_model_forward(
                 hidden_states,
                 attention_mask,
                 position_ids,
-                None,
+                past_key_value,
                 output_attentions,
                 None,
+                padding_mask,
                 cu_seqlens,
                 max_seqlen,
             )
@@ -533,6 +671,7 @@ def llama_model_forward(
                 past_key_value=past_key_value,
                 output_attentions=output_attentions,
                 use_cache=use_cache,
+                padding_mask=padding_mask,
                 cu_seqlens=cu_seqlens,
                 max_seqlen=max_seqlen,
             )
@@ -579,6 +718,7 @@ class LlamaDecoderLayer(OriginalLlamaDecoderLayer):
         past_key_value: Optional[Tuple[torch.Tensor]] = None,
         output_attentions: Optional[bool] = False,
         use_cache: Optional[bool] = False,
+        padding_mask: Optional[torch.LongTensor] = None,
         cu_seqlens: Optional[torch.Tensor] = None,
         max_seqlen: Optional[torch.Tensor] = None,
     ) -> Tuple[
@@ -611,6 +751,7 @@ class LlamaDecoderLayer(OriginalLlamaDecoderLayer):
             past_key_value=past_key_value,
             output_attentions=output_attentions,
             use_cache=use_cache,
+            padding_mask=padding_mask,
             cu_seqlens=cu_seqlens,
             max_seqlen=max_seqlen,
         )

src/axolotl/monkeypatch/mistral_attn_hijack_flash.py

@@ -0,0 +1,640 @@
+"""Flash attention monkey patch for mistral model"""
+# pylint: disable=duplicate-code
+
+import logging
+from typing import List, Optional, Tuple, Union
+
+import torch
+import transformers
+from einops import rearrange
+from flash_attn.bert_padding import pad_input, unpad_input
+from flash_attn.flash_attn_interface import (  # pylint: disable=ungrouped-imports
+    flash_attn_kvpacked_func,
+    flash_attn_varlen_kvpacked_func,
+    flash_attn_varlen_qkvpacked_func,
+)
+from transformers.modeling_outputs import BaseModelOutputWithPast
+from transformers.models.mistral.modeling_mistral import (
+    MistralAttention as OriginalMistralAttention,
+)
+from transformers.models.mistral.modeling_mistral import (
+    MistralDecoderLayer as OriginalMistralDecoderLayer,
+)
+from transformers.models.mistral.modeling_mistral import apply_rotary_pos_emb, repeat_kv
+
+from axolotl.monkeypatch.utils import get_cu_seqlens_from_pos_ids
+
+LOG = logging.getLogger("axolotl.monkeypatch.mistral")
+
+
+def replace_mistral_attn_with_flash_attn(
+    packed: Optional[bool] = False,
+):
+    transformers.models.mistral.modeling_mistral.MistralModel._prepare_decoder_attention_mask = (  # pylint: disable=protected-access
+        _prepare_decoder_attention_mask
+    )
+    transformers.models.mistral.modeling_mistral.MistralAttention.forward = (
+        flashattn_forward
+    )
+    if packed:
+        transformers.models.mistral.modeling_mistral.MistralDecoderLayer = (
+            MistralDecoderLayer
+        )
+        transformers.models.mistral.modeling_mistral.MistralModel.forward = (
+            mistral_model_forward
+        )
+
+
+@torch.jit.script
+def _make_sliding_window_causal_mask(
+    bsz: int,
+    tgt_len: int,
+    dtype: torch.dtype,
+    device: torch.device,
+    past_key_values_length: int = 0,
+    sliding_window: int = 4096,
+):
+    """
+    Make causal mask used for sliding window attention
+    """
+    tensor = torch.full(
+        (tgt_len, tgt_len),
+        fill_value=1,
+        device=device,
+    )
+    mask = torch.tril(tensor, diagonal=0)
+    # make the mask banded to account for sliding window
+    # NOTE: HF implementation is wrong as of 14-10-2023 for torch.triu, needs +1
+    mask = torch.triu(mask, diagonal=-sliding_window + 1)
+    mask = torch.log(mask).to(dtype)
+
+    if past_key_values_length > 0:
+        mask = torch.cat(
+            [
+                torch.zeros(
+                    tgt_len, past_key_values_length, dtype=dtype, device=device
+                ),
+                mask,
+            ],
+            dim=-1,
+        )
+    return mask[None, None, :, :].expand(
+        bsz, 1, tgt_len, tgt_len + past_key_values_length
+    )
+
+
+# Disable the transformation of the attention mask in LlamaModel as the flash attention
+# requires the attention mask to be the same as the key_padding_mask
+def _prepare_decoder_attention_mask(
+    self,
+    attention_mask,
+    input_shape,
+    inputs_embeds,
+    past_key_values_length,
+    sliding_window,
+):  # pylint: disable=unused-argument
+    # [bsz, seq_len]
+    if attention_mask is None:
+        return attention_mask
+
+    # NOTE: attention mask and sliding masks are only broadcastable in certain scenarios.
+    # Without attention_mask.shape[0] == 1, error will trigger after eval loss but only when wandb is enabled.
+    if input_shape[-1] > 1 and attention_mask.shape[0] == 1:
+        sliding_window_mask = _make_sliding_window_causal_mask(
+            bsz=input_shape[0],
+            tgt_len=input_shape[1],
+            dtype=inputs_embeds.dtype,
+            device=inputs_embeds.device,
+            past_key_values_length=past_key_values_length,
+            sliding_window=sliding_window,
+        )
+        attention_mask = attention_mask + sliding_window_mask
+    else:
+        LOG.info("skipping sliding window mask, not broadcastable with attention mask")
+
+    return attention_mask
+
+
+def flashattn_forward(
+    self: OriginalMistralAttention,
+    hidden_states: torch.Tensor,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_value: Optional[Tuple[torch.Tensor]] = None,
+    output_attentions: bool = False,
+    use_cache: bool = False,
+    cu_seqlens: Optional[torch.Tensor] = None,
+    max_seqlen: Optional[torch.Tensor] = None,
+) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+    bsz, q_len, _ = hidden_states.size()
+
+    query_states = self.q_proj(hidden_states)
+    key_states = self.k_proj(hidden_states)
+    value_states = self.v_proj(hidden_states)
+
+    query_states = query_states.view(
+        bsz, q_len, self.num_heads, self.head_dim
+    ).transpose(1, 2)
+    key_states = key_states.view(
+        bsz, q_len, self.num_key_value_heads, self.head_dim
+    ).transpose(1, 2)
+    value_states = value_states.view(
+        bsz, q_len, self.num_key_value_heads, self.head_dim
+    ).transpose(1, 2)
+
+    kv_seq_len = key_states.shape[-2]
+    if past_key_value is not None:
+        kv_seq_len += past_key_value[0].shape[-2]
+    cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+    query_states, key_states = apply_rotary_pos_emb(
+        query_states, key_states, cos, sin, position_ids
+    )
+
+    use_sliding_windows = (
+        hasattr(self.config, "sliding_window") is not None
+        and kv_seq_len > self.config.sliding_window
+    )
+
+    if use_sliding_windows:
+        window_size = (self.config.sliding_window, self.config.sliding_window)
+    else:
+        window_size = (-1, -1)
+
+    if past_key_value is not None:
+        # Activate slicing cache only if the config has a value `sliding_windows` attribute
+        if (
+            hasattr(self.config, "sliding_window")
+            and kv_seq_len > self.config.sliding_window
+        ):
+            slicing_tokens = kv_seq_len - self.config.sliding_window
+
+            past_key = past_key_value[0]
+            past_value = past_key_value[1]
+
+            past_key = past_key[:, :, slicing_tokens:, :].contiguous()
+            past_value = past_value[:, :, slicing_tokens:, :].contiguous()
+
+            if past_key.shape[-2] != self.config.sliding_window - 1:
+                raise ValueError(
+                    f"past key much have a shape of (`batch_size, num_heads, self.config.sliding_window-1, head_dim`), got"
+                    f" {past_key.shape}"
+                )
+
+            past_key_value = (past_key, past_value) if use_cache else None
+
+        if past_key_value is not None:
+            key_states = torch.cat([past_key_value[0], key_states], dim=2)
+            value_states = torch.cat([past_key_value[1], value_states], dim=2)
+
+    past_key_value = (key_states, value_states) if use_cache else None
+
+    # repeat k/v heads if n_kv_heads < n_heads
+    key_states = repeat_kv(key_states, self.num_key_value_groups)
+    value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+    if self.training:
+        # during training q,k,v always have same seqlen
+        assert key_states.shape == query_states.shape
+        is_causal = True
+    else:
+        # turn off FA causal mask after first inference autoregressive iteration
+        # only on first autoregressive step q,k,v have same seqlen
+        is_causal = key_states.shape == query_states.shape
+
+    if cu_seqlens is not None and max_seqlen is not None and cu_seqlens.dim() == 1:
+        # special handling using sample packing
+        qkv = torch.stack(
+            [query_states, key_states, value_states], dim=2
+        )  # [bsz, nh, 3, q_len, hd]
+        qkv = qkv.transpose(1, 3)  # [bsz, q_len, 3, nh, hd]
+        qkv = rearrange(qkv, "b s ... -> (b s) ...")
+
+        output = flash_attn_varlen_qkvpacked_func(
+            qkv,
+            cu_seqlens,
+            max_seqlen,
+            0.0,
+            softmax_scale=None,
+            causal=True,
+            window_size=window_size,
+        )
+        output = rearrange(output, "(b s) ... -> b s ...", b=bsz)
+    elif query_states.shape == key_states.shape:
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+        qkv_unpad, cu_seqlens_q, max_seqlen_q, _, output_pad_fn = generate_qkv(
+            query_states,
+            key_states,
+            value_states,
+            qkvpacked=True,
+            # We have disabled _prepare_decoder_attention_mask in LlamaModel
+            # the attention_mask should be the same as the key_padding_mask
+            key_padding_mask=attention_mask,
+            query_padding_mask=attention_mask[:, -query_states.size(1) :]
+            if attention_mask is not None
+            else None,
+        )
+        output_unpad = flash_attn_varlen_qkvpacked_func(
+            qkv_unpad,
+            cu_seqlens_q,
+            max_seqlen_q,
+            0.0,
+            softmax_scale=None,
+            causal=is_causal,
+            window_size=window_size,
+        )
+        output = output_pad_fn(output_unpad)
+    else:
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+        if attention_mask is None or attention_mask.all().item():
+            output = flash_attn_kvpacked_func(
+                query_states,
+                torch.stack([key_states, value_states], 2),
+                causal=is_causal,
+                window_size=window_size,
+            )
+        else:
+            (  # pylint: disable=unbalanced-tuple-unpacking
+                q_unpad,
+                kv_unpad,
+                cu_seqlens_q,
+                cu_seqlens_k,
+                max_seqlen_q,
+                max_seqlen_k,
+                _,
+                _,
+                output_pad_fn,
+            ) = generate_qkv(
+                query_states,
+                key_states,
+                value_states,
+                kvpacked=True,
+                key_padding_mask=attention_mask,
+                query_padding_mask=attention_mask[:, -query_states.size(1) :]
+                if attention_mask is not None
+                else None,
+            )
+            if q_unpad.dtype != kv_unpad.dtype:
+                kv_unpad = kv_unpad.to(q_unpad.dtype)
+            output_unpad = flash_attn_varlen_kvpacked_func(
+                q_unpad,
+                kv_unpad,
+                cu_seqlens_q,
+                cu_seqlens_k,
+                max_seqlen_q,
+                max_seqlen_k,
+                0.0,
+                softmax_scale=None,
+                causal=is_causal,
+                window_size=window_size,
+            )
+            output = output_pad_fn(output_unpad)
+
+    attn_output = output
+    if attn_output.size() != (bsz, q_len, self.num_heads, self.head_dim):
+        raise ValueError(
+            f"`attn_output` should be of size {(bsz, q_len, self.num_heads, self.head_dim)}, but is"
+            f" {attn_output.size()}"
+        )
+    attn_output = rearrange(attn_output, "b s h d -> b s (h d)")
+
+    attn_output = self.o_proj(attn_output)
+
+    if not output_attentions:
+        attn_weights = None
+
+    return attn_output, attn_weights, past_key_value
+
+
+# based on https://github.com/Dao-AILab/flash-attention/blob/364a5b/tests/test_flash_attn.py#L38
+def generate_qkv(
+    q,
+    k,
+    v,
+    query_padding_mask=None,
+    key_padding_mask=None,
+    kvpacked=False,
+    qkvpacked=False,
+):  # pylint: disable=invalid-name,unnecessary-lambda-assignment
+    """
+    Arguments:
+        q: (batch_size, seqlen_q, nheads, d)
+        k: (batch_size, seqlen_k, nheads_k, d)
+        v: (batch_size, seqlen_k, nheads_k, d)
+        query_padding_mask: (batch_size, seqlen), bool
+        key_padding_mask: (batch_size, seqlen), bool
+    """
+    assert not (kvpacked and qkvpacked)
+    batch_size, seqlen_q, nheads, d = q.shape
+    _, seqlen_k, nheads_k, _ = k.shape
+    assert k.shape == (batch_size, seqlen_k, nheads_k, d)
+    assert v.shape == (batch_size, seqlen_k, nheads_k, d)
+
+    if query_padding_mask is not None:
+        q_unpad, indices_q, cu_seqlens_q, max_seqlen_q = unpad_input(
+            q, query_padding_mask
+        )
+
+        output_pad_fn = lambda output_unpad: pad_input(  # noqa: E731
+            output_unpad, indices_q, batch_size, seqlen_q
+        )
+
+    else:
+        q_unpad = rearrange(q, "b s h d -> (b s) h d")
+        cu_seqlens_q = torch.arange(
+            0,
+            (batch_size + 1) * seqlen_q,
+            step=seqlen_q,
+            dtype=torch.int32,
+            device=q_unpad.device,
+        )
+        max_seqlen_q = seqlen_q
+
+        output_pad_fn = lambda output_unpad: rearrange(  # noqa: E731
+            output_unpad, "(b s) h d -> b s h d", b=batch_size
+        )
+
+    if key_padding_mask is not None:
+        k_unpad, _, cu_seqlens_k, max_seqlen_k = unpad_input(k, key_padding_mask)
+        v_unpad, _, _, _ = unpad_input(v, key_padding_mask)
+    else:
+        k_unpad = rearrange(k, "b s h d -> (b s) h d")
+        v_unpad = rearrange(v, "b s h d -> (b s) h d")
+        cu_seqlens_k = torch.arange(
+            0,
+            (batch_size + 1) * seqlen_k,
+            step=seqlen_k,
+            dtype=torch.int32,
+            device=k_unpad.device,
+        )
+        max_seqlen_k = seqlen_k
+
+    if qkvpacked:
+        assert nheads == nheads_k
+        qkv_unpad = torch.stack([q_unpad, k_unpad, v_unpad], dim=1)
+        qkv = torch.stack([q, k, v], dim=2)
+        return (qkv_unpad, cu_seqlens_q, max_seqlen_q, qkv, output_pad_fn)
+
+    if kvpacked:
+        kv_unpad = torch.stack([k_unpad, v_unpad], dim=1)
+        kv = torch.stack([k, v], dim=2)
+        return (
+            q_unpad,
+            kv_unpad,
+            cu_seqlens_q,
+            cu_seqlens_k,
+            max_seqlen_q,
+            max_seqlen_k,
+            q,
+            kv,
+            output_pad_fn,
+        )
+
+    return (
+        q_unpad,
+        k_unpad,
+        v_unpad,
+        cu_seqlens_q,
+        cu_seqlens_k,
+        max_seqlen_q,
+        max_seqlen_k,
+        q,
+        k,
+        v,
+        output_pad_fn,
+    )
+
+
+def mistral_model_forward(
+    self,
+    input_ids: torch.LongTensor = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[List[torch.FloatTensor]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    return_dict: Optional[bool] = None,
+) -> Union[Tuple, BaseModelOutputWithPast]:
+    output_attentions = (
+        output_attentions
+        if output_attentions is not None
+        else self.config.output_attentions
+    )
+    output_hidden_states = (
+        output_hidden_states
+        if output_hidden_states is not None
+        else self.config.output_hidden_states
+    )
+    use_cache = use_cache if use_cache is not None else self.config.use_cache
+
+    return_dict = (
+        return_dict if return_dict is not None else self.config.use_return_dict
+    )
+
+    # retrieve input_ids and inputs_embeds
+    if input_ids is not None and inputs_embeds is not None:
+        raise ValueError(
+            "You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time"
+        )
+    if input_ids is not None:
+        batch_size, seq_length = input_ids.shape
+    elif inputs_embeds is not None:
+        batch_size, seq_length, _ = inputs_embeds.shape
+    else:
+        raise ValueError(
+            "You have to specify either decoder_input_ids or decoder_inputs_embeds"
+        )
+
+    seq_length_with_past = seq_length
+    past_key_values_length = 0
+
+    if past_key_values is not None:
+        past_key_values_length = past_key_values[0][0].shape[2]
+        seq_length_with_past = seq_length_with_past + past_key_values_length
+
+    cu_seqlens = None
+    max_seqlen = None
+    if position_ids is None:
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+        position_ids = torch.arange(
+            past_key_values_length,
+            seq_length + past_key_values_length,
+            dtype=torch.long,
+            device=device,
+        )
+        position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+    else:
+        position_ids = position_ids.view(-1, seq_length).long()
+        cu_seqlens, max_seqlen = get_cu_seqlens_from_pos_ids(position_ids)
+        cu_seqlens = cu_seqlens.squeeze()
+
+    if inputs_embeds is None:
+        inputs_embeds = self.embed_tokens(input_ids)
+    # embed positions
+    if attention_mask is None:
+        attention_mask = torch.ones(
+            (batch_size, seq_length_with_past),
+            dtype=torch.bool,
+            device=inputs_embeds.device,
+        )
+    attention_mask = (
+        self._prepare_decoder_attention_mask(  # pylint: disable=protected-access
+            attention_mask,
+            (batch_size, seq_length),
+            inputs_embeds,
+            past_key_values_length,
+            sliding_window=self.config.sliding_window,
+        )
+    )
+
+    hidden_states = inputs_embeds
+
+    if self.gradient_checkpointing and self.training:
+        if use_cache:
+            transformers.logger.warning_once(
+                "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+            )
+            use_cache = False
+
+    # decoder layers
+    all_hidden_states = () if output_hidden_states else None
+    all_self_attns = () if output_attentions else None
+    next_decoder_cache = () if use_cache else None
+
+    for idx, decoder_layer in enumerate(self.layers):
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        past_key_value = past_key_values[idx] if past_key_values is not None else None
+
+        if self.gradient_checkpointing and self.training:
+
+            def create_custom_forward(module):
+                def custom_forward(*inputs):
+                    # None for past_key_value
+                    return module(*inputs)
+
+                return custom_forward
+
+            layer_outputs = torch.utils.checkpoint.checkpoint(
+                create_custom_forward(decoder_layer),
+                hidden_states,
+                attention_mask,
+                position_ids,
+                past_key_value,
+                output_attentions,
+                None,
+                cu_seqlens,
+                max_seqlen,
+            )
+        else:
+            layer_outputs = decoder_layer(
+                hidden_states,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_value,
+                output_attentions=output_attentions,
+                use_cache=use_cache,
+                cu_seqlens=cu_seqlens,
+                max_seqlen=max_seqlen,
+            )
+
+        hidden_states = layer_outputs[0]
+
+        if use_cache:
+            next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
+
+        if output_attentions:
+            all_self_attns += (layer_outputs[1],)
+
+    hidden_states = self.norm(hidden_states)
+
+    # add hidden states from the last decoder layer
+    if output_hidden_states:
+        all_hidden_states += (hidden_states,)
+
+    next_cache = next_decoder_cache if use_cache else None
+    if not return_dict:
+        return tuple(
+            v
+            for v in [hidden_states, next_cache, all_hidden_states, all_self_attns]
+            if v is not None
+        )
+    return BaseModelOutputWithPast(
+        last_hidden_state=hidden_states,
+        past_key_values=next_cache,
+        hidden_states=all_hidden_states,
+        attentions=all_self_attns,
+    )
+
+
+class MistralDecoderLayer(OriginalMistralDecoderLayer):
+    """
+    patched version of MistralDecoderLayer to pass through the precalculated cu_seqlens
+    """
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+        cu_seqlens: Optional[torch.Tensor] = None,
+        max_seqlen: Optional[torch.Tensor] = None,
+    ) -> Tuple[
+        torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]
+    ]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
+                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            use_cache (`bool`, *optional*):
+                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
+                (see `past_key_values`).
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+            cu_seqlens (`torch.Tensor`, *optional*) cumulative sequence len when packing
+        """
+
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+
+        # Self Attention
+        hidden_states, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+            cu_seqlens=cu_seqlens,
+            max_seqlen=max_seqlen,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+
+        return outputs

src/axolotl/monkeypatch/neft_embeddings.py

@@ -0,0 +1,65 @@
+"""
+patches implemented through the trainer hooks to enable NEFT/noisy embeddings per https://arxiv.org/abs/2310.05914
+"""
+import torch
+from peft import PeftModel
+from transformers import PreTrainedModel
+
+
+def patch_neft(alpha, model):
+    embeddings = None
+    if isinstance(model, PreTrainedModel):
+        embeddings = model.get_input_embeddings()
+    if isinstance(model, PeftModel):
+        embeddings = model.base_model.get_input_embeddings()
+    if not embeddings:
+        raise ValueError(f"unhandled model class for neft: {model.__class__.__name__}")
+    embeddings.noisy_embedding_alpha = alpha
+    old_forward = embeddings.forward
+
+    # This hack seems to be needed to properly use a custom forward pass
+    # all credits to: https://discuss.pytorch.org/t/how-can-i-replace-the-forward-method-of-a-predefined-torchvision-model-with-my-customized-forward-function/54224/11
+    bound_method = neft_forward.__get__(  # pylint: disable=no-value-for-parameter
+        embeddings, embeddings.__class__
+    )
+    setattr(embeddings, "forward", bound_method)
+
+    embeddings._old_forward = old_forward  # pylint: disable=protected-access
+    return model
+
+
+def unpatch_neft(model):
+    embeddings = None
+    if isinstance(model, PreTrainedModel):
+        embeddings = model.get_input_embeddings()
+    if isinstance(model, PeftModel):
+        embeddings = model.base_model.get_input_embeddings()
+    if not embeddings:
+        raise ValueError(f"unhandled model class for neft: {model.__class__.__name__}")
+    if hasattr(embeddings, "_old_forward"):
+        embeddings.forward = embeddings._old_forward  # pylint: disable=protected-access
+        del embeddings._old_forward  # pylint: disable=protected-access
+        del embeddings.noisy_embedding_alpha
+
+
+def neft_forward(self, inputs: torch.Tensor):
+    embeddings = self._old_forward(inputs)  # pylint: disable=protected-access
+
+    if self.training:
+        dims = torch.tensor(embeddings.size(1) * embeddings.size(2))
+        mag_norm = self.noisy_embedding_alpha / torch.sqrt(dims)
+        embeddings = embeddings + torch.zeros_like(embeddings).uniform_(
+            -mag_norm, mag_norm
+        )
+
+    return embeddings
+
+
+def pretrain_hook(cfg, trainer):
+    if cfg.noisy_embedding_alpha:
+        trainer.model = patch_neft(cfg.noisy_embedding_alpha, trainer.model)
+
+
+def post_train_hook(cfg, trainer):
+    if cfg.noisy_embedding_alpha:
+        unpatch_neft(trainer.model)

src/axolotl/monkeypatch/stablelm_attn_hijack_flash.py

@@ -0,0 +1,415 @@
+# coding=utf-8
+# Copyright 2023 Stability AI, EleutherAI, and The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+# This code is based off the following work:
+# https://github.com/huggingface/transformers/blob/main/src/transformers/models/llama/modeling_llama.py
+# https://github.com/huggingface/transformers/blob/main/src/transformers/models/gpt_neox/modeling_gpt_neox.py
+""" PyTorch StableLM Epoch model. """
+import importlib
+import math
+from typing import Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+from accelerate import init_empty_weights
+from einops import rearrange
+from flash_attn.flash_attn_interface import (  # pylint: disable=ungrouped-imports
+    flash_attn_varlen_qkvpacked_func,
+)
+from torch import nn
+from transformers import AutoConfig, AutoModelForCausalLM
+from transformers.modeling_outputs import BaseModelOutputWithPast
+from transformers.utils import logging
+
+from axolotl.monkeypatch.utils import get_cu_seqlens_from_pos_ids
+
+logger = logging.get_logger(__name__)
+
+
+def replace_stablelm_attn_with_flash_attn(model_name="stabilityai/stablelm-3b-4e1t"):
+    # this is a wonky hack to get the remotely loaded module
+    model_config = AutoConfig.from_pretrained(model_name, trust_remote_code=True)
+    # we need to load the model here in order for modeling_stablelm_epoch to be available
+    with init_empty_weights():
+        AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True)
+    module_name = model_config.__class__.__module__.replace(
+        ".configuration_stablelm_epoch", ".modeling_stablelm_epoch"
+    )
+    modeling_stablelm = importlib.import_module(module_name)
+    modeling_stablelm.Attention.forward = (  # pylint: disable=protected-access
+        flashattn_attn
+    )
+    modeling_stablelm.StableLMEpochModel.forward = (  # pylint: disable=protected-access
+        stablelm_model_forward
+    )
+    modeling_stablelm.DecoderLayer.forward = (  # pylint: disable=protected-access
+        decoder_layer_forward
+    )
+
+
+def rotate_half(x: torch.Tensor):
+    """Rotates half the hidden dims of the input."""
+    # pylint: disable=invalid-name
+    x1, x2 = torch.chunk(x, 2, dim=-1)
+    return torch.cat((-x2, x1), dim=-1)
+
+
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids):
+    # The first two dimensions of cos and sin are always 1, so we can `squeeze` them.
+    # pylint: disable=invalid-name
+    cos = cos.squeeze(1).squeeze(0)  # [seq_len, dim]
+    sin = sin.squeeze(1).squeeze(0)  # [seq_len, dim]
+    cos = cos[position_ids].unsqueeze(1)  # [batch_size, 1, seq_len, dim]
+    sin = sin[position_ids].unsqueeze(1)  # [batch_size, 1, seq_len, dim]
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """
+    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(
+        batch, num_key_value_heads, n_rep, slen, head_dim
+    )
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+def flashattn_attn(
+    self,
+    hidden_states: torch.FloatTensor,
+    attention_mask: torch.FloatTensor,
+    position_ids: torch.LongTensor,
+    past_key_value: Optional[Tuple[torch.Tensor]] = None,
+    output_attentions: Optional[bool] = False,  # pylint: disable=unused-argument
+    use_cache: Optional[bool] = False,
+    cu_seqlens: Optional[torch.Tensor] = None,
+    max_seqlen: Optional[torch.Tensor] = None,
+) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+    bsz, q_len, _ = hidden_states.size()
+
+    query_states = self.q_proj(hidden_states)
+    key_states = self.k_proj(hidden_states)
+    value_states = self.v_proj(hidden_states)
+
+    query_states = query_states.view(
+        bsz, q_len, self.num_heads, self.head_dim
+    ).transpose(1, 2)
+    key_states = key_states.view(
+        bsz, q_len, self.num_key_value_heads, self.head_dim
+    ).transpose(1, 2)
+    value_states = value_states.view(
+        bsz, q_len, self.num_key_value_heads, self.head_dim
+    ).transpose(1, 2)
+
+    query_rot = query_states[..., : self.rotary_ndims]
+    query_pass = query_states[..., self.rotary_ndims :]
+    key_rot = key_states[..., : self.rotary_ndims]
+    key_pass = key_states[..., self.rotary_ndims :]
+
+    kv_seq_len = key_states.shape[-2]
+    if past_key_value is not None:
+        kv_seq_len += past_key_value[0].shape[-2]
+    cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+    query_states, key_states = apply_rotary_pos_emb(
+        query_rot, key_rot, cos, sin, position_ids
+    )
+
+    # [batch_size, num_heads, seq_len, head_dim]
+    query_states = torch.cat((query_states, query_pass), dim=-1)
+    key_states = torch.cat((key_states, key_pass), dim=-1)
+
+    if past_key_value is not None:
+        # Reuse k, v, self_attention
+        key_states = torch.cat((past_key_value[0], key_states), dim=2)
+        value_states = torch.cat((past_key_value[1], value_states), dim=2)
+
+    past_key_value = (key_states, value_states) if use_cache else None
+
+    # Repeat k/v heads if n_kv_heads < n_heads
+    key_states = repeat_kv(key_states, self.num_key_value_groups)
+    value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+    if cu_seqlens is not None and max_seqlen is not None and cu_seqlens.dim() == 1:
+        # special handling using sample packing
+        qkv = torch.stack(
+            [query_states, key_states, value_states], dim=2
+        )  # [bsz, nh, 3, q_len, hd]
+        qkv = qkv.transpose(1, 3)  # [bsz, q_len, 3, nh, hd]
+        qkv = rearrange(qkv, "b s ... -> (b s) ...")
+        softmax_scale = None
+
+        output = flash_attn_varlen_qkvpacked_func(
+            qkv, cu_seqlens, max_seqlen, 0.0, softmax_scale=softmax_scale, causal=True
+        )
+
+        attn_output = rearrange(output, "(b s) ... -> b s ...", b=bsz)
+        attn_output = rearrange(attn_output, "b s h d -> b s (h d)")
+    else:
+        attn_weights = torch.matmul(
+            query_states, key_states.transpose(2, 3)
+        ) / math.sqrt(self.head_dim)
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz, self.num_heads, q_len, kv_seq_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+            attn_weights = attn_weights + attention_mask
+
+        # Upcast attention to fp32
+        attn_weights = nn.functional.softmax(
+            attn_weights, dim=-1, dtype=torch.float32
+        ).to(query_states.dtype)
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        # Merge heads
+        attn_output = attn_output.transpose(1, 2).contiguous()
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+
+    # Final linear projection
+    attn_output = self.o_proj(attn_output)
+
+    return attn_output, None, past_key_value
+
+
+def decoder_layer_forward(
+    self,
+    hidden_states: Optional[torch.FloatTensor],
+    attention_mask: Optional[torch.FloatTensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_value: Optional[Tuple[torch.Tensor]] = None,
+    output_attentions: Optional[bool] = False,
+    use_cache: Optional[bool] = False,
+    cu_seqlens: Optional[torch.Tensor] = None,
+    max_seqlen: Optional[torch.Tensor] = None,
+) -> Union[
+    Tuple[torch.Tensor], Optional[Tuple[torch.Tensor, Tuple[torch.FloatTensor, ...]]]
+]:
+    # pylint: disable=duplicate-code
+    residual = hidden_states
+
+    hidden_states = self.input_layernorm(hidden_states)
+
+    # Self Attention
+    hidden_states, self_attn_weights, present_key_value = self.self_attn(
+        hidden_states=hidden_states,
+        attention_mask=attention_mask,
+        position_ids=position_ids,
+        past_key_value=past_key_value,
+        output_attentions=output_attentions,
+        use_cache=use_cache,
+        cu_seqlens=cu_seqlens,
+        max_seqlen=max_seqlen,
+    )
+    hidden_states = residual + hidden_states
+
+    # Fully Connected
+    residual = hidden_states
+    hidden_states = self.post_attention_layernorm(hidden_states)
+    hidden_states = self.mlp(hidden_states)
+    hidden_states = residual + hidden_states
+
+    outputs = (hidden_states,)
+
+    if output_attentions:
+        outputs += (self_attn_weights,)
+
+    if use_cache:
+        outputs += (present_key_value,)
+
+    return outputs
+
+
+def stablelm_model_forward(
+    self,
+    input_ids: Optional[torch.LongTensor] = None,
+    attention_mask: Optional[torch.FloatTensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
+    inputs_embeds: Optional[torch.FloatTensor] = None,
+    use_cache: Optional[bool] = None,
+    output_attentions: Optional[bool] = None,
+    output_hidden_states: Optional[bool] = None,
+    return_dict: Optional[bool] = None,
+) -> Union[Tuple, BaseModelOutputWithPast]:
+    # pylint: disable=duplicate-code
+    output_attentions = (
+        output_attentions
+        if output_attentions is not None
+        else self.config.output_attentions
+    )
+    output_hidden_states = (
+        output_hidden_states
+        if output_hidden_states is not None
+        else self.config.output_hidden_states
+    )
+    use_cache = use_cache if use_cache is not None else self.config.use_cache
+
+    return_dict = (
+        return_dict if return_dict is not None else self.config.use_return_dict
+    )
+
+    # Retrieve input_ids and inputs_embeds
+    if input_ids is not None and inputs_embeds is not None:
+        raise ValueError(
+            "You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time"
+        )
+    if input_ids is not None:
+        batch_size, seq_length = input_ids.shape
+    elif inputs_embeds is not None:
+        batch_size, seq_length, _ = inputs_embeds.shape
+    else:
+        raise ValueError(
+            "You have to specify either decoder_input_ids or decoder_inputs_embeds"
+        )
+
+    seq_length_with_past = seq_length
+    past_key_values_length = 0
+
+    if past_key_values is not None:
+        past_key_values_length = past_key_values[0][0].shape[2]
+        seq_length_with_past = seq_length_with_past + past_key_values_length
+
+    cu_seqlens = None
+    max_seqlen = None
+    if position_ids is None:
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+        position_ids = torch.arange(
+            past_key_values_length,
+            seq_length + past_key_values_length,
+            dtype=torch.long,
+            device=device,
+        )
+        position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+    else:
+        position_ids = position_ids.view(-1, seq_length).long()
+        cu_seqlens, max_seqlen = get_cu_seqlens_from_pos_ids(position_ids)
+        cu_seqlens = cu_seqlens.squeeze()
+
+    if inputs_embeds is None:
+        inputs_embeds = self.embed_tokens(input_ids)
+    # Embed positions
+    if attention_mask is None:
+        attention_mask = torch.ones(
+            (batch_size, seq_length_with_past),
+            dtype=torch.bool,
+            device=inputs_embeds.device,
+        )
+    attention_mask = (
+        self._prepare_decoder_attention_mask(  # pylint: disable=protected-access
+            attention_mask,
+            (batch_size, seq_length),
+            inputs_embeds,
+            past_key_values_length,
+        )
+    )
+
+    hidden_states = inputs_embeds
+
+    if self.gradient_checkpointing and self.training:
+        if use_cache:
+            logger.warning(
+                "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+            )
+            use_cache = False
+
+    # Decoder layers
+    all_hidden_states = () if output_hidden_states else None
+    all_self_attns = () if output_attentions else None
+    next_decoder_cache = () if use_cache else None
+
+    for idx, decoder_layer in enumerate(self.layers):
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        past_key_value = past_key_values[idx] if past_key_values is not None else None
+
+        if self.gradient_checkpointing and self.training:
+
+            def create_custom_forward(module):
+                def custom_forward(*inputs):
+                    # None for past_key_value
+                    return module(*inputs)
+
+                return custom_forward
+
+            layer_outputs = torch.utils.checkpoint.checkpoint(
+                create_custom_forward(decoder_layer),
+                hidden_states,
+                attention_mask,
+                position_ids,
+                past_key_value,
+                output_attentions,
+                None,
+                cu_seqlens,
+                max_seqlen,
+            )
+        else:
+            layer_outputs = decoder_layer(
+                hidden_states,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_value,
+                output_attentions=output_attentions,
+                use_cache=use_cache,
+                cu_seqlens=cu_seqlens,
+                max_seqlen=max_seqlen,
+            )
+
+        hidden_states = layer_outputs[0]
+
+        if use_cache:
+            next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
+
+        if output_attentions:
+            all_self_attns += (layer_outputs[1],)
+
+    hidden_states = self.norm(hidden_states)
+
+    # Add hidden states from the last decoder layer
+    if output_hidden_states:
+        all_hidden_states += (hidden_states,)
+
+    next_cache = next_decoder_cache if use_cache else None
+    if not return_dict:
+        return tuple(
+            v
+            for v in [hidden_states, next_cache, all_hidden_states, all_self_attns]
+            if v is not None
+        )
+    return BaseModelOutputWithPast(
+        last_hidden_state=hidden_states,
+        past_key_values=next_cache,
+        hidden_states=all_hidden_states,
+        attentions=all_self_attns,
+    )

src/axolotl/monkeypatch/utils.py

@@ -101,3 +101,16 @@ def get_cu_seqlens_from_pos_ids(position_ids):
         max_seq_lens.append(max_seq_len)
 
     return torch.stack(results).to(dtype=torch.int32), torch.stack(max_seq_lens)
+
+
+def set_module_name(model, name, value):
+    if "." in name:
+        parent_name = name.rsplit(".", 1)[0]
+        child_name = name[len(parent_name) + 1 :]
+        parent = model.get_submodule(parent_name)
+    else:
+        parent_name = ""
+        parent = model
+        child_name = name
+
+    setattr(parent, child_name, value)

src/axolotl/prompt_strategies/alpaca_chat.py

@@ -1,6 +1,6 @@
-"""Module containing the AlpacaQAPromptTokenizingStrategy class"""
+"""Module for Alpaca prompt strategy classes"""
 
-from typing import Tuple
+from typing import Any, Dict, Optional, Tuple
 
 from axolotl.prompt_tokenizers import (
     AlpacaPromptTokenizingStrategy,
@@ -9,9 +9,13 @@ from axolotl.prompt_tokenizers import (
 from axolotl.prompters import AlpacaPrompter, PromptStyle, UnpromptedPrompter
 
 
-def load(tokenizer, cfg):
+def load(tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None):
+    prompt_style = PromptStyle.CHAT.value
+    if ds_cfg and "conversation" in ds_cfg:
+        prompt_style = ds_cfg["conversation"]
+
     return AlpacaPromptTokenizingStrategy(
-        AlpacaPrompter(PromptStyle.CHAT.value),
+        AlpacaPrompter(prompt_style),
         tokenizer,
         cfg.train_on_inputs,
         cfg.sequence_len,

src/axolotl/prompt_strategies/completion.py

@@ -1,10 +1,81 @@
 """
 Basic completion text
 """
-from typing import Any, Dict, Optional
+from collections import defaultdict
+from typing import Any, Dict, Generator, Optional, Tuple
 
-from axolotl.prompt_tokenizers import CompletionPromptTokenizingStrategy
-from axolotl.prompters import CompletionPrompter
+from axolotl.prompt_tokenizers import InstructionPromptTokenizingStrategy
+
+
+class CompletionPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
+    """
+    Tokenizing strategy for Completion prompts.
+    """
+
+    _field: str = "text"
+
+    def __init__(self, *args, max_length=None, **kwargs):
+        super().__init__(*args, **kwargs)
+        if max_length is not None:
+            self.max_length = max_length
+
+    @property
+    def supports_batched(self):
+        return True
+
+    @property
+    def field(self) -> str:
+        return self._field
+
+    @field.setter
+    def field(self, new_field: str):
+        self._field = new_field
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
+        return (
+            prompt[self.field],
+            "",
+            "",
+        )
+
+    def tokenize_prompt(self, prompt):
+        res = defaultdict(lambda: [])
+        feature_names = list(prompt.keys())
+        for row in zip(*prompt.values()):
+            prompt_row = dict(zip(feature_names, row))
+            (
+                instruction,
+                _,
+                _,
+            ) = self.parse_instruction_fields(prompt_row)
+
+            full_prompt = self._build_full_prompt(instruction, None, None)
+            tokenized_full_prompt = self._tokenize(full_prompt)
+
+            for key, val in tokenized_full_prompt.items():
+                for i in range(0, len(val), self.sequence_len):
+                    res[key].append(val[i : i + self.sequence_len])
+
+        return dict(res)
+
+    def _build_full_prompt(
+        self, instruction, input, response
+    ):  # pylint: disable=redefined-builtin
+        return next(iter(self.prompter.build_prompt(instruction, input, response)))
+
+
+class CompletionPrompter:
+    """
+    Prompter for completion
+    """
+
+    def build_prompt(
+        self,
+        instruction: str,
+        input=None,  # pylint: disable=redefined-builtin, unused-argument
+        output=None,  # pylint: disable=unused-argument
+    ) -> Generator[str, None, None]:
+        yield instruction
 
 
 def load(tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None):
@@ -13,6 +84,7 @@ def load(tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None):
         tokenizer,
         cfg.train_on_inputs,
         cfg.sequence_len,
+        max_length=cfg.sequence_len * 64,
     )
     if ds_cfg and "field" in ds_cfg:
         strat.field = ds_cfg["field"]

src/axolotl/prompt_strategies/context_qa.py

@@ -24,6 +24,15 @@ def load(tokenizer, cfg):
     )
 
 
+def load_v2(tokenizer, cfg):
+    return ContextQaV2PromptTokenizingStrategy(
+        ContextV2Prompter(),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
 class AlpacaContextPrompter(AlpacaPrompter):
     """
     Customized system prompted for concise QA
@@ -50,6 +59,38 @@ class AlpacaContextPromptTokenizingStrategy(InstructionPromptTokenizingStrategy)
         )
 
 
+class ContextQaV2PromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
+    """
+    Tokenization Strategy to combine in-context article with a question and answer
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
+        return (
+            "Context: "
+            + prompt["context"]
+            + "\nQuestion: "
+            + prompt["question"]
+            + "\n",
+            "",
+            "Answer: " + prompt["answer"],
+        )
+
+
+class ContextV2Prompter(AlpacaPrompter):
+    """
+    Customized system prompted for concise QA
+    """
+
+    system_prompt = ""
+    system_no_input_prompt = ""
+
+    def match_prompt_style(self):
+        # pylint: disable=duplicate-code
+        self.turn_format = "{instruction}\n{input}"
+        self.turn_no_input_format = "{instruction}"
+        self.system_format = "{system}"
+
+
 class AlpacaMissingInfoContextPromptTokenizingStrategy(
     InstructionPromptTokenizingStrategy
 ):

src/axolotl/prompt_strategies/sharegpt.py

@@ -1,21 +1,47 @@
 """Module containing the SimpleShareGPTPromptTokenizingStrategy class"""
+from typing import Any, Dict, Optional
+
+from fastchat.conversation import Conversation, SeparatorStyle, register_conv_template
 
 from axolotl.prompt_tokenizers import ShareGPTPromptTokenizingStrategy
-from axolotl.prompters import PromptStyle, ShareGPTPrompter
+from axolotl.prompters import ShareGPTPrompterV2
+
+register_conv_template(
+    Conversation(
+        name="chatml",
+        system_template="<|im_start|>system\n{system_message}",
+        system_message="You are a helpful assistant.",
+        roles=["<|im_start|>user", "<|im_start|>assistant"],
+        sep_style=SeparatorStyle.CHATML,
+        sep="<|im_end|>\n",
+    )
+)
 
 
-def load(tokenizer, cfg):
-    return SimpleShareGPTPromptTokenizingStrategy(
-        ShareGPTPrompter(PromptStyle.CHAT.value),
+def load(tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None):
+    conversation = (
+        ds_cfg["conversation"] if ds_cfg and "conversation" in ds_cfg else None
+    )
+    field_human = ds_cfg["field_human"] if ds_cfg and "field_human" in ds_cfg else None
+    field_model = ds_cfg["field_model"] if ds_cfg and "field_model" in ds_cfg else None
+    strategy = SimpleShareGPTPromptTokenizingStrategy(
+        ShareGPTPrompterV2(
+            conversation=conversation,
+            role_key_model=field_model,
+            role_key_human=field_human,
+        ),
         tokenizer,
         cfg.train_on_inputs,
         cfg.sequence_len,
     )
+    if ds_cfg and "strict" in ds_cfg:
+        strategy.strict = ds_cfg["strict"]
+    return strategy
 
 
 def load_role(tokenizer, cfg):
     return SimpleRoleShareGPTPromptTokenizingStrategy(
-        ShareGPTPrompter(PromptStyle.CHAT.value),
+        ShareGPTPrompterV2(),
         tokenizer,
         cfg.train_on_inputs,
         cfg.sequence_len,
@@ -24,7 +50,7 @@ def load_role(tokenizer, cfg):
 
 def load_guanaco(tokenizer, cfg):
     return GuanacoShareGPTPromptTokenizingStrategy(
-        ShareGPTPrompter(PromptStyle.CHAT.value),
+        ShareGPTPrompterV2(),
         tokenizer,
         cfg.train_on_inputs,
         cfg.sequence_len,
@@ -36,8 +62,26 @@ class SimpleShareGPTPromptTokenizingStrategy(ShareGPTPromptTokenizingStrategy):
     basic sharegpt strategy to grab conversations from the sample row
     """
 
+    _strict = True
+
+    @property
+    def strict(self):
+        return self._strict
+
+    @strict.setter
+    def strict(self, strict):
+        self._strict = strict
+
     def get_conversation_thread(self, prompt):
-        return prompt["conversations"]
+        conversations = prompt["conversations"]
+        if self.strict:
+            return conversations
+        # remap roles - allow for assistant turn
+        role_map = {"human": "human", "assistant": "gpt", "gpt": "gpt"}
+        turns = [
+            {"from": role_map[t["from"]], "value": t["value"]} for t in conversations
+        ]
+        return turns
 
 
 class SimpleRoleShareGPTPromptTokenizingStrategy(ShareGPTPromptTokenizingStrategy):

src/axolotl/prompt_strategies/sharegpt_jokes.py

@@ -1,11 +1,11 @@
 """Module for Jokes prompts using sharegpt style """
 from axolotl.prompt_tokenizers import ShareGPTPromptTokenizingStrategy
-from axolotl.prompters import PromptStyle, ShareGPTPrompter
+from axolotl.prompters import ShareGPTPrompterV2
 
 
 def load(tokenizer, cfg):
     return SimpleJokesShareGPTPromptTokenizingStrategy(
-        ShareGPTPrompter(PromptStyle.CHAT.value),
+        ShareGPTPrompterV2(),
         tokenizer,
         cfg.train_on_inputs,
         cfg.sequence_len,

src/axolotl/prompt_tokenizers.py

@@ -2,12 +2,15 @@
 
 import abc
 import copy
-import functools
 import logging
 from typing import Dict, List, Tuple, Union
 
+from fastchat.conversation import Conversation
 from transformers import BatchEncoding, PreTrainedTokenizer
 
+from axolotl.monkeypatch.fastchat_conversation_turns import (
+    add_get_turns_to_conversation,
+)
 from axolotl.prompters import IGNORE_TOKEN_ID
 
 LOG = logging.getLogger("axolotl")
@@ -18,6 +21,8 @@ LLAMA_DEFAULT_EOS_TOKEN = "</s>"  # nosec
 LLAMA_DEFAULT_BOS_TOKEN = "<s>"  # nosec
 LLAMA_DEFAULT_UNK_TOKEN = "<unk>"  # nosec
 
+add_get_turns_to_conversation()
+
 
 class InvalidDataException(Exception):
     """
@@ -40,63 +45,47 @@ class PromptTokenizingStrategy(abc.ABC):
         self.prompter = prompter
         self.tokenizer: PreTrainedTokenizer = tokenizer
         self.train_on_inputs = train_on_inputs
+        # sequence_len and max_length can be different for CompletionPromptTokenizingStrategy.
+        # TODO: Document how they are different.
         self.sequence_len = sequence_len
+        self.max_length = sequence_len
 
     @abc.abstractmethod
     def tokenize_prompt(self, prompt):
         pass
 
-    @functools.lru_cache(maxsize=128)
-    def _get_user_token(self):
-        try:
-            id_or_ids = self.tokenizer.convert_tokens_to_ids("<|USER|>")
-            if isinstance(id_or_ids, (int,)):
-                return id_or_ids
-        except KeyError:
-            pass
-        return False
-
-    @functools.lru_cache(maxsize=128)
-    def _get_assistant_token(self):
-        try:
-            id_or_ids = self.tokenizer.convert_tokens_to_ids("<|ASSISTANT|>")
-            if isinstance(id_or_ids, (int,)):
-                return id_or_ids
-        except KeyError:
-            pass
+    @property
+    def supports_batched(self):
         return False
 
     def _tokenize(
         self, prompt: str, add_eos_token: bool = True, strip_bos_token: bool = False
     ) -> BatchEncoding:
-        result: BatchEncoding
-        if not prompt.strip():
+        empty = BatchEncoding(data={"input_ids": [], "attention_mask": []})
+        if not prompt:
             LOG.warning("Empty text requested for tokenization.")
-            result = BatchEncoding(data={"input_ids": [], "attention_mask": []})
-        else:
-            result = self.tokenizer(
-                prompt,
-                truncation=True,
-                max_length=self.sequence_len,
-                padding=False,
-                return_tensors=None,
-            )
+            return empty
+
+        result = self.tokenizer(
+            prompt,
+            truncation=True,
+            max_length=self.max_length,
+            padding=False,
+            return_tensors=None,
+        )
         if len(result["input_ids"]) == 0:
             LOG.warning("Tokenizer result is empty. You may want to audit your dataset")
+            return empty
+
         if (
-            len(result["input_ids"]) > 0
-            and result["input_ids"][-1] != self.tokenizer.eos_token_id
-            and len(result["input_ids"]) < self.sequence_len
+            result["input_ids"][-1] != self.tokenizer.eos_token_id
+            and len(result["input_ids"]) < self.max_length
             and add_eos_token
         ):
             result["input_ids"].append(self.tokenizer.eos_token_id)
             result["attention_mask"].append(1)
 
-        if (
-            len(result["input_ids"]) > 0
-            and result["input_ids"][0] == self.tokenizer.bos_token_id
-            and strip_bos_token
-        ):
+        if result["input_ids"][0] == self.tokenizer.bos_token_id and strip_bos_token:
             result["input_ids"] = result["input_ids"][1:]
             result["attention_mask"] = result["attention_mask"][1:]
 
@@ -132,7 +121,7 @@ class InstructionPromptTokenizingStrategy(PromptTokenizingStrategy):
         if not self.train_on_inputs:
             user_prompt_len = len(tokenized_prompt["input_ids"])
             # TODO this could be sped up using numpy array slicing
-            tokenized_prompt["labels"] = [-100] * user_prompt_len
+            tokenized_prompt["labels"] = [IGNORE_INDEX] * user_prompt_len
         tokenized_res_prompt = self._tokenize(
             response, strip_bos_token=True, add_eos_token=True
         )
@@ -247,46 +236,6 @@ class NomicGPT4AllPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
         )
 
 
-class CompletionPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
-    """
-    Tokenizing strategy for Completion prompts.
-    """
-
-    _field: str = "text"
-
-    @property
-    def field(self) -> str:
-        return self._field
-
-    @field.setter
-    def field(self, new_field: str):
-        self._field = new_field
-
-    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
-        return (
-            prompt[self.field],
-            "",
-            "",
-        )
-
-    def tokenize_prompt(self, prompt):
-        (
-            instruction,
-            _,
-            _,
-        ) = self.parse_instruction_fields(prompt)
-
-        full_prompt = self._build_full_prompt(instruction, None, None)
-        tokenized_full_prompt = self._tokenize(full_prompt)
-
-        return tokenized_full_prompt
-
-    def _build_full_prompt(
-        self, instruction, input, response
-    ):  # pylint: disable=redefined-builtin
-        return next(iter(self.prompter.build_prompt(instruction, input, response)))
-
-
 class ReflectionPromptTokenizingStrategy(PromptTokenizingStrategy):
     """
     Tokenizing strategy for Reflection prompts.
@@ -296,6 +245,7 @@ class ReflectionPromptTokenizingStrategy(PromptTokenizingStrategy):
         raise NotImplementedError
 
     def tokenize_prompt(self, prompt):
+        # pylint: disable=duplicate-code
         (
             instruction,
             input,  # pylint: disable=redefined-builtin
@@ -320,7 +270,7 @@ class ReflectionPromptTokenizingStrategy(PromptTokenizingStrategy):
             user_prompt_len = len(tokenized_user_prompt["input_ids"])
             # TODO this could be sped up using numpy array slicing
             tokenized_full_prompt["labels"] = [
-                -100
+                IGNORE_INDEX
             ] * user_prompt_len + tokenized_full_prompt["labels"][user_prompt_len:]
 
         return tokenized_full_prompt
@@ -384,52 +334,89 @@ class ShareGPTPromptTokenizingStrategy(PromptTokenizingStrategy):
         return prompt["conversations"]
 
     def tokenize_prompt(self, prompt):
+        # Initial values. We will append to these as we go through the conversation.
         result, current_len = tokenize_prompt_default()
-        user_token = self._get_user_token()
-        assistant_token = self._get_assistant_token()
+        conversation: Conversation = (
+            self.prompter._conversation.copy()  # pylint: disable=protected-access
+        )
+
+        # support for custom roles from the dataset, only useful for vicuna style prompts/roles
+        role_remap = []
+        if (
+            conversation.name == "vicuna_v1.1"
+            and "roles" in prompt
+            and len(prompt["roles"]) >= 2
+        ):
+            role_remap = [
+                {"from": conversation.roles[0], "to": prompt["roles"][0]},
+                {"from": conversation.roles[1], "to": prompt["roles"][1]},
+            ]
+
         try:
             for _, part in enumerate(
                 self.prompter.build_prompt(self.get_conversation_thread(prompt))
             ):
-                if isinstance(part, tuple):
-                    if part[0] == "USER:":
-                        part = part[0] + part[1] if not user_token else part[1]
-                        # this is still the user query, we should
-                        res = self._tokenize(
-                            part.strip(),
-                            add_eos_token=False,
-                            strip_bos_token=True,
-                        )
-                        if user_token:
-                            res["input_ids"] = [user_token, *res["input_ids"]]
-                        # everything from this is masked out from the labels
-                        labels = [IGNORE_TOKEN_ID] * len(res["input_ids"])
-                    elif part[0] == "ASSISTANT:":
-                        # TODO label assistant token/tokens w/ IGNORE_TOKEN_ID
-                        part = part[0] + part[1] if not assistant_token else part[1]
-                        # this should be the assistent response, should end with an eos token
-                        res = self._tokenize(
-                            part.strip(),
-                            add_eos_token=True,
-                            strip_bos_token=True,
-                        )
-                        if assistant_token:
-                            res["input_ids"] = [
-                                assistant_token,
-                                *res["input_ids"],
-                            ]
-                        # not masked out from labels
-                        labels = copy.deepcopy(res["input_ids"])
-                    elif part[0] == "SYSTEM:":
-                        part = part[1]  # Ignore the system role from preamble
-                        # this is only ever the first part, should include the bos token and the user query
-                        res = self._tokenize(
-                            part.strip(), add_eos_token=False, strip_bos_token=False
-                        )
-                        # everything from this is masked out from the labels
-                        labels = [IGNORE_TOKEN_ID] * len(res["input_ids"])
-                    else:
-                        LOG.warning(f"unhandled role: {part[0]}")
+                if not isinstance(part, tuple):
+                    LOG.warning(f"expected tuple, got {part}")
+                    continue
+
+                user, assistant = conversation.roles
+                role, content = part
+
+                # Uses "in" because role contains extra characters
+                if user in role:
+                    role = (
+                        role.replace(role_remap[0]["from"], role_remap[0]["to"])
+                        if role_remap
+                        else role
+                    )
+                    turn = role + content
+                    # this is still the user query, we should
+                    if not content.strip():
+                        LOG.warning(f"user turn has empty text: {prompt}")
+                    res = self._tokenize(
+                        turn,
+                        add_eos_token=False,
+                        strip_bos_token=True,
+                    )
+                    # everything from this is masked out from the labels
+                    labels = [IGNORE_TOKEN_ID] * len(res["input_ids"])
+                elif assistant in role:
+                    # TODO label assistant token/tokens w/ IGNORE_TOKEN_ID
+                    role = (
+                        role.replace(role_remap[1]["from"], role_remap[1]["to"])
+                        if role_remap
+                        else role
+                    )
+                    turn = role + content
+                    # this should be the assistant response, should end with an eos token
+                    if not content.strip():
+                        LOG.warning(f"assistant turn has empty text: {prompt}")
+                    res = self._tokenize(
+                        turn,
+                        add_eos_token=True,
+                        strip_bos_token=True,
+                    )
+                    role_res = self._tokenize(
+                        role.rstrip(),
+                        add_eos_token=False,
+                        strip_bos_token=True,
+                    )
+                    # not masked out from labels
+                    labels = copy.deepcopy(res["input_ids"])
+                    len_role = len(role_res["input_ids"])
+                    labels[:len_role] = [IGNORE_TOKEN_ID] * min(len_role, len(labels))
+                elif role == "":
+                    turn = content
+                    # this is only ever the first part, should include the bos token and the user query
+                    res = self._tokenize(
+                        turn, add_eos_token=False, strip_bos_token=False
+                    )
+                    # everything from this is masked out from the labels
+                    labels = [IGNORE_TOKEN_ID] * len(res["input_ids"])
+                else:
+                    LOG.warning(f"unhandled role: {role}")
+                    continue
 
                 # pylint: disable=duplicate-code
                 result, current_len = parse_tokenized_to_result(
@@ -443,29 +430,6 @@ class ShareGPTPromptTokenizingStrategy(PromptTokenizingStrategy):
         except (KeyError, AssertionError, IndexError) as err:
             raise InvalidDataException(str(err)) from err
 
-    def _tokenize(self, prompt, add_eos_token=True, strip_bos_token=False):
-        result = self.tokenizer(
-            prompt,
-            truncation=True,
-            max_length=self.sequence_len,
-            padding=False,
-            return_tensors=None,
-        )
-        if (
-            result["input_ids"][-1] != self.tokenizer.eos_token_id
-            and len(result["input_ids"]) < self.sequence_len
-            and add_eos_token
-        ):
-            result["input_ids"].append(self.tokenizer.eos_token_id)
-            result["attention_mask"].append(1)
-
-        if result["input_ids"][0] == self.tokenizer.bos_token_id and strip_bos_token:
-            result["input_ids"] = result["input_ids"][1:]
-            result["attention_mask"] = result["attention_mask"][1:]
-
-        result["labels"] = result["input_ids"].copy()
-        return result
-
 
 def tokenize_prompt_default() -> Tuple[Dict[str, List[int]], int]:
     """

src/axolotl/prompters.py

@@ -1,12 +1,15 @@
 """Module containing prompters"""
 
-import dataclasses
 import logging
-from enum import Enum, auto
-from typing import Generator, List, Optional, Tuple, Union
+from enum import Enum
+from typing import Generator, Optional, Union
+
+from colorama import Fore
+from fastchat.conversation import Conversation, get_conv_template
 
 LOG = logging.getLogger("axolotl")
 IGNORE_TOKEN_ID = -100
+REPR_TEMPLATE = "\n<start>\n" + Fore.CYAN + "{full_prompt}" + Fore.RESET + "\n<end>\n"
 
 
 class PromptStyle(Enum):
@@ -54,20 +57,15 @@ class AlpacaPrompter:
             )
             self.system_format = "<|im_start|>system\n{system}<|im_end|>\n"
 
-    def build_prompt(
-        self,
-        instruction: str,
-        input: Union[None, str] = None,  # pylint: disable=redefined-builtin
-        output: Union[None, str] = None,
-    ) -> Generator[str, None, None]:
+    def _build_result(self, instruction, input_text, output):
         # returns the full prompt from instruction and optional input
         # if a label (=response, =output) is provided, it's also appended.
-        if input:
+        if input_text:
             res = (
                 self.system_format.format(system=self.system_prompt)
                 if self.system_prompt
                 else ""
-            ) + self.turn_format.format(instruction=instruction, input=input)
+            ) + self.turn_format.format(instruction=instruction, input=input_text)
         else:
             res = (
                 self.system_format.format(system=self.system_no_input_prompt)
@@ -76,7 +74,21 @@ class AlpacaPrompter:
             ) + self.turn_no_input_format.format(instruction=instruction)
         if output:
             res = f"{res}{output}"
-        yield res
+
+        return res
+
+    def build_prompt(
+        self,
+        instruction: str,
+        input: Union[None, str] = None,  # pylint: disable=redefined-builtin
+        output: Union[None, str] = None,
+    ) -> Generator[str, None, None]:
+        yield self._build_result(instruction, input, output)
+
+    def __repr__(self) -> str:
+        return REPR_TEMPLATE.format(
+            full_prompt=self._build_result("{instruction}", "{input}", "{output}")
+        )
 
 
 class UnpromptedPrompter(AlpacaPrompter):
@@ -135,20 +147,6 @@ class SummarizeTLDRPrompter(AlpacaPrompter):
         self.turn_no_input_format = "USER: Summarize the following article as a TL;DR.\n{instruction}\nASSISTANT:"
 
 
-class CompletionPrompter:
-    """
-    Prompter for completion
-    """
-
-    def build_prompt(
-        self,
-        instruction: str,
-        input=None,  # pylint: disable=redefined-builtin, unused-argument
-        output=None,  # pylint: disable=unused-argument
-    ) -> Generator[str, None, None]:
-        yield instruction
-
-
 class GPTeacherPrompter(AlpacaPrompter):
     """
     Prompter for GPTeacher
@@ -204,14 +202,14 @@ class ReflectAlpacaPrompter:
             )
             self.response_split = "ASSISTANT:"
 
-    def build_prompt(
+    def _build_result(
         self,
         instruction: str,
         input: Union[None, str] = None,  # pylint: disable=redefined-builtin
         output: Union[None, str] = None,
         reflection: Union[None, str] = None,
         corrected: Union[None, str] = None,
-    ) -> Generator[str, None, None]:
+    ):
         # returns the full prompt from instruction and optional input
         # if a label (=response, =output) is provided, it's also appended.
         if input:
@@ -225,54 +223,30 @@ class ReflectAlpacaPrompter:
                 corrected=corrected,
             )
             res = f"{res}{label}"
-        yield res
 
+        return res
 
-class SeparatorStyle(Enum):
-    """Different separator style."""
-
-    SINGLE = auto()
-    TWO = auto()
-    DOLLY = auto()
-
-
-# TODO clean this 💩 up
-@dataclasses.dataclass
-class Conversation:
-    """A class that keeps all conversation history."""
-
-    system: str
-    roles: List[str]
-    messages: List[List[str]]
-    offset: int
-    sep_style: SeparatorStyle = SeparatorStyle.SINGLE
-    sep: str = "###"
-    sep2: Optional[str] = None
-
-    def get_prompt(self) -> Generator[Tuple[str, str], None, None]:
-        # seps = [self.sep, self.sep2]
-        preamble = self.system + self.sep
-        yield ("SYSTEM:", preamble)
-        for _, (role, message) in enumerate(self.messages):
-            if message:
-                yield (role + ":", " " + message)
-            else:
-                LOG.warning(f"role with empty message: {role}")
-                yield (role + ":", "")
-
-    def copy(self):
-        return Conversation(
-            system=self.system,
-            roles=self.roles,
-            messages=[[x, y] for x, y in self.messages],
-            offset=self.offset,
-            sep_style=self.sep_style,
-            sep=self.sep,
-            sep2=self.sep2,
+    def build_prompt(
+        self,
+        instruction: str,
+        input: Union[None, str] = None,  # pylint: disable=redefined-builtin
+        output: Union[None, str] = None,
+        reflection: Union[None, str] = None,
+        corrected: Union[None, str] = None,
+    ) -> Generator[str, None, None]:
+        # pylint: disable=duplicate-code
+        yield self._build_result(
+            instruction,
+            input,
+            output,
+            reflection,
+            corrected,
         )
 
-    def append_message(self, role, message):
-        self.messages.append([role, message])
+    def __repr__(self) -> str:
+        return REPR_TEMPLATE.format(
+            full_prompt=self._build_result("{instruction}", "{input}", "{output}")
+        )
 
 
 SHAREGPT_ASSERTION_FAILED_ROLE = (
@@ -285,30 +259,29 @@ class ShareGPTPrompter:  # pylint: disable=too-few-public-methods
     A prompter that generates prompts for the ShareGPT
     """
 
-    def __init__(self, prompt_style=None, system_prompt: Optional[str] = None):
-        if prompt_style != PromptStyle.CHAT.value:
-            raise ValueError(
-                f"unsupported prompt_style for ShareGPTPrompter({prompt_style})"
-            )
-        system: str = (
-            system_prompt
-            if system_prompt
-            else (
-                "A chat between a curious user and an artificial intelligence assistant. "
-                "The assistant gives helpful, detailed, and polite answers to the user's questions."
-            )
-        )
-        self._conversation = Conversation(
-            system=system,
-            roles=["USER", "ASSISTANT"],
-            messages=[],
-            offset=0,
-            sep_style=SeparatorStyle.TWO,
-            sep=" ",
-            sep2=" ",
-        )
+    role_key_human = "human"
+    role_key_model = "gpt"
 
-    def build_prompt(self, source) -> Generator[str, None, None]:
+    def __init__(
+        self,
+        prompt_style=None,  # pylint: disable=unused-argument
+        conversation: Optional[Union[str, Conversation]] = None,
+        role_key_human: Optional[str] = None,
+        role_key_model: Optional[str] = None,
+    ):
+        if conversation:
+            if isinstance(conversation, Conversation):
+                self._conversation = conversation
+            else:
+                self._conversation = get_conv_template(conversation)
+        else:
+            self._conversation = get_conv_template("vicuna_v1.1")
+        if role_key_human:
+            self.role_key_human = role_key_human
+        if role_key_model:
+            self.role_key_model = role_key_model
+
+    def _build_result(self, source):
         if len(source) < 2:
             # If there isn't a back and forth conversation, ignore it
             # also happens on the data splitting leaving empty conversations
@@ -320,17 +293,14 @@ class ShareGPTPrompter:  # pylint: disable=too-few-public-methods
 
         # Add the conversation system prompt if provided, otherwise use the default one
         if source[0]["from"] == "system":
-            conv.system = source[0]["value"]
+            conv.set_system_message(source[0]["value"])
             source.pop(0)
 
-        roles = {"human": conv.roles[0], "gpt": conv.roles[1]}
+        roles = {self.role_key_human: conv.roles[0], self.role_key_model: conv.roles[1]}
 
         try:
             # Apply prompt templates
-            if (
-                source[0]["from"] not in roles
-                or roles[source[0]["from"]] != conv.roles[0]
-            ):
+            if source[0]["from"] not in roles:
                 # Skip the first one if it is not from human
                 source = source[1:]
         except IndexError as err:
@@ -338,10 +308,54 @@ class ShareGPTPrompter:  # pylint: disable=too-few-public-methods
             raise err
 
         conv.messages = []
-        for j, sentence in enumerate(source):
+        for _, sentence in enumerate(source):
             role = roles[sentence["from"]]
-            assert role == conv.roles[j % 2], SHAREGPT_ASSERTION_FAILED_ROLE
+            if len(conv.messages) > 0 and (
+                (role == conv.messages[-1][0]) or (role not in conv.roles)
+            ):
+                LOG.warning(f"{SHAREGPT_ASSERTION_FAILED_ROLE}: {sentence}")
             conv.append_message(role, sentence["value"])
 
-        for part in conv.get_prompt():
+        return conv.get_turns()
+
+    def build_prompt(self, source) -> Generator[str, None, None]:
+        turns = self._build_result(source)
+
+        for part in turns:
+            if part[0] and not part[1]:
+                LOG.warning(f"role with empty message: {part[0]}")
             yield part
+
+    def __repr__(self) -> str:
+        turns = self._build_result([{"from": "{from}", "value": "{value}"}])
+        return "\n".join([REPR_TEMPLATE.format(full_prompt=part) for part in turns])
+
+
+class ShareGPTPrompterV2(ShareGPTPrompter):
+    """
+    A V2 prompter that generates prompts for the ShareGPT
+    """
+
+    def __init__(
+        self,
+        conversation: Optional[Union[str, Conversation]] = None,
+        role_key_human: Optional[str] = None,
+        role_key_model: Optional[str] = None,
+    ):
+        super().__init__(
+            conversation=conversation,
+            role_key_human=role_key_human,
+            role_key_model=role_key_model,
+        )
+
+
+class UnsupportedPrompter:
+    """
+    A dummy class for custom prompters
+    """
+
+    def __init__(self) -> None:
+        pass
+
+    def __repr__(self):
+        return "Pre-tokenized or custom dataset types are unsupported for logging"

src/axolotl/train.py

@@ -1,6 +1,5 @@
 """Prepare and train a model on a dataset. Can also infer from a model or merge lora"""
 
-import logging
 import os
 import signal
 import sys
@@ -9,13 +8,15 @@ from pathlib import Path
 from typing import Optional
 
 import torch
-
-# add src to the pythonpath so we don't need to pip install this
+import transformers.modelcard
+from accelerate.logging import get_logger
 from datasets import Dataset
 from optimum.bettertransformer import BetterTransformer
+from transformers.deepspeed import is_deepspeed_zero3_enabled
 
 from axolotl.common.cli import TrainerCliArgs
 from axolotl.logging_config import configure_logging
+from axolotl.monkeypatch import neft_embeddings
 from axolotl.utils.dict import DictDefault
 from axolotl.utils.models import load_model, load_tokenizer
 from axolotl.utils.trainer import setup_trainer
@@ -25,7 +26,7 @@ src_dir = os.path.join(project_root, "src")
 sys.path.insert(0, src_dir)
 
 configure_logging()
-LOG = logging.getLogger("axolotl.train")
+LOG = get_logger("axolotl.train")
 
 
 @dataclass
@@ -40,13 +41,13 @@ class TrainDatasetMeta:
 
 
 def train(
-    *,
-    cfg: DictDefault,
-    cli_args: TrainerCliArgs,
-    dataset_meta: TrainDatasetMeta,
+    *, cfg: DictDefault, cli_args: TrainerCliArgs, dataset_meta: TrainDatasetMeta
 ):
     # load the tokenizer first
-    LOG.info(f"loading tokenizer... {cfg.tokenizer_config or cfg.base_model_config}")
+    LOG.debug(
+        f"loading tokenizer... {cfg.tokenizer_config or cfg.base_model_config}",
+        main_process_only=True,
+    )
     tokenizer = load_tokenizer(cfg)
 
     train_dataset = dataset_meta.train_dataset
@@ -54,7 +55,10 @@ def train(
     total_num_steps = dataset_meta.total_num_steps
 
     # Load the model and tokenizer
-    LOG.info("loading model and (optionally) peft_config...")
+    msg = "loading model"
+    if cfg.adapter:
+        msg += " and peft_config..."
+    LOG.debug(msg)
     model, peft_config = load_model(cfg, tokenizer, inference=cli_args.inference)
 
     safe_serialization = cfg.save_safetensors is True
@@ -103,10 +107,14 @@ def train(
             signal.SIGINT, lambda signum, frame: terminate_handler(signum, frame, model)
         )
 
+    badge_markdown = """[<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)"""
+    transformers.modelcard.AUTOGENERATED_TRAINER_COMMENT += f"\n{badge_markdown}"
+
     LOG.info("Starting trainer...")
     if cfg.group_by_length:
         LOG.info("hang tight... sorting dataset for group_by_length")
 
+    pretrain_hooks(cfg, trainer)
     if cfg.flash_optimum:
         with torch.backends.cuda.sdp_kernel(
             enable_flash=True, enable_math=True, enable_mem_efficient=True
@@ -114,9 +122,15 @@ def train(
             trainer.train(resume_from_checkpoint=resume_from_checkpoint)
     else:
         trainer.train(resume_from_checkpoint=resume_from_checkpoint)
+    post_train_hooks(cfg, trainer)
 
     LOG.info(f"Training Completed!!! Saving pre-trained model to {cfg.output_dir}")
 
+    # post training
+    for name, module in model.named_modules():
+        if hasattr(module, "_post_training"):
+            module._post_training(model, name)  # pylint: disable=protected-access
+
     if trainer.is_fsdp_enabled:
         trainer.accelerator.state.fsdp_plugin.set_state_dict_type("FULL_STATE_DICT")
         LOG.info("Set FSDP state dict type to FULL_STATE_DICT for saving.")
@@ -132,10 +146,49 @@ def train(
     # only save on rank 0, otherwise it corrupts output on multi-GPU when multiple processes attempt to write the same file
     if cfg.fsdp:
         trainer.save_model(cfg.output_dir)
+    elif cfg.deepspeed and is_deepspeed_zero3_enabled():
+        # Copied over from: https://github.com/huggingface/accelerate/blob/5ae611118057232f441055f7ef9ba0b0f2b8d533/docs/source/usage_guides/deepspeed.md#saving-and-loading
+        trainer.accelerator.wait_for_everyone()
+        unwrapped_model = trainer.accelerator.unwrap_model(trainer.model_wrapped)
+
+        # Saves the whole/unpartitioned fp16 model when in ZeRO Stage-3 to the output directory if
+        # `stage3_gather_16bit_weights_on_model_save` is True in DeepSpeed Config file or
+        # `zero3_save_16bit_model` is True in DeepSpeed Plugin.
+        # For Zero Stages 1 and 2, models are saved as usual in the output directory.
+        # The model name saved is `pytorch_model.bin`
+        unwrapped_model.save_pretrained(
+            cfg.output_dir,
+            is_main_process=trainer.accelerator.is_main_process,
+            save_function=trainer.accelerator.save,
+            state_dict=trainer.accelerator.get_state_dict(trainer.model_wrapped),
+        )
     elif cfg.local_rank == 0:
         if cfg.flash_optimum:
             model = BetterTransformer.reverse(model)
 
         model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)
 
+    if not cfg.hub_model_id:
+        trainer.create_model_card(model_name=cfg.output_dir.lstrip("./"))
+
     return model, tokenizer
+
+
+def pretrain_hooks(cfg, trainer):
+    """
+    Run hooks right before kicking off the training
+    :param cfg:
+    :param trainer:
+    :return:
+    """
+    neft_embeddings.pretrain_hook(cfg, trainer)
+
+
+def post_train_hooks(cfg, trainer):
+    """
+    Run hooks right after training completes
+    :param cfg:
+    :param trainer:
+    :return:
+    """
+    neft_embeddings.post_train_hook(cfg, trainer)

src/axolotl/utils/bench.py

@@ -1,14 +1,44 @@
 """Benchmarking and measurement utilities"""
+import functools
 
 import pynvml
 import torch
 from pynvml.nvml import NVMLError
 
 
+def check_cuda_device(default_value):
+    """
+    wraps a function and returns the default value instead of running the
+    wrapped function if cuda isn't available or the device is auto
+    :param default_value:
+    :return:
+    """
+
+    def deco(func):
+        @functools.wraps(func)
+        def wrapper(*args, **kwargs):
+            device = kwargs.get("device", args[0] if args else None)
+
+            if (
+                not torch.cuda.is_available()
+                or device == "auto"
+                or torch.device(device).type == "cpu"
+            ):
+                return default_value
+
+            return func(*args, **kwargs)
+
+        return wrapper
+
+    return deco
+
+
+@check_cuda_device(0.0)
 def gpu_memory_usage(device=0):
     return torch.cuda.memory_allocated(device) / 1024.0**3
 
 
+@check_cuda_device((0.0, 0.0, 0.0))
 def gpu_memory_usage_all(device=0):
     usage = torch.cuda.memory_allocated(device) / 1024.0**3
     reserved = torch.cuda.memory_reserved(device) / 1024.0**3
@@ -16,6 +46,7 @@ def gpu_memory_usage_all(device=0):
     return usage, reserved - usage, max(0, smi - reserved)
 
 
+@check_cuda_device(0.0)
 def gpu_memory_usage_smi(device=0):
     if isinstance(device, torch.device):
         device = device.index
@@ -31,9 +62,6 @@ def gpu_memory_usage_smi(device=0):
 
 
 def log_gpu_memory_usage(log, msg, device):
-    if not torch.cuda.is_available() or device == "auto":
-        return (0, 0, 0)
-
     usage, cache, misc = gpu_memory_usage_all(device)
     extras = []
     if cache > 0:

src/axolotl/utils/callbacks.py

@@ -37,32 +37,32 @@ from axolotl.utils.distributed import (
 )
 
 if TYPE_CHECKING:
-    from axolotl.utils.trainer import AxolotlTrainingArguments
+    from axolotl.core.trainer_builder import AxolotlTrainingArguments
 
 LOG = logging.getLogger("axolotl.callbacks")
 IGNORE_INDEX = -100
 
 
-class SavePeftModelCallback(TrainerCallback):  # pylint: disable=too-few-public-methods
-    """Callback to save the PEFT adapter"""
+class EvalFirstStepCallback(
+    TrainerCallback
+):  # pylint: disable=too-few-public-methods disable=unused-argument
+    """
+    Callback to trigger evals on the first step
+    """
 
-    def on_save(
+    def on_step_end(
         self,
         args: TrainingArguments,
         state: TrainerState,
         control: TrainerControl,
         **kwargs,
     ):
-        checkpoint_folder = os.path.join(
-            args.output_dir,
-            f"{PREFIX_CHECKPOINT_DIR}-{state.global_step}",
-        )
-
-        peft_model_path = os.path.join(checkpoint_folder, "adapter_model")
-        kwargs["model"].save_pretrained(
-            peft_model_path, save_safetensors=args.save_safetensors
-        )
-
+        if (
+            args.evaluation_strategy == IntervalStrategy.STEPS
+            and args.eval_steps < 1.0
+            and state.global_step == 1
+        ):
+            control.should_evaluate = True
         return control
 
 
@@ -514,3 +514,27 @@ def log_prediction_callback_factory(trainer: Trainer, tokenizer):
             return control
 
     return LogPredictionCallback
+
+
+class SaveAxolotlConfigtoWandBCallback(TrainerCallback):
+    """Callback to save axolotl config to wandb"""
+
+    def __init__(self, axolotl_config_path):
+        self.axolotl_config_path = axolotl_config_path
+
+    def on_train_begin(
+        self,
+        args: AxolotlTrainingArguments,  # pylint: disable=unused-argument
+        state: TrainerState,  # pylint: disable=unused-argument
+        control: TrainerControl,
+        **kwargs,  # pylint: disable=unused-argument
+    ):
+        if is_main_process():
+            try:
+                artifact = wandb.Artifact(name="axolotl-config", type="config")
+                artifact.add_file(local_path=self.axolotl_config_path)
+                wandb.run.log_artifact(artifact)
+                LOG.info("Axolotl config has been saved to WandB as an artifact.")
+            except (FileNotFoundError, ConnectionError) as err:
+                LOG.warning(f"Error while saving Axolotl config to WandB: {err}")
+        return control

src/axolotl/utils/collators.py

@@ -119,3 +119,30 @@ class DataCollatorForSeq2Seq:
             features["decoder_input_ids"] = decoder_input_ids
 
         return features
+
+
+@dataclass
+class BatchSamplerDataCollatorForSeq2Seq(DataCollatorForSeq2Seq):
+    """
+    Collator for multipack specific to the using the BatchSampler
+    """
+
+    def __call__(self, features, return_tensors=None):
+        chunked_data = {}
+        for feature in features[0].keys():
+            if feature == "length":
+                continue
+            if feature == "attention_mask":
+                arrays = [
+                    (1) * np.array(item[feature])
+                    for item in features
+                    if feature in item
+                ]
+                chunked_data[feature] = np.concatenate(arrays)
+            else:
+                arrays = [
+                    np.array(item[feature]) for item in features if feature in item
+                ]
+                chunked_data[feature] = np.concatenate(arrays)
+        features = [chunked_data]
+        return super().__call__(features, return_tensors=return_tensors)

src/axolotl/utils/config.py

@@ -49,6 +49,8 @@ def normalize_config(cfg):
     cfg.batch_size = (
         cfg.batch_size or cfg.micro_batch_size * cfg.gradient_accumulation_steps
     )
+    if cfg.eval_batch_size is None:
+        cfg.eval_batch_size = cfg.micro_batch_size
     cfg.world_size = int(os.environ.get("WORLD_SIZE", 1))
     cfg.local_rank = int(os.environ.get("LOCAL_RANK", 0))
     cfg.eval_table_size = cfg.eval_table_size or 0
@@ -68,13 +70,20 @@ def normalize_config(cfg):
     else:
         torch.backends.cuda.matmul.allow_tf32 = cfg.tf32 or False
 
-    if cfg.bf16 or cfg.bfloat16:
+    if cfg.fp8:
+        cfg.torch_dtype = torch.bfloat16
+    elif cfg.bf16 or cfg.bfloat16:
         cfg.torch_dtype = torch.bfloat16
     elif cfg.load_in_8bit or cfg.fp16 or cfg.float16:
         cfg.torch_dtype = torch.float16
     else:
         cfg.torch_dtype = torch.float32
 
+    cfg.dataset_processes = cfg.dataset_processes or os.cpu_count()
+
+    if not cfg.base_model_config:
+        cfg.base_model_config = cfg.base_model
+
     model_config = load_model_config(cfg)
     cfg.model_config_type = model_config.model_type
 
@@ -82,10 +91,42 @@ def normalize_config(cfg):
     cfg.is_llama_derived_model = (
         (hasattr(model_config, "model_type") and model_config.model_type == "llama")
         or cfg.is_llama_derived_model
-        or "llama" in cfg.base_model
+        or "llama" in cfg.base_model.lower()
         or (cfg.model_type and "llama" in cfg.model_type.lower())
     )
 
+    # figure out if the model is falcon
+    cfg.is_falcon_derived_model = (
+        (
+            hasattr(model_config, "model_type")
+            and model_config.model_type
+            in [
+                "falcon",
+                "RefinedWebModel",
+                "RefinedWeb",
+            ]
+        )
+        or cfg.is_falcon_derived_model
+        or "falcon" in cfg.base_model.lower()
+        or (cfg.model_type and "rwforcausallm" in cfg.model_type.lower())
+    )
+
+    cfg.is_mistral_derived_model = (
+        (
+            hasattr(model_config, "model_type")
+            and model_config.model_type
+            in [
+                "mistral",
+            ]
+        )
+        or cfg.is_mistral_derived_model
+        or "mistral" in cfg.base_model.lower()
+        or (cfg.model_type and "mistral" in cfg.model_type.lower())
+    )
+
+    if isinstance(cfg.learning_rate, str):
+        cfg.learning_rate = float(cfg.learning_rate)
+
     log_gpu_memory_usage(LOG, "baseline", cfg.device)
 
 
@@ -94,7 +135,7 @@ def validate_config(cfg):
         if not cfg.bf16 and not cfg.bfloat16:
             LOG.info("bf16 support detected, but not enabled for this configuration.")
     else:
-        if cfg.bf16 or cfg.bfloat16:
+        if not cfg.merge_lora and (cfg.bf16 or cfg.bfloat16):
             raise ValueError(
                 "bf16 requested, but AMP is not supported on this GPU. Requires Ampere series or above."
             )
@@ -126,6 +167,11 @@ def validate_config(cfg):
             "batch_size is not recommended. Please use gradient_accumulation_steps instead.",
             "To calculate the equivalent gradient_accumulation_steps, divide batch_size / micro_batch_size / number of gpus.",
         )
+    if cfg.eval_batch_size != cfg.micro_batch_size:
+        LOG.warning(
+            "eval_batch_size != micro_batch_size. This can lead to VRAM instability."
+        )
+
     if cfg.load_4bit:
         raise ValueError("cfg.load_4bit parameter has been deprecated")
 
@@ -151,9 +197,15 @@ def validate_config(cfg):
             if not cfg.load_in_4bit:
                 raise ValueError("Require cfg.load_in_4bit to be True for qlora")
 
+        if cfg.flash_attn_fuse_qkv or cfg.flash_attn_fuse_mlp:
+            raise ValueError("Fused modules are not supported with QLoRA")
+
     if not cfg.load_in_8bit and cfg.adapter == "lora":
         LOG.warning("We recommend setting `load_in_8bit: true` for LORA finetuning")
 
+    if cfg.adapter == "lora" and (cfg.flash_attn_fuse_qkv or cfg.flash_attn_fuse_mlp):
+        raise ValueError("Fused modules are not supported with LoRA")
+
     if cfg.relora_steps:
         if cfg.adapter not in ("lora", "qlora"):
             raise ValueError("cfg.adapter must be lora or qlora to use ReLoRA")
@@ -167,6 +219,9 @@ def validate_config(cfg):
         if cfg.lr_scheduler == "one_cycle":
             raise ValueError("ReLoRA is not compatible with the one_cycle scheduler")
 
+        if cfg.flash_attn_fuse_qkv or cfg.flash_attn_fuse_mlp:
+            raise ValueError("Fused modules are not supported with ReLoRA")
+
     if cfg.trust_remote_code:
         LOG.warning(
             "`trust_remote_code` is set to true. Please make sure that you reviewed the remote code/model."
@@ -262,6 +317,60 @@ def validate_config(cfg):
                     "`model_type: MixFormerSequentialForCausalLM` required for sample_packing"
                 )
 
+    if cfg.datasets:
+        for idx, ds_cfg in enumerate(cfg.datasets):
+            if not ds_cfg.type:
+                continue
+            if ds_cfg.type == "sharegpt:chat":
+                LOG.warning(
+                    PendingDeprecationWarning(
+                        "`type: sharegpt:chat` will soon be deprecated. simply use `type: sharegpt` instead."
+                    )
+                )
+                cfg.datasets[idx].type = "sharegpt"
+            if "sharegpt_simple" in ds_cfg.type:
+                LOG.warning(
+                    PendingDeprecationWarning(
+                        "`type: sharegpt_simple` will soon be deprecated. simply use `type: sharegpt` instead."
+                    )
+                )
+                cfg.datasets[idx].type = cfg.datasets[idx].type.replace(
+                    "sharegpt_simple", "sharegpt"
+                )
+    if cfg.save_strategy and cfg.save_steps and cfg.save_strategy != "steps":
+        raise ValueError(
+            "save_strategy and save_steps mismatch. Please set save_strategy to 'steps' or remove save_steps."
+        )
+
+    if (
+        cfg.evaluation_strategy
+        and cfg.eval_steps
+        and cfg.evaluation_strategy != "steps"
+    ):
+        raise ValueError(
+            "evaluation_strategy and eval_steps mismatch. Please set evaluation_strategy to 'steps' or remove eval_steps."
+        )
+
+    if cfg.val_set_size == 0 and (cfg.eval_steps or cfg.evaluation_strategy):
+        raise ValueError(
+            "eval_steps and evaluation_strategy are not supported with val_set_size == 0"
+        )
+
+    if (
+        cfg.sample_packing
+        and cfg.eval_table_size
+        and cfg.eval_sample_packing is not False
+    ):
+        raise ValueError(
+            "eval_table_size and eval_sample_packing are not supported together with sample_packing. Please set 'eval_sample_packing' to false."
+        )
+
+    if not cfg.adapter and (cfg.load_in_8bit or cfg.load_in_4bit):
+        raise ValueError(
+            "load_in_8bit and load_in_4bit are not supported without setting an adapter."
+            "If you want to full finetune, please turn off load_in_8bit and load_in_4bit."
+        )
+
     # TODO
     # MPT 7b
     # https://github.com/facebookresearch/bitsandbytes/issues/25

src/axolotl/utils/data.py

@@ -3,7 +3,7 @@ import functools
 import hashlib
 import logging
 from pathlib import Path
-from typing import Dict, List, Tuple, Union
+from typing import Any, Dict, List, Tuple, Union
 
 import torch
 from datasets import (
@@ -16,6 +16,7 @@ from datasets import (
 from huggingface_hub import hf_hub_download
 from transformers import PreTrainedTokenizerBase
 
+from axolotl.common.const import DEFAULT_DATASET_PREPARED_PATH
 from axolotl.datasets import ConstantLengthDataset, TokenizedPromptDataset
 from axolotl.prompt_strategies import load
 from axolotl.prompt_tokenizers import (
@@ -25,7 +26,6 @@ from axolotl.prompt_tokenizers import (
     GPTeacherPromptTokenizingStrategy,
     JeopardyPromptTokenizingStrategy,
     OpenAssistantPromptTokenizingStrategy,
-    ShareGPTPromptTokenizingStrategy,
     SummarizeTLDRPromptTokenizingStrategy,
 )
 from axolotl.prompters import (
@@ -35,8 +35,8 @@ from axolotl.prompters import (
     MultipleChoiceConcisePrompter,
     MultipleChoiceExplainPrompter,
     ReflectAlpacaPrompter,
-    ShareGPTPrompter,
     SummarizeTLDRPrompter,
+    UnsupportedPrompter,
 )
 from axolotl.utils.dict import DictDefault
 from axolotl.utils.distributed import is_main_process, zero_first
@@ -46,7 +46,6 @@ from axolotl.utils.trainer import (
 )
 
 LOG = logging.getLogger("axolotl")
-DEFAULT_DATASET_PREPARED_PATH = "last_run_prepared"
 
 
 def md5(to_hash: str, encoding: str = "utf-8") -> str:
@@ -57,9 +56,10 @@ def md5(to_hash: str, encoding: str = "utf-8") -> str:
 
 
 def prepare_dataset(cfg, tokenizer):
+    prompters = []
     if not cfg.pretraining_dataset:
         with zero_first(is_main_process()):
-            train_dataset, eval_dataset = load_prepare_datasets(
+            train_dataset, eval_dataset, prompters = load_prepare_datasets(
                 tokenizer, cfg, DEFAULT_DATASET_PREPARED_PATH
             )
     else:
@@ -72,20 +72,20 @@ def prepare_dataset(cfg, tokenizer):
         # https://discuss.huggingface.co/t/how-to-use-huggingface-trainer-streaming-datasets-without-wrapping-it-with-torchdatas-iterablewrapper/25230
         train_dataset = train_dataset.with_format("torch")
         eval_dataset = None
-        return train_dataset, eval_dataset, cfg.max_steps
+        return train_dataset, eval_dataset, cfg.max_steps, prompters
 
     with zero_first(is_main_process()):
         train_dataset, eval_dataset = process_datasets_for_packing(
-            cfg, train_dataset, eval_dataset
+            cfg, train_dataset, eval_dataset, tokenizer
         )
     if cfg.max_steps:
         total_num_steps = min(
-            calculate_total_num_steps(cfg, train_dataset, tokenizer), cfg.max_steps
+            calculate_total_num_steps(cfg, train_dataset), cfg.max_steps
         )
         LOG.info(f"Maximum number of steps set at {total_num_steps}")
     else:
-        total_num_steps = calculate_total_num_steps(cfg, train_dataset, tokenizer)
-    return train_dataset, eval_dataset, total_num_steps
+        total_num_steps = calculate_total_num_steps(cfg, train_dataset)
+    return train_dataset, eval_dataset, total_num_steps, prompters
 
 
 def load_tokenized_prepared_datasets(
@@ -111,12 +111,13 @@ def load_tokenized_prepared_datasets(
         else Path(default_dataset_prepared_path) / ds_hash
     )
     dataset = None
+    prompters = []
     use_auth_token = cfg.hf_use_auth_token
     try:
         if cfg.push_dataset_to_hub:
             dataset = load_dataset(
                 f"{cfg.push_dataset_to_hub}/{ds_hash}",
-                use_auth_token=use_auth_token,
+                token=use_auth_token,
             )
             dataset = dataset["train"]
     except Exception:  # pylint: disable=broad-except # nosec
@@ -124,7 +125,7 @@ def load_tokenized_prepared_datasets(
 
     if dataset:
         ...
-    elif any(prepared_ds_path.glob("*")):
+    elif cfg.dataset_prepared_path and any(prepared_ds_path.glob("*")):
         LOG.info(f"Loading prepared dataset from disk at {prepared_ds_path}...")
         dataset = load_from_disk(str(prepared_ds_path))
         LOG.info("Prepared dataset loaded from disk...")
@@ -149,48 +150,48 @@ def load_tokenized_prepared_datasets(
                     yield dataset
 
         # pylint: disable=invalid-name
-        for d in for_d_in_datasets(cfg.datasets):
+        for config_dataset in for_d_in_datasets(cfg.datasets):
             ds: Union[Dataset, DatasetDict] = None
             ds_from_hub = False
             try:
                 load_dataset(
-                    d.path,
-                    name=d.name,
+                    config_dataset.path,
+                    name=config_dataset.name,
                     streaming=True,
-                    use_auth_token=use_auth_token,
+                    token=use_auth_token,
                 )
                 ds_from_hub = True
-            except FileNotFoundError:
+            except (FileNotFoundError, ConnectionError):
                 pass
 
             # prefer local dataset, even if hub exists
-            local_path = Path(d.path)
+            local_path = Path(config_dataset.path)
             if local_path.exists():
                 if local_path.is_dir():
                     # TODO dirs with arrow or parquet files could be loaded with `load_from_disk`
                     ds = load_dataset(
-                        d.path,
-                        name=d.name,
-                        data_files=d.data_files,
+                        config_dataset.path,
+                        name=config_dataset.name,
+                        data_files=config_dataset.data_files,
                         streaming=False,
                         split=None,
                     )
                 elif local_path.is_file():
                     ds_type = "json"
-                    if d.ds_type:
-                        ds_type = d.ds_type
-                    elif ".parquet" in d.path:
+                    if config_dataset.ds_type:
+                        ds_type = config_dataset.ds_type
+                    elif ".parquet" in config_dataset.path:
                         ds_type = "parquet"
-                    elif ".arrow" in d.path:
+                    elif ".arrow" in config_dataset.path:
                         ds_type = "arrow"
-                    elif ".csv" in d.path:
+                    elif ".csv" in config_dataset.path:
                         ds_type = "csv"
-                    elif ".txt" in d.path:
+                    elif ".txt" in config_dataset.path:
                         ds_type = "text"
                     ds = load_dataset(
                         ds_type,
-                        name=d.name,
-                        data_files=d.path,
+                        name=config_dataset.name,
+                        data_files=config_dataset.path,
                         streaming=False,
                         split=None,
                     )
@@ -200,143 +201,83 @@ def load_tokenized_prepared_datasets(
                     )
             elif ds_from_hub:
                 ds = load_dataset(
-                    d.path,
-                    name=d.name,
+                    config_dataset.path,
+                    name=config_dataset.name,
                     streaming=False,
-                    data_files=d.data_files,
-                    use_auth_token=use_auth_token,
+                    data_files=config_dataset.data_files,
+                    token=use_auth_token,
                 )
             else:
-                fp = hf_hub_download(
-                    repo_id=d.path,
-                    repo_type="dataset",
-                    filename=d.data_files,
-                )
+                if isinstance(config_dataset.data_files, str):
+                    fp = hf_hub_download(
+                        repo_id=config_dataset.path,
+                        repo_type="dataset",
+                        filename=config_dataset.data_files,
+                    )
+                elif isinstance(config_dataset.data_files, list):
+                    fp = []
+                    for file in config_dataset.data_files:
+                        fp.append(
+                            hf_hub_download(
+                                repo_id=config_dataset.path,
+                                repo_type="dataset",
+                                filename=file,
+                            )
+                        )
+                else:
+                    raise ValueError(
+                        "data_files must be either a string or list of strings"
+                    )
                 ds = load_dataset(
-                    "json", name=d.name, data_files=fp, streaming=False, split=None
+                    "json",
+                    name=config_dataset.name,
+                    data_files=fp,
+                    streaming=False,
+                    split=None,
                 )
             if not ds:
                 raise ValueError("unhandled dataset load")
             # support for using a subset of the data
-            if d.shards:
+            if config_dataset.shards:
                 if "train" in ds:
                     ds = ds.shuffle(seed=seed)["train"].shard(
-                        num_shards=d.shards, index=0
+                        num_shards=config_dataset.shards, index=0
                     )
                 else:
-                    ds = ds.shuffle(seed=seed).shard(num_shards=d.shards, index=0)
+                    ds = ds.shuffle(seed=seed).shard(
+                        num_shards=config_dataset.shards, index=0
+                    )
 
             d_base_type = d_prompt_style = None
-            d_type = d.type
+            d_type = config_dataset.type
             if isinstance(d_type, str):
                 d_type_split = d_type.split(":")
                 d_base_type = d_type_split[0]
                 d_prompt_style = d_type_split[1] if len(d_type_split) > 1 else None
             if "train" in ds:
                 ds = ds["train"]
-            if (
-                "input_ids" in ds.features
-                and "attention_mask" in ds.features
-                and "labels" in ds.features
+            elif (
+                isinstance(ds, DatasetDict)
+                and config_dataset.train_on_split
+                and config_dataset.train_on_split in ds
             ):
-                # dataset is already tokenized, just drop it straight in
-                datasets.append(ds)
-            elif isinstance(d.type, DictDefault):
-                ds_strategy = load("user_defined", tokenizer, cfg, d.type.to_dict())
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
-                datasets.append(ds_wrapper)
-            elif ds_strategy := load(d.type, tokenizer, cfg, d):
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
-                datasets.append(ds_wrapper)
-            elif d_base_type == "alpaca":
-                ds_strategy = AlpacaPromptTokenizingStrategy(
-                    AlpacaPrompter(d_prompt_style),
-                    tokenizer,
-                    cfg.train_on_inputs,
-                    cfg.sequence_len,
-                )
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
-                datasets.append(ds_wrapper)
-            elif d_base_type == "explainchoice":
-                ds_strategy = AlpacaMultipleChoicePromptTokenizingStrategy(
-                    MultipleChoiceExplainPrompter(d_prompt_style),
-                    tokenizer,
-                    cfg.train_on_inputs,
-                    cfg.sequence_len,
-                )
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
-                datasets.append(ds_wrapper)
-            elif d_base_type == "concisechoice":
-                ds_strategy = AlpacaMultipleChoicePromptTokenizingStrategy(
-                    MultipleChoiceConcisePrompter(d_prompt_style),
-                    tokenizer,
-                    cfg.train_on_inputs,
-                    cfg.sequence_len,
-                )
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
-                datasets.append(ds_wrapper)
-            elif d_base_type == "summarizetldr":
-                ds_strategy = SummarizeTLDRPromptTokenizingStrategy(
-                    SummarizeTLDRPrompter(d_prompt_style),
-                    tokenizer,
-                    cfg.train_on_inputs,
-                    cfg.sequence_len,
-                )
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
-                datasets.append(ds_wrapper)
-            elif d_base_type == "jeopardy":
-                ds_strategy = JeopardyPromptTokenizingStrategy(
-                    JeopardyPrompter(d_prompt_style),
-                    tokenizer,
-                    cfg.train_on_inputs,
-                    cfg.sequence_len,
-                )
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
-                datasets.append(ds_wrapper)
-            elif d_base_type == "oasst":
-                ds_strategy = OpenAssistantPromptTokenizingStrategy(
-                    AlpacaPrompter(d_prompt_style),
-                    tokenizer,
-                    cfg.train_on_inputs,
-                    cfg.sequence_len,
-                )
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
-                datasets.append(ds_wrapper)
-            elif d_base_type == "gpteacher":
-                ds_strategy = GPTeacherPromptTokenizingStrategy(
-                    GPTeacherPrompter(d_prompt_style),
-                    tokenizer,
-                    cfg.train_on_inputs,
-                    cfg.sequence_len,
-                )
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
-                datasets.append(ds_wrapper)
-            elif d_base_type == "reflection":
-                ds_strategy = AlpacaReflectionPTStrategy(
-                    ReflectAlpacaPrompter(d_prompt_style),
-                    tokenizer,
-                    cfg.train_on_inputs,
-                    cfg.sequence_len,
-                )
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
-                datasets.append(ds_wrapper)
-            elif d_base_type == "sharegpt":
-                ds_strategy = ShareGPTPromptTokenizingStrategy(
-                    ShareGPTPrompter(d_prompt_style),
-                    tokenizer,
-                    cfg.train_on_inputs,
-                    cfg.sequence_len,
-                )
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
-                datasets.append(ds_wrapper)
-            else:
-                suffix = ""
-                if ":load_" in d.type:
-                    suffix = f" Did you mean {d.type.replace(':load_', '.load_')}?"
-                LOG.error(f"unhandled prompt tokenization strategy: {d.type}. {suffix}")
+                ds = ds[config_dataset.train_on_split]
+            elif isinstance(ds, DatasetDict):
                 raise ValueError(
-                    f"unhandled prompt tokenization strategy: {d.type} {suffix}"
+                    f"no train split found for dataset {config_dataset.path}, you may specify a split with 'train_on_split: `"
                 )
+
+            dataset_wrapper, dataset_prompter = get_dataset_wrapper(
+                config_dataset=config_dataset,
+                dataset=ds,
+                tokenizer=tokenizer,
+                cfg=cfg,
+                d_base_type=d_base_type,
+                d_prompt_style=d_prompt_style,
+            )
+            datasets.append(dataset_wrapper)
+            prompters.append(dataset_prompter)
+
         LOG.info("merging datasets")
         dataset = concatenate_datasets(datasets)
 
@@ -354,14 +295,14 @@ def load_tokenized_prepared_datasets(
                     f"{cfg.push_dataset_to_hub}/{ds_hash}", private=True
                 )
 
-    return dataset
+    return dataset, prompters
 
 
 def load_prepare_datasets(
     tokenizer: PreTrainedTokenizerBase,
     cfg,
     default_dataset_prepared_path,
-) -> Tuple[Dataset, Dataset]:
+) -> Tuple[Dataset, Dataset, List[Any]]:
     max_packed_sequence_len = (
         cfg.max_packed_sequence_len if cfg.max_packed_sequence_len else cfg.sequence_len
     )
@@ -370,6 +311,7 @@ def load_prepare_datasets(
     )  # make sure we don't accidentally set it larger than sequence_len
 
     tokenizer_name = tokenizer.__class__.__name__
+    prompters = []
     if cfg.max_packed_sequence_len is not None:
         # see if we can go ahead and load the stacked dataset
         seed = f"@{str(cfg.seed)}" if cfg.seed else ""
@@ -403,7 +345,7 @@ def load_prepare_datasets(
                 )
                 dataset = load_dataset(
                     f"{cfg.push_dataset_to_hub}/{ds_hash}",
-                    use_auth_token=use_auth_token,
+                    token=use_auth_token,
                 )
                 dataset = dataset["train"]
         except Exception:  # pylint: disable=broad-except # nosec
@@ -411,7 +353,7 @@ def load_prepare_datasets(
 
         if dataset:
             ...
-        elif any(prepared_ds_path.glob("*")):
+        elif cfg.dataset_prepared_path and any(prepared_ds_path.glob("*")):
             LOG.info(
                 f"Loading prepared packed dataset from disk at {prepared_ds_path}..."
             )
@@ -425,7 +367,7 @@ def load_prepare_datasets(
                     f"{cfg.push_dataset_to_hub}/{ds_hash}", private=True
                 )
         else:
-            dataset = load_tokenized_prepared_datasets(
+            dataset, prompters = load_tokenized_prepared_datasets(
                 tokenizer, cfg, default_dataset_prepared_path
             )
 
@@ -467,7 +409,7 @@ def load_prepare_datasets(
                         private=True,
                     )
     else:
-        dataset = load_tokenized_prepared_datasets(
+        dataset, prompters = load_tokenized_prepared_datasets(
             tokenizer, cfg, default_dataset_prepared_path
         )
 
@@ -518,7 +460,144 @@ def load_prepare_datasets(
         train_dataset = dataset
         eval_dataset = None
 
-    return train_dataset, eval_dataset
+    return train_dataset, eval_dataset, prompters
+
+
+def get_dataset_wrapper(
+    config_dataset, dataset, tokenizer, cfg, d_base_type, d_prompt_style
+):
+    dataset_wrapper = None
+    dataset_prompter = None
+
+    if (
+        "input_ids" in dataset.features
+        and "attention_mask" in dataset.features
+        and "labels" in dataset.features
+    ):
+        # dataset is already tokenized, just drop it straight in
+        dataset_prompter = UnsupportedPrompter()
+        dataset_wrapper = dataset
+    elif isinstance(config_dataset.type, DictDefault):
+        ds_strategy = load(
+            "user_defined", tokenizer, cfg, config_dataset.type.to_dict()
+        )
+        dataset_prompter = UnsupportedPrompter()
+        dataset_wrapper = TokenizedPromptDataset(
+            ds_strategy, dataset, process_count=cfg.dataset_processes
+        )
+    elif ds_strategy := load(config_dataset.type, tokenizer, cfg, config_dataset):
+        dataset_prompter = UnsupportedPrompter()
+        dataset_wrapper = TokenizedPromptDataset(
+            ds_strategy, dataset, process_count=cfg.dataset_processes
+        )
+    elif d_base_type == "alpaca":
+        dataset_prompter = AlpacaPrompter(d_prompt_style)
+        ds_strategy = AlpacaPromptTokenizingStrategy(
+            dataset_prompter,
+            tokenizer,
+            cfg.train_on_inputs,
+            cfg.sequence_len,
+        )
+        ds_wrapper = TokenizedPromptDataset(
+            ds_strategy, dataset, process_count=cfg.dataset_processes
+        )
+        dataset_wrapper = ds_wrapper
+    elif d_base_type == "explainchoice":
+        dataset_prompter = MultipleChoiceExplainPrompter(d_prompt_style)
+        ds_strategy = AlpacaMultipleChoicePromptTokenizingStrategy(
+            dataset_prompter,
+            tokenizer,
+            cfg.train_on_inputs,
+            cfg.sequence_len,
+        )
+        ds_wrapper = TokenizedPromptDataset(
+            ds_strategy, dataset, process_count=cfg.dataset_processes
+        )
+        dataset_wrapper = ds_wrapper
+    elif d_base_type == "concisechoice":
+        dataset_prompter = MultipleChoiceConcisePrompter(d_prompt_style)
+        ds_strategy = AlpacaMultipleChoicePromptTokenizingStrategy(
+            dataset_prompter,
+            tokenizer,
+            cfg.train_on_inputs,
+            cfg.sequence_len,
+        )
+        ds_wrapper = TokenizedPromptDataset(
+            ds_strategy, dataset, process_count=cfg.dataset_processes
+        )
+        dataset_wrapper = ds_wrapper
+    elif d_base_type == "summarizetldr":
+        dataset_prompter = SummarizeTLDRPrompter(d_prompt_style)
+        ds_strategy = SummarizeTLDRPromptTokenizingStrategy(
+            dataset_prompter,
+            tokenizer,
+            cfg.train_on_inputs,
+            cfg.sequence_len,
+        )
+        ds_wrapper = TokenizedPromptDataset(
+            ds_strategy, dataset, process_count=cfg.dataset_processes
+        )
+        dataset_wrapper = ds_wrapper
+    elif d_base_type == "jeopardy":
+        dataset_prompter = JeopardyPrompter(d_prompt_style)
+        ds_strategy = JeopardyPromptTokenizingStrategy(
+            dataset_prompter,
+            tokenizer,
+            cfg.train_on_inputs,
+            cfg.sequence_len,
+        )
+        ds_wrapper = TokenizedPromptDataset(
+            ds_strategy, dataset, process_count=cfg.dataset_processes
+        )
+        dataset_wrapper = ds_wrapper
+    elif d_base_type == "oasst":
+        dataset_prompter = AlpacaPrompter(d_prompt_style)
+        ds_strategy = OpenAssistantPromptTokenizingStrategy(
+            dataset_prompter,
+            tokenizer,
+            cfg.train_on_inputs,
+            cfg.sequence_len,
+        )
+        ds_wrapper = TokenizedPromptDataset(
+            ds_strategy, dataset, process_count=cfg.dataset_processes
+        )
+        dataset_wrapper = ds_wrapper
+    elif d_base_type == "gpteacher":
+        dataset_prompter = GPTeacherPrompter(d_prompt_style)
+        ds_strategy = GPTeacherPromptTokenizingStrategy(
+            dataset_prompter,
+            tokenizer,
+            cfg.train_on_inputs,
+            cfg.sequence_len,
+        )
+        ds_wrapper = TokenizedPromptDataset(
+            ds_strategy, dataset, process_count=cfg.dataset_processes
+        )
+        dataset_wrapper = ds_wrapper
+    elif d_base_type == "reflection":
+        dataset_prompter = ReflectAlpacaPrompter(d_prompt_style)
+        ds_strategy = AlpacaReflectionPTStrategy(
+            dataset_prompter,
+            tokenizer,
+            cfg.train_on_inputs,
+            cfg.sequence_len,
+        )
+        ds_wrapper = TokenizedPromptDataset(
+            ds_strategy, dataset, process_count=cfg.dataset_processes
+        )
+        dataset_wrapper = ds_wrapper
+    else:
+        suffix = ""
+        if ":load_" in config_dataset.type:
+            suffix = f" Did you mean {config_dataset.type.replace(':load_', '.load_')}?"
+        LOG.error(
+            f"unhandled prompt tokenization strategy: {config_dataset.type}. {suffix}"
+        )
+        raise ValueError(
+            f"unhandled prompt tokenization strategy: {config_dataset.type} {suffix}"
+        )
+
+    return dataset_wrapper, dataset_prompter
 
 
 def encode_pretraining(

src/axolotl/utils/dataloader.py

@@ -3,6 +3,9 @@ import hashlib
 import itertools
 import logging
 import math
+import time
+from queue import Queue
+from threading import Thread
 from typing import Any, Callable, List, Union
 
 import numba
@@ -149,6 +152,8 @@ class MultipackDistributedDataloader:
         packing_efficiency_estimate: float = 1.0,
         sample_packing_seq_len_multiplier: int = 1,
         device_count: int = 1,
+        prefetch_max: int = 1000,
+        num_epochs: int = 1,
     ):
         # Dataset
         self.dataset = dataset
@@ -167,6 +172,7 @@ class MultipackDistributedDataloader:
         self.seq_max_length = seq_max_length
         self.batch_max_length = batch_size * seq_max_length
         self.collate_fn = collate_fn
+        self.num_epochs = num_epochs
 
         self.num_replicas = 1
         self.rank = 0
@@ -177,6 +183,44 @@ class MultipackDistributedDataloader:
         self.packing_efficiency_estimate = packing_efficiency_estimate or 1.0
         self.device_count = device_count
 
+        # maxsize is maximum number of samples in queue
+        self.prefetch_max = prefetch_max
+        self.queue: Queue = Queue(maxsize=prefetch_max)
+        self.thread = None
+
+    def _worker(self):
+        LOG.info(
+            f"[WORKER] Epochs: {self.num_epochs}, Samples: {self.len_w_stats()*self.batch_size}"
+        )
+        for epoch in range(self.num_epochs):
+            for sample in self._internal_batch_generator():
+                while True:
+                    if self.queue.full():
+                        time.sleep(1)
+                    else:
+                        break
+                self.queue.put(sample)
+
+            # stop the queue when epoch is done
+            self.queue.put(None)
+
+    def __iter__(self):
+        if hasattr(self.sampler, "set_epoch"):
+            new_epoch = self.sampler.epoch + 1
+            self.sampler.set_epoch(new_epoch)
+            LOG.info(f"calling sampler.set_epoch({new_epoch})")
+
+        if self.thread is None:
+            self.thread = Thread(target=self._worker, daemon=True)
+            self.thread.start()
+
+        while True:
+            item = self.queue.get()
+
+            if item is None:
+                break
+            yield item
+
     def generate_batches(self, set_stats=False):
         LOG.info("generating packed batches")
         if self.sampler:
@@ -206,11 +250,7 @@ class MultipackDistributedDataloader:
 
         return batches, totseqs
 
-    def __iter__(self):
-        if hasattr(self.sampler, "set_epoch"):
-            new_epoch = self.sampler.epoch + 1
-            self.sampler.set_epoch(new_epoch)
-            LOG.info(f"calling sampler.set_epoch({new_epoch})")
+    def _internal_batch_generator(self):
         all_batches, _ = self.generate_batches(set_stats=True)
         features = self.dataset.features.keys()
         len_remaining = self._len_est()

src/axolotl/utils/distributed.py

@@ -50,6 +50,17 @@ def get_world_size():
     return int(os.getenv("WORLD_SIZE", "1"))
 
 
+@contextmanager
+def zero_only():
+    """
+    Context manager that only runs the enclosed block on the main rank.
+    """
+    if is_main_process():
+        yield
+    else:
+        yield None
+
+
 @contextmanager
 def zero_first(is_main):
     """
@@ -77,7 +88,9 @@ def gather_scalar_from_all_ranks(fn, world_size=1):  # pylint: disable=invalid-n
     value_scalar = fn()
     if not is_distributed():
         return [value_scalar]
-    value_tensor = torch.tensor(value_scalar, device=dist.get_rank()).float()
+    value_tensor = torch.tensor(
+        value_scalar, device=torch.cuda.current_device()
+    ).float()
 
     if not is_main_process():
         dist.gather(value_tensor, dst=0)
@@ -137,9 +150,13 @@ def compute_and_broadcast(fn):  # pylint: disable=invalid-name
     """
     if is_main_process():
         value_scalar = fn()
-        value_tensor = torch.tensor(value_scalar, device=dist.get_rank()).float()
+        value_tensor = torch.tensor(
+            value_scalar, device=torch.cuda.current_device()
+        ).float()
     else:
-        value_tensor = torch.tensor(0.0, device=dist.get_rank())  # Placeholder tensor
+        value_tensor = torch.tensor(
+            0.0, device=torch.cuda.current_device()
+        )  # Placeholder tensor
 
     # Broadcast the tensor to all processes.
     barrier()
@@ -164,7 +181,9 @@ def gather_from_all_ranks(fn, world_size=1):  # pylint: disable=invalid-name
     - A list of computed values from all ranks if on the gathering rank, otherwise None.
     """
     value_scalar = fn()
-    value_tensor = torch.tensor(value_scalar, device=dist.get_rank()).float()
+    value_tensor = torch.tensor(
+        value_scalar, device=torch.cuda.current_device()
+    ).float()
 
     # Placeholder tensor for gathering results
     if is_main_process():

src/axolotl/utils/models.py

@@ -1,5 +1,4 @@
 """Module for models and model loading"""
-import importlib
 import logging
 import math
 import os
@@ -10,7 +9,9 @@ import torch
 import transformers
 from optimum.bettertransformer import BetterTransformer
 from peft import PeftConfig, prepare_model_for_kbit_training
+from peft.tuners.lora import QuantLinear
 from transformers import (  # noqa: F401
+    AddedToken,
     AutoConfig,
     AutoModelForCausalLM,
     AutoTokenizer,
@@ -30,7 +31,7 @@ LOG = logging.getLogger("axolotl")
 
 def load_model_config(cfg):
     model_config_name = cfg.base_model_config or cfg.base_model
-    trust_remote_code: bool = False or cfg.trust_remote_code
+    trust_remote_code = cfg.trust_remote_code is True
     return AutoConfig.from_pretrained(
         model_config_name, trust_remote_code=trust_remote_code
     )
@@ -71,20 +72,31 @@ def load_tokenizer(cfg):
         # set a pad_token, but use eos_token so we don't add a new token
         tokenizer.pad_token = LLAMA_DEFAULT_EOS_TOKEN
 
-    LOG.debug(f"EOS: {tokenizer.eos_token_id} / {tokenizer.eos_token}")
-    LOG.debug(f"BOS: {tokenizer.bos_token_id} / {tokenizer.bos_token}")
-    LOG.debug(f"PAD: {tokenizer.pad_token_id} / {tokenizer.pad_token}")
-    LOG.debug(f"UNK: {tokenizer.unk_token_id} / {tokenizer.unk_token}")
-
     if tokenizer.__class__.__name__ == "GPTNeoXTokenizerFast":
         tokenizer.add_special_tokens({"pad_token": "[PAD]"})
         os.environ["TOKENIZERS_PARALLELISM"] = "false"
 
+    # Mistral's official FA implementation requires left padding
+    if cfg.is_mistral_derived_model and cfg.flash_attention and not cfg.sample_packing:
+        tokenizer.padding_side = "left"
+
     if cfg.special_tokens:
         for k, val in cfg.special_tokens.items():
-            tokenizer.add_special_tokens({k: val})
+            tokenizer.add_special_tokens(
+                {k: AddedToken(val, rstrip=False, lstrip=False, normalized=False)}
+            )
     if cfg.tokens:
-        tokenizer.add_tokens(list(cfg.tokens))
+        tokenizer.add_tokens(
+            [
+                AddedToken(token, rstrip=False, lstrip=False, normalized=False)
+                for token in cfg.tokens
+            ]
+        )
+
+    LOG.debug(f"EOS: {tokenizer.eos_token_id} / {tokenizer.eos_token}")
+    LOG.debug(f"BOS: {tokenizer.bos_token_id} / {tokenizer.bos_token}")
+    LOG.debug(f"PAD: {tokenizer.pad_token_id} / {tokenizer.pad_token}")
+    LOG.debug(f"UNK: {tokenizer.unk_token_id} / {tokenizer.unk_token}")
 
     return tokenizer
 
@@ -113,26 +125,29 @@ def load_model(
 
             replace_btlm_attn_with_flash_attn(cfg.base_model)
 
-    if hasattr(model_config, "model_type") and model_config.model_type in [
-        "falcon",
-        "RefinedWebModel",
-        "RefinedWeb",
-    ]:
-        if cfg.flash_attention:
-            from axolotl.monkeypatch.falcon_attn_hijack_flash import (
-                replace_falcon_attn_with_flash_attn,
+    if (
+        hasattr(model_config, "model_type")
+        and model_config.model_type == "stablelm_epoch"
+    ):
+        if cfg.flash_attention and cfg.sample_packing:
+            from axolotl.monkeypatch.stablelm_attn_hijack_flash import (
+                replace_stablelm_attn_with_flash_attn,
             )
 
-            replace_falcon_attn_with_flash_attn()
+            replace_stablelm_attn_with_flash_attn(cfg.base_model)
 
-    if cfg.is_llama_derived_model and cfg.flash_attention:
+    if cfg.is_llama_derived_model and cfg.flash_attention and cfg.sample_packing:
         if cfg.device not in ["mps", "cpu"] and not inference:
             from axolotl.monkeypatch.llama_attn_hijack_flash import (
                 replace_llama_attn_with_flash_attn,
             )
 
-            LOG.info("patching with flash attention")
-            replace_llama_attn_with_flash_attn(packed=cfg.sample_packing)
+            LOG.info("patching with flash attention for sample packing")
+            replace_llama_attn_with_flash_attn(
+                packed=cfg.sample_packing,
+                cross_entropy=cfg.flash_attn_cross_entropy,
+                rms_norm=cfg.flash_attn_rms_norm,
+            )
     elif cfg.is_llama_derived_model and cfg.xformers_attention:
         from axolotl.monkeypatch.llama_attn_hijack_xformers import (
             hijack_llama_attention,
@@ -157,6 +172,14 @@ def load_model(
         # Note: This might overwrite previous additional_special_tokens
         tokenizer.add_special_tokens({"additional_special_tokens": [MEM_TOKEN]})
 
+    if cfg.is_mistral_derived_model and cfg.flash_attention and cfg.sample_packing:
+        from axolotl.monkeypatch.mistral_attn_hijack_flash import (
+            replace_mistral_attn_with_flash_attn,
+        )
+
+        LOG.info("patching with flash attention")
+        replace_mistral_attn_with_flash_attn(packed=cfg.sample_packing)
+
     if cfg.is_llama_derived_model and cfg.xpos_rope:
         from axolotl.monkeypatch.xpos_rope_llama_monkey_patch import (
             replace_llama_rope_with_xpos_rope,
@@ -175,21 +198,11 @@ def load_model(
         LOG.info("patching _expand_mask")
         hijack_expand_mask()
 
-    # special handling b/c remote MixFormers code doesn't have _no_split_modules set
-    if (
-        "MixFormerSequentialConfig" in model_config.__class__.__name__
-        and cfg.model_type == "AutoModelForCausalLM"
-    ):
-        module_name = model_config.__class__.__module__.replace(
-            ".configuration_mixformer_sequential", ".modeling_mixformer_sequential"
-        )
-        modeling_phi = importlib.import_module(module_name)
-        # pylint:disable=protected-access
-        modeling_phi.MixFormerSequentialForCausalLM._no_split_modules = [
-            "ParallelBlock"
-        ]
-
     model_kwargs = {}
+
+    model_kwargs["device_map"] = cfg.device_map
+    model_kwargs["torch_dtype"] = cfg.torch_dtype
+
     if cfg.model_revision:
         model_kwargs["revision"] = cfg.model_revision
     if cfg.gptq:
@@ -212,6 +225,15 @@ def load_model(
             bnb_4bit_use_double_quant=True,
             bnb_4bit_quant_type="nf4",
         )
+    # sample packing uses custom FA2 patch
+    if cfg.flash_attention and not cfg.sample_packing:
+        if (
+            cfg.is_llama_derived_model
+            or cfg.is_falcon_derived_model
+            or cfg.is_mistral_derived_model
+        ):
+            model_kwargs["use_flash_attention_2"] = True
+
     try:
         if cfg.is_llama_derived_model and not cfg.trust_remote_code and not cfg.gptq:
             from transformers import LlamaForCausalLM
@@ -226,12 +248,24 @@ def load_model(
             model = LlamaForCausalLM.from_pretrained(
                 base_model,
                 config=config,
-                device_map=cfg.device_map,
                 load_in_8bit=cfg.load_in_8bit and cfg.adapter is not None,
                 load_in_4bit=cfg.load_in_4bit and cfg.adapter is not None,
-                torch_dtype=cfg.torch_dtype,
                 **model_kwargs,
             )
+
+            if cfg.flash_attention and not inference:
+                from axolotl.monkeypatch.llama_attn_hijack_flash import (
+                    replace_llama_mlp_with_swiglu,
+                    replace_llama_qkv_with_fused,
+                )
+
+                if cfg.flash_attn_fuse_mlp:
+                    LOG.info("patching with SwiGLU")
+                    replace_llama_mlp_with_swiglu(model)
+
+                if cfg.flash_attn_fuse_qkv:
+                    LOG.info("patching with fused QKV")
+                    replace_llama_qkv_with_fused(model)
         # elif model_type == "GPTNeoXForCausalLM" and cfg.flash_attention:
         #     This is a WIP, still an issue with the backward pass
         #     RuntimeError: grad can be implicitly created only for scalar outputs
@@ -263,28 +297,22 @@ def load_model(
 
             model = MixFormerSequentialForCausalLM.from_pretrained(
                 base_model,
-                device_map=cfg.device_map,
                 load_in_8bit=cfg.load_in_8bit and cfg.adapter is not None,
                 load_in_4bit=cfg.load_in_4bit and cfg.adapter is not None,
-                torch_dtype=cfg.torch_dtype,
                 **model_kwargs,
             )
         elif model_type and not cfg.trust_remote_code:
             if cfg.gptq:
                 model = AutoModelForCausalLM.from_pretrained(
                     base_model,
-                    device_map=cfg.device_map,
-                    torch_dtype=cfg.torch_dtype,
                     trust_remote_code=cfg.trust_remote_code or False,
                     **model_kwargs,
                 )
             else:
                 model = getattr(transformers, model_type).from_pretrained(
                     base_model,
-                    device_map=cfg.device_map,
                     load_in_8bit=cfg.load_in_8bit and cfg.adapter is not None,
                     load_in_4bit=cfg.load_in_4bit and cfg.adapter is not None,
-                    torch_dtype=cfg.torch_dtype,
                     trust_remote_code=cfg.trust_remote_code or False,
                     **model_kwargs,
                 )
@@ -309,16 +337,22 @@ def load_model(
             ):
                 config.max_sequence_length = cfg.sequence_len
                 LOG.warning(f"increasing context length to {cfg.sequence_len}")
-            model = AutoModelForCausalLM.from_pretrained(
-                base_model,
-                config=config,
-                device_map=cfg.device_map,
-                load_in_8bit=cfg.load_in_8bit and cfg.adapter is not None,
-                load_in_4bit=cfg.load_in_4bit and cfg.adapter is not None,
-                torch_dtype=cfg.torch_dtype,
-                trust_remote_code=cfg.trust_remote_code or False,
-                **model_kwargs,
-            )
+            if cfg.gptq:
+                model = AutoModelForCausalLM.from_pretrained(
+                    base_model,
+                    config=config,
+                    trust_remote_code=cfg.trust_remote_code or False,
+                    **model_kwargs,
+                )
+            else:
+                model = AutoModelForCausalLM.from_pretrained(
+                    base_model,
+                    config=config,
+                    load_in_8bit=cfg.load_in_8bit and cfg.adapter is not None,
+                    load_in_4bit=cfg.load_in_4bit and cfg.adapter is not None,
+                    trust_remote_code=cfg.trust_remote_code or False,
+                    **model_kwargs,
+                )
     except Exception as err:  # pylint: disable=broad-exception-caught
         LOG.error(
             "Exception raised attempting to load model, retrying with AutoModelForCausalLM"
@@ -326,10 +360,8 @@ def load_model(
         LOG.exception(err)
         model = AutoModelForCausalLM.from_pretrained(
             base_model,
-            device_map=cfg.device_map,
             load_in_8bit=cfg.load_in_8bit and cfg.adapter is not None,
             load_in_4bit=cfg.load_in_4bit and cfg.adapter is not None,
-            torch_dtype=cfg.torch_dtype,
             trust_remote_code=cfg.trust_remote_code or False,
             **model_kwargs,
         )
@@ -354,6 +386,20 @@ def load_model(
         )
         model.config.max_position_embeddings = cfg.sequence_len
 
+    if (
+        hasattr(model.config, "bos_token_id")
+        and model.config.bos_token_id
+        and model.config.bos_token_id != tokenizer.bos_token_id
+    ):
+        model.config.bos_token_id = tokenizer.bos_token_id
+
+    if (
+        hasattr(model.config, "eos_token_id")
+        and model.config.eos_token_id
+        and model.config.eos_token_id != tokenizer.eos_token_id
+    ):
+        model.config.eos_token_id = tokenizer.eos_token_id
+
     if model.device.type == "cuda":
         log_gpu_memory_usage(LOG, "after model load", model.device)
 
@@ -396,14 +442,7 @@ def load_model(
     if cfg.ddp and not load_in_8bit:
         model.to(f"cuda:{cfg.local_rank}")
 
-    if (
-        torch.cuda.device_count() > 1
-        and int(os.getenv("WORLD_SIZE", "1")) > 1
-        and (cfg.load_in_4bit)
-    ):
-        # llama is PROBABLY model parallelizable, but the default isn't that it is
-        # so let's only set it for the 4bit, see
-        # https://github.com/johnsmith0031/alpaca_lora_4bit/blob/08b3fca4a4a9e0d3945be1bab4529f100a428636/finetune.py#L130-L133
+    if torch.cuda.device_count() > 1 and int(os.getenv("WORLD_SIZE", "1")) == 1:
         setattr(model, "is_parallelizable", True)
         setattr(model, "model_parallel", True)
 
@@ -466,10 +505,14 @@ def load_llama_adapter(model, cfg):
 
 
 def find_all_linear_names(model):
-    cls = (bnb.nn.Linear4bit, bnb.nn.Linear8bitLt, torch.nn.Linear)
+    cls = (bnb.nn.Linear4bit, bnb.nn.Linear8bitLt, torch.nn.Linear, QuantLinear)
     lora_module_names = set()
     for name, module in model.named_modules():
-        if isinstance(module, cls):
+        if (
+            isinstance(module, cls)
+            or "Linear" in module.__class__.__name__
+            and module.__class__.__name__ not in ("LlamaLinearScalingRotaryEmbedding",)
+        ):
             names = name.split(".")
             lora_module_names.add(names[0] if len(names) == 1 else names[-1])
 

src/axolotl/utils/samplers/__init__.py

@@ -0,0 +1,4 @@
+"""
+axolotl samplers module
+"""
+from .multipack import MultipackBatchSampler  # noqa: F401

src/axolotl/utils/samplers/multipack.py

@@ -0,0 +1,193 @@
+# pylint: skip-file
+"""
+Multipack Batch Sampler
+"""
+import logging
+import math
+import os
+from typing import Any, Iterable, List, Union
+
+import numba
+import numpy as np
+from torch.utils.data import BatchSampler, Sampler
+
+LOG = logging.getLogger("axolotl.utils.samplers.multipack")
+
+
+@numba.njit
+def ffd_check(a: np.ndarray, c: int, n: int):
+    # First-fit-decreasing bin packing
+    # Check if a[] could fit in n bins with capacity c
+    # https://en.wikipedia.org/wiki/First-fit-decreasing_bin_packing
+
+    a = np.sort(a)[::-1]
+    bins = np.full((n,), c, dtype=a.dtype)
+    for size in a:
+        not_found = True
+        for idx in range(n):
+            if bins[idx] >= size:
+                bins[idx] -= size
+                not_found = False
+                break
+
+        if not_found:
+            return False
+
+    return True
+
+
+@numba.njit
+def ffd_with_result(a: np.ndarray, c: int, start_index: int):
+    # First-fit-decreasing bin packing (with result return)
+
+    indices = np.argsort(a)[::-1]
+    a = a[indices]
+
+    bins: List[Any] = []
+    bins_result: List[Any] = []
+    for a_id, size in enumerate(a):
+        add_new = True
+        for idx in range(len(bins)):
+            if bins[idx] >= size:
+                bins[idx] -= size
+                bins_result[idx].append(indices[a_id] + start_index)
+                add_new = False
+                break
+
+        if add_new:
+            bins.append(c - size)
+            bins_result.append([indices[a_id] + start_index])
+
+    return bins_result
+
+
+@numba.njit
+def allocate(
+    lengths: np.ndarray, lengths_cumsum: np.ndarray, rank: int, c: int, n: int
+):
+    # Dynamic batch allocator, similar to Multifit
+    # https://en.wikipedia.org/wiki/Multifit_algorithm
+    # ~99.5% efficiency on OpenChat training set (12 * 2048 ctx len)
+
+    s = 0
+    start_index = 0
+    result = []
+
+    while True:
+        # binary search [l, r)
+        left = 1
+        right = 1 + np.searchsorted(lengths_cumsum[start_index:], s + c * n, "right")
+
+        while right - left > 1:
+            mid = (left + right) // 2
+            if ffd_check(lengths[start_index : start_index + mid], c, n):
+                left = mid
+            else:
+                right = mid
+
+        # use length l
+        batch = ffd_with_result(
+            lengths[start_index : start_index + left], c, start_index
+        )
+        assert len(batch) <= n
+        if len(batch) < n:
+            break
+
+        start_index += left
+        s = lengths_cumsum[start_index - 1]
+
+        # add local rank
+        result.append(batch[rank])
+
+    return result, s, len(result) * c * n
+
+
+class MultipackBatchSampler(BatchSampler):
+    """
+    Batch Sampler class for multipack
+    """
+
+    def __init__(
+        self,
+        sampler: Union[Sampler[int], Iterable[int]],
+        batch_size: int,
+        drop_last: bool,
+        batch_max_len: int,
+        lengths: np.ndarray,
+        packing_efficiency_estimate: float = 1.0,
+    ):
+        super().__init__(sampler, batch_size, drop_last)
+        self.batch_size = None
+        self.batch_max_len = batch_max_len
+        self.lengths: np.ndarray = lengths
+        self.packing_efficiency_estimate = packing_efficiency_estimate or 1.0
+
+        assert isinstance(self.lengths, np.ndarray)
+
+        self.epoch = 0
+
+        # statistics
+        self.eff_total_used = 0
+        self.eff_total_slots = 0
+
+    def set_epoch(self, epoch: int):
+        self.epoch = epoch
+
+    def generate_batches(self, set_stats=False):
+        indices = [idx for idx in self.sampler]
+
+        lengths = self.lengths[indices]
+        lengths_cumsum = np.cumsum(lengths)
+
+        batches, total_used, total_slots = allocate(
+            lengths=lengths,
+            lengths_cumsum=lengths_cumsum,
+            rank=0,
+            c=self.batch_max_len,
+            n=1,
+        )
+
+        batches = [[indices[b_idx] for b_idx in batch] for batch in batches]
+
+        # statistics
+        if set_stats:
+            self.eff_total_used += total_used
+            self.eff_total_slots += total_slots
+
+        return batches
+
+    def __iter__(self):
+        batches = self.generate_batches(set_stats=True)
+        return iter(batches)
+
+    def num_batches(self):
+        batches = self.generate_batches(set_stats=True)
+        return len(batches)
+
+    def efficiency(self):
+        return self.eff_total_used / self.eff_total_slots
+
+    def __len__(self):
+        self.num_batches()
+        return self._len_est()
+
+    def _len_est(self):
+        world_size = int(os.getenv("WORLD_SIZE", "1"))
+        lengths_sum = np.sum(self.lengths)
+        lengths_sum_per_device = lengths_sum // world_size
+        LOG.info(
+            f"packing_efficiency_estimate: {self.packing_efficiency_estimate} "
+            f"total_num_tokens per device: {lengths_sum_per_device}"
+        )
+
+        # shave off 1% + 1 for dealing with variance in packing from random sampler to sampler
+        return (
+            world_size
+            * math.floor(
+                0.99
+                * lengths_sum_per_device
+                / self.packing_efficiency_estimate
+                // self.batch_max_len
+            )
+            - 1
+        )

src/axolotl/utils/tokenization.py

@@ -31,7 +31,8 @@ def check_example_labels(example, tokenizer, text_only=False):
         )
         colored_tokens.append(colored_token)
 
-    LOG.info(" ".join(colored_tokens))
+    delimiter = "" if text_only else " "
+    LOG.info(delimiter.join(colored_tokens))
     LOG.info("\n\n\n")
 
     return " ".join(colored_tokens)

src/axolotl/utils/trainer.py

@@ -1,50 +1,22 @@
 """Module containing the Trainer class and related functions"""
-import importlib
-import logging
 import math
 import os
-import sys
 from contextlib import contextmanager
-from dataclasses import dataclass, field
 from functools import partial
-from pathlib import Path
-from typing import List, Optional, Union
+from typing import List
 
 import numpy as np
 import torch
 import torch.cuda
-import torch.distributed as dist
-import transformers
-from datasets import Dataset, set_caching_enabled
-from torch.optim.lr_scheduler import OneCycleLR
-from torch.utils.data import (
-    DataLoader,
-    DistributedSampler,
-    RandomSampler,
-    SequentialSampler,
-)
-from transformers import EarlyStoppingCallback, Trainer, TrainingArguments
-from transformers.trainer_pt_utils import SequentialDistributedSampler
+from accelerate.logging import get_logger
+from datasets import set_caching_enabled
+from torch.utils.data import DataLoader, RandomSampler
 
-from axolotl.monkeypatch.relora import ReLoRACallback, ReLoRAScheduler
-from axolotl.utils.callbacks import (
-    GPUStatsCallback,
-    SaveBetterTransformerModelCallback,
-    SavePeftModelCallback,
-    bench_eval_callback_factory,
-    log_prediction_callback_factory,
-)
-from axolotl.utils.collators import DataCollatorForSeq2Seq
-from axolotl.utils.dataloader import MultipackDistributedDataloader
-from axolotl.utils.distributed import (
-    is_distributed,
-    is_main_process,
-    reduce_and_broadcast,
-    zero_first,
-)
-from axolotl.utils.schedulers import get_cosine_schedule_with_quadratic_warmup
+from axolotl.core.trainer_builder import HFCausalTrainerBuilder
+from axolotl.utils.distributed import is_main_process, reduce_and_broadcast, zero_first
+from axolotl.utils.samplers import MultipackBatchSampler
 
-LOG = logging.getLogger("axolotl")
+LOG = get_logger("axolotl")
 
 
 @torch.jit.script
@@ -109,269 +81,6 @@ def trainer_weighted_loss(model_output, labels, shift_labels=True):
     return weighted_cross_entropy(logits, labels, weights)
 
 
-@dataclass
-class AxolotlTrainingArguments(TrainingArguments):
-    """
-    Extend the base TrainingArguments for axolotl helpers
-    """
-
-    lr_quadratic_warmup: bool = field(
-        default=False,
-        metadata={"help": "Use quadratic warmup for cosine scheduling."},
-    )
-    sample_packing: bool = field(
-        default=False,
-        metadata={"help": "Use sample packing for efficient training."},
-    )
-    eval_sample_packing: Optional[bool] = field(
-        default=None,
-        metadata={"help": "Use sample packing for efficient evals."},
-    )
-    sample_packing_efficiency: float = field(
-        default=1.0,
-        metadata={"help": "Sample packing efficiency for calculating batch length."},
-    )
-    max_seq_length: int = field(
-        default=2048,
-        metadata={"help": "The maximum sequence length the model can handle"},
-    )
-    sample_packing_seq_len_multiplier: int = field(
-        default=1,
-        metadata={"help": "the multiplier for the max len for packed sequences"},
-    )
-    relora_steps: Optional[int] = field(
-        default=None,
-        metadata={"help": "how often to reset for ReLoRA"},
-    )
-    relora_warmup_steps: Optional[int] = field(
-        default=None,
-        metadata={"help": "how many warmup steps to take after reset for ReLoRA"},
-    )
-    bench_split: Optional[str] = field(
-        default="eval", metadata={"help": "The benchmark split to run on"}
-    )
-    bench_dataset: Optional[str] = field(
-        default="pharaouk/dharma-1/dharma_1_mini.json",
-        metadata={
-            "help": "Benchmark dataset to use: options are `mmlu-zs`, `mmlu-fs`, or the full path to the dataset file"
-        },
-    )
-    do_bench_eval: Optional[bool] = field(
-        default=False, metadata={"help": "Whether to run the Benchmark evaluation."}
-    )
-    max_bench_samples: Optional[int] = field(
-        default=None,
-        metadata={
-            "help": "If set, only evaluates on `max_bench_samples` of the benchmark dataset."
-        },
-    )
-    bench_source_max_len: int = field(
-        default=2048, metadata={"help": "Maximum source sequence length for bench."}
-    )
-
-
-class AxolotlTrainer(Trainer):
-    """
-    Extend the base Trainer for axolotl helpers
-    """
-
-    args = None  # type: AxolotlTrainingArguments
-
-    def __init__(self, *args, bench_data_collator=None, **kwargs):
-        self.bench_data_collator = bench_data_collator
-        super().__init__(*args, **kwargs)
-
-    def create_scheduler(
-        self, num_training_steps: int, optimizer: torch.optim.Optimizer = None
-    ):
-        """
-        Setup the scheduler. The optimizer of the trainer must have been set up either before this method is called or
-        passed as an argument.
-
-        Args:
-            num_training_steps (int): The number of training steps to do.
-            optimizer (torch.optim.Optimizer): The training optimizer
-        """
-
-        # fmt: off
-        if self.lr_scheduler is None:  # type: ignore  # pylint: disable=access-member-before-definition
-            # fmt: on
-            if (
-                self.args.lr_scheduler_type == "cosine"
-                and self.args.lr_quadratic_warmup is True
-            ):
-                self.lr_scheduler = get_cosine_schedule_with_quadratic_warmup(  # pylint: disable=attribute-defined-outside-init
-                    optimizer,
-                    num_warmup_steps=self.args.get_warmup_steps(num_training_steps),
-                    num_training_steps=num_training_steps,
-                )
-            else:
-                return super().create_scheduler(num_training_steps, optimizer)
-        return self.lr_scheduler
-
-    def _get_train_sampler(self) -> Optional[torch.utils.data.Sampler]:
-        if self.args.world_size > 1 and self.args.sample_packing:
-            return DistributedSampler(
-                self.train_dataset,
-                num_replicas=self.args.world_size,
-                rank=self.args.process_index,
-                seed=self.args.seed,
-            )
-        return super()._get_train_sampler()
-
-    def _get_eval_sampler(
-        self, eval_dataset: Dataset
-    ) -> Optional[torch.utils.data.Sampler]:
-        if (
-            self.args.world_size > 1
-            and self.args.sample_packing
-            and self.args.eval_sample_packing is not False
-        ):
-            return SequentialDistributedSampler(
-                eval_dataset,
-                num_replicas=self.args.world_size,
-                rank=self.args.process_index,
-                batch_size=self.args.per_device_eval_batch_size,
-            )
-        return super()._get_eval_sampler(eval_dataset)
-
-    def get_train_dataloader(self) -> Union[DataLoader, MultipackDistributedDataloader]:
-        if self.args.sample_packing:
-            train_sampler = self._get_train_sampler()
-            return self.accelerator.prepare(
-                MultipackDistributedDataloader(
-                    self.train_dataset,
-                    batch_size=self._train_batch_size,
-                    seq_max_length=self.args.max_seq_length,
-                    collate_fn=self.data_collator,
-                    sampler=train_sampler,
-                    packing_efficiency_estimate=self.args.sample_packing_efficiency,
-                    sample_packing_seq_len_multiplier=self.args.sample_packing_seq_len_multiplier,
-                    device_count=int(os.environ.get("WORLD_SIZE", 1)),
-                )
-            )
-        return super().get_train_dataloader()
-
-    def get_eval_dataloader(
-        self, eval_dataset: Optional[Dataset] = None
-    ) -> Union[DataLoader, MultipackDistributedDataloader]:
-        if self.args.sample_packing and self.args.eval_sample_packing is not False:
-            eval_dataset = (
-                eval_dataset if eval_dataset is not None else self.eval_dataset
-            )
-
-            eval_sampler = self._get_eval_sampler(eval_dataset)
-            return self.accelerator.prepare(
-                MultipackDistributedDataloader(
-                    eval_dataset,
-                    batch_size=self.args.eval_batch_size,
-                    seq_max_length=self.args.max_seq_length,
-                    collate_fn=self.data_collator,
-                    sampler=eval_sampler,
-                    packing_efficiency_estimate=self.args.sample_packing_efficiency,
-                    sample_packing_seq_len_multiplier=self.args.eval_batch_size,
-                    device_count=int(os.environ.get("WORLD_SIZE", 1)),
-                )
-            )
-        return super().get_eval_dataloader(eval_dataset)
-
-    def _get_bench_sampler(
-        self, bench_dataset: Dataset
-    ) -> Optional[torch.utils.data.Sampler]:
-        if self.args.world_size <= 1:
-            return SequentialSampler(bench_dataset)
-        return None
-
-    def get_bench_dataloader(
-        self,
-        bench_dataset: Dataset,
-    ) -> Union[DataLoader, MultipackDistributedDataloader]:
-        dataloader_params = {
-            "batch_size": self.args.eval_batch_size,
-            "collate_fn": self.bench_data_collator,
-            "num_workers": self.args.dataloader_num_workers,
-            "pin_memory": self.args.dataloader_pin_memory,
-        }
-
-        if not isinstance(bench_dataset, torch.utils.data.IterableDataset):
-            dataloader_params["sampler"] = self._get_bench_sampler(bench_dataset)
-            dataloader_params["drop_last"] = self.args.dataloader_drop_last
-
-        return DataLoader(bench_dataset, **dataloader_params)
-        # return self.accelerator.prepare(DataLoader(bench_dataset, **dataloader_params))
-
-    def compute_loss(self, model, inputs, return_outputs=False):
-        # use one's weighted cross entropy loss calc
-        # if self.args.sample_packing:
-        #     labels = inputs.pop("labels")
-        #     outputs = model(**inputs)
-        #     loss = trainer_weighted_loss(outputs, labels, shift_labels=True)
-        #     return (loss, outputs) if return_outputs else loss
-        return super().compute_loss(model, inputs, return_outputs=return_outputs)
-
-
-class OneCycleLRSchedulerTrainer(AxolotlTrainer):
-    """
-    Trainer subclass that uses the OneCycleLR scheduler
-    """
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        self.lr_scheduler = None
-
-    def create_scheduler(
-        self,
-        num_training_steps: int,
-        optimizer: Optional[torch.optim.Optimizer] = None,
-    ):
-        optimizer = self.optimizer if optimizer is None else optimizer
-        num_warmup_steps = self.args.get_warmup_steps(num_training_steps)
-        pct_start = num_warmup_steps / num_training_steps
-
-        self.lr_scheduler = OneCycleLR(
-            optimizer,
-            max_lr=self.args.learning_rate,
-            total_steps=num_training_steps,
-            pct_start=pct_start,
-            div_factor=6,
-        )
-
-        return self.lr_scheduler
-
-
-class ReLoRATrainer(AxolotlTrainer):
-    """
-    Trainer subclass that uses the OneCycleLR scheduler
-    """
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        self.lr_scheduler = None
-
-    def create_scheduler(
-        self,
-        num_training_steps: int,
-        optimizer: Optional[torch.optim.Optimizer] = None,
-    ):
-        optimizer = self.optimizer if optimizer is None else optimizer
-        lr_scheduler = super().create_scheduler(num_training_steps, optimizer)
-
-        if self.args.relora_steps:
-            warmup_steps = (
-                self.args.relora_warmup_steps if self.args.relora_warmup_steps else 10
-            )
-            self.lr_scheduler = ReLoRAScheduler(
-                optimizer,
-                lr_scheduler,
-                self.args.relora_steps,
-                warmup_steps,
-            )
-        else:
-            self.lr_scheduler = lr_scheduler
-
-        return self.lr_scheduler
-
-
 def add_position_ids(sample):
     sample_len = len(sample["input_ids"])
     sample["position_ids"] = torch.arange(len(sample["input_ids"]))
@@ -397,39 +106,53 @@ def disable_datasets_caching():
         set_caching_enabled(True)
 
 
-def process_datasets_for_packing(cfg, train_dataset, eval_dataset):
+def process_datasets_for_packing(cfg, train_dataset, eval_dataset, tokenizer):
     drop_long = partial(drop_long_seq, sequence_len=cfg.sequence_len)
     with zero_first(is_main_process()):
-        train_dataset = train_dataset.filter(drop_long, num_proc=os.cpu_count())
+        train_dataset = train_dataset.filter(drop_long, num_proc=cfg.dataset_processes)
         if eval_dataset:
-            eval_dataset = eval_dataset.filter(drop_long, num_proc=os.cpu_count())
+            eval_dataset = eval_dataset.filter(
+                drop_long, num_proc=cfg.dataset_processes
+            )
 
         if cfg.group_by_length:
-            train_dataset = train_dataset.map(add_length, num_proc=os.cpu_count())
+            train_dataset = train_dataset.map(
+                add_length, num_proc=cfg.dataset_processes
+            )
 
         if cfg.sample_packing:
-            train_dataset = train_dataset.map(add_position_ids, num_proc=os.cpu_count())
+            train_dataset = train_dataset.map(
+                add_position_ids, num_proc=cfg.dataset_processes
+            )
             if cfg.eval_sample_packing is not False:
                 if eval_dataset:
                     eval_dataset = eval_dataset.map(
-                        add_position_ids, num_proc=os.cpu_count()
+                        add_position_ids, num_proc=cfg.dataset_processes
                     )
+
+        # Phi doesn't want the attention_mask feature when training
+        if "CodeGenTokenizer" in tokenizer.__class__.__name__ or (
+            cfg.is_mistral_derived_model and cfg.flash_attention
+        ):
+            train_dataset = train_dataset.remove_columns("attention_mask")
+            if eval_dataset:
+                eval_dataset = eval_dataset.remove_columns("attention_mask")
+
     return train_dataset, eval_dataset
 
 
-def calculate_total_num_steps(cfg, train_dataset, tokenizer):
+def calculate_total_num_steps(cfg, train_dataset):
     if cfg.sample_packing:
         # we have to drop anything longer then sequence len otherwise
         # flash attention with position ids fails
         if not cfg.total_num_tokens:
-            LOG.info("calculating total_num_tokens")
             total_num_tokens = np.sum(
                 train_dataset.data.column("input_ids")
                 .to_pandas()
                 .apply(lambda x: len(x))  # pylint: disable=unnecessary-lambda
                 .values
             )
-            LOG.info(f"total_num_tokens: {total_num_tokens}")
+            LOG.debug(f"total_num_tokens: {total_num_tokens}", main_process_only=True)
             cfg.total_num_tokens = total_num_tokens
 
         if not cfg.total_supervised_tokens:
@@ -439,7 +162,10 @@ def calculate_total_num_steps(cfg, train_dataset, tokenizer):
                 .apply(lambda x: np.sum(np.array(x) != -100))
                 .sum()
             )
-            LOG.info(f"`total_supervised_tokens: {total_supervised_tokens}`")
+            LOG.debug(
+                f"`total_supervised_tokens: {total_supervised_tokens}`",
+                main_process_only=True,
+            )
             cfg.total_supervised_tokens = total_supervised_tokens
 
         if cfg.sample_packing_eff_est:
@@ -458,40 +184,41 @@ def calculate_total_num_steps(cfg, train_dataset, tokenizer):
                 )
                 * cfg.num_epochs
             )
-            LOG.info(
-                f"total_num_tokens: {cfg.total_num_tokens}, total_num_steps: {total_num_steps}"
+            LOG.debug(
+                f"total_num_tokens: {cfg.total_num_tokens}, total_num_steps: {total_num_steps}",
+                main_process_only=True,
             )
         else:
-            if cfg.world_size > 1 and is_distributed():
-                sampler = DistributedSampler(
-                    train_dataset,
-                    num_replicas=cfg.world_size,
-                    rank=dist.get_rank(),
-                    seed=cfg.seed or 42,
-                )
-            else:
-                sampler = RandomSampler(train_dataset)
-
-            data_loader = MultipackDistributedDataloader(
-                train_dataset,
+            sampler = MultipackBatchSampler(
+                sampler=RandomSampler(train_dataset),
                 batch_size=cfg.micro_batch_size,
-                seq_max_length=cfg.max_packed_sequence_len or cfg.sequence_len,
-                collate_fn=DataCollatorForSeq2Seq(
-                    tokenizer,
-                    return_tensors="pt",
-                    padding="longest",
+                drop_last=True,
+                batch_max_len=cfg.micro_batch_size
+                * (cfg.max_packed_sequence_len or cfg.sequence_len),
+                lengths=(
+                    train_dataset.data.column("position_ids")
+                    .to_pandas()
+                    .apply(lambda x: x[-1] + 1)
+                    .values
                 ),
-                sampler=sampler,
-                packing_efficiency_estimate=cfg.sample_packing_eff_est,
-                sample_packing_seq_len_multiplier=cfg.micro_batch_size,
-                device_count=int(os.environ.get("WORLD_SIZE", 1)),
             )
-            data_loader_len = data_loader.len_w_stats()
-            actual_eff = data_loader.efficiency()
-            LOG.info(f"data_loader_len: {data_loader_len}")
+
+            data_loader = DataLoader(
+                train_dataset.remove_columns(["length"]),
+                batch_sampler=sampler,
+            )
+            data_loader_len = len(data_loader)
+            actual_eff = sampler.efficiency()
+            LOG.debug(f"data_loader_len: {data_loader_len}", main_process_only=True)
             # FIXME: is there a bug here somewhere? the total num steps depends
             # on the agreed on value for sample_packing_eff_est
-            total_num_steps = int(math.floor(data_loader_len * cfg.num_epochs))
+            total_num_steps = int(
+                math.floor(
+                    data_loader_len
+                    * cfg.num_epochs
+                    / int(os.environ.get("WORLD_SIZE", 1))
+                )
+            )
 
             def calc_sample_packing_eff_est(estimates: List[float]):
                 LOG.info(f"sample_packing_eff_est across ranks: {repr(estimates)}")
@@ -505,12 +232,20 @@ def calculate_total_num_steps(cfg, train_dataset, tokenizer):
                 math.ceil(sample_packing_actual_eff_all * 100.0) / 100.0
             )
             cfg.sample_packing_eff_est = sample_packing_eff_est
-            LOG.info(f"sample_packing_eff_est: {cfg.sample_packing_eff_est}")
+            LOG.debug(
+                f"sample_packing_eff_est: {cfg.sample_packing_eff_est}",
+                main_process_only=True,
+            )
     else:
         total_num_steps = int(
-            math.ceil(len(train_dataset) * cfg.num_epochs / cfg.batch_size)
+            math.ceil(
+                len(train_dataset)
+                * cfg.num_epochs
+                / int(os.environ.get("WORLD_SIZE", 1))
+                / cfg.batch_size
+            )
         )
-    LOG.info(f"total_num_steps: {total_num_steps}")
+    LOG.debug(f"total_num_steps: {total_num_steps}", main_process_only=True)
     return total_num_steps
 
 
@@ -533,256 +268,11 @@ def setup_trainer(cfg, train_dataset, eval_dataset, model, tokenizer, total_num_
         setup_fsdp_envs(cfg)
     elif cfg.deepspeed:
         os.environ["ACCELERATE_USE_DEEPSPEED"] = "true"
+    if cfg.fp8:
+        os.environ["ACCELERATE_MIXED_PRECISION"] = "fp8"
 
-    warmup_steps = (
-        cfg.warmup_steps
-        if cfg.warmup_steps is not None
-        else min(int(0.03 * total_num_steps), 100)
-    )
-    logging_steps = (
-        cfg.logging_steps
-        if cfg.logging_steps is not None
-        else max(min(int(0.005 * total_num_steps), 10), 1)
-    )
+    trainer_builder = HFCausalTrainerBuilder(cfg, model, tokenizer)
+    trainer_builder.train_dataset = train_dataset
+    trainer_builder.eval_dataset = eval_dataset
 
-    training_arguments_kwargs = {}
-    if cfg.bf16 == "full":
-        training_arguments_kwargs["bf16_full_eval"] = True
-    else:
-        training_arguments_kwargs["bf16"] = cfg.bf16
-    training_arguments_kwargs["fp16"] = (cfg.fp16 and not cfg.bf16) or False
-    training_arguments_kwargs["tf32"] = cfg.tf32
-    training_arguments_kwargs["warmup_steps"] = warmup_steps
-    training_arguments_kwargs["logging_steps"] = logging_steps
-
-    if cfg.seed:
-        training_arguments_kwargs["seed"] = cfg.seed
-
-    if cfg.gradient_checkpointing:
-        training_arguments_kwargs["gradient_checkpointing"] = cfg.gradient_checkpointing
-    if cfg.fsdp:
-        training_arguments_kwargs["fsdp"] = cfg.fsdp
-        if cfg.fsdp_config:
-            training_arguments_kwargs["fsdp_config"] = dict(cfg.fsdp_config)
-
-    # deepspeed
-    if cfg.deepspeed:
-        training_arguments_kwargs["deepspeed"] = cfg.deepspeed
-
-    if cfg.lr_quadratic_warmup is not None:
-        training_arguments_kwargs["lr_quadratic_warmup"] = cfg.lr_quadratic_warmup
-
-    if cfg.adam_beta1:
-        training_arguments_kwargs["adam_beta1"] = cfg.adam_beta1
-    if cfg.adam_beta2:
-        training_arguments_kwargs["adam_beta2"] = cfg.adam_beta2
-    if cfg.adam_epsilon:
-        training_arguments_kwargs["adam_epsilon"] = cfg.adam_epsilon
-    if cfg.max_grad_norm:
-        training_arguments_kwargs["max_grad_norm"] = cfg.max_grad_norm
-
-    if cfg.hub_model_id:
-        training_arguments_kwargs["hub_model_id"] = cfg.hub_model_id
-        training_arguments_kwargs["push_to_hub"] = True
-        training_arguments_kwargs["hub_private_repo"] = True
-
-        if cfg.hub_strategy:
-            training_arguments_kwargs["hub_strategy"] = cfg.hub_strategy
-
-    if cfg.save_safetensors:
-        training_arguments_kwargs["save_safetensors"] = cfg.save_safetensors
-
-    if cfg.sample_packing_eff_est:
-        training_arguments_kwargs[
-            "sample_packing_efficiency"
-        ] = cfg.sample_packing_eff_est
-
-    if cfg.eval_steps and cfg.evaluation_strategy:
-        # assume if the user set both, they know what they're doing
-        training_arguments_kwargs["evaluation_strategy"] = cfg.evaluation_strategy
-        training_arguments_kwargs["eval_steps"] = cfg.eval_steps
-    elif cfg.val_set_size == 0:
-        # no eval set, so don't eval
-        training_arguments_kwargs["evaluation_strategy"] = "no"
-    elif cfg.evaluation_strategy and cfg.evaluation_strategy in ["epoch", "no"]:
-        # if explicitly set for epoch, just set, and eval steps don't matter
-        training_arguments_kwargs["evaluation_strategy"] = cfg.evaluation_strategy
-    elif cfg.eval_steps:
-        # steps isn't used w/ epochs
-        training_arguments_kwargs["evaluation_strategy"] = "steps"
-        training_arguments_kwargs["eval_steps"] = cfg.eval_steps
-    else:
-        # we have an eval set, but no steps defined, default to use epoch
-        training_arguments_kwargs["evaluation_strategy"] = "epoch"
-
-    if cfg.save_steps:
-        # save_steps implies save_strategy of steps
-        training_arguments_kwargs["save_strategy"] = "steps"
-        training_arguments_kwargs["save_steps"] = cfg.save_steps
-    elif cfg.save_strategy:
-        training_arguments_kwargs["save_strategy"] = cfg.save_strategy
-    else:
-        # default to saving each epoch if not defined
-        training_arguments_kwargs["save_strategy"] = "epoch"
-
-    if cfg.do_bench_eval:
-        training_arguments_kwargs["do_bench_eval"] = cfg.do_bench_eval
-        if cfg.bench_dataset:
-            training_arguments_kwargs["bench_dataset"] = cfg.bench_dataset
-    if cfg.metric_for_best_model:
-        training_arguments_kwargs["metric_for_best_model"] = cfg.metric_for_best_model
-    if cfg.greater_is_better:
-        training_arguments_kwargs["greater_is_better"] = cfg.greater_is_better
-
-    if cfg.torch_compile:
-        if torch.__version__ < "2.1.0":  # pylint: disable=protected-access
-            LOG.warning("torch>=2.1.0 required for torch_compile to work properly")
-        else:
-            import torch._dynamo  # pylint: disable=redefined-outer-name
-
-            torch._dynamo.config.suppress_errors = (  # pylint: disable=protected-access
-                True
-            )
-            training_arguments_kwargs["torch_compile"] = cfg.torch_compile
-            if cfg.torch_compile_backend:
-                training_arguments_kwargs[
-                    "torch_compile_backend"
-                ] = cfg.torch_compile_backend
-
-    # DDP Config
-    if cfg.ddp_timeout:
-        training_arguments_kwargs["ddp_timeout"] = cfg.ddp_timeout
-    # see https://pytorch.org/docs/stable/generated/torch.nn.parallel.DistributedDataParallel.html
-    if cfg.ddp_bucket_cap_mb:
-        training_arguments_kwargs["ddp_bucket_cap_mb"] = cfg.ddp_bucket_cap_mb
-    if cfg.ddp_broadcast_buffers is not None:
-        training_arguments_kwargs["ddp_broadcast_buffers"] = cfg.ddp_broadcast_buffers
-
-    training_args = AxolotlTrainingArguments(  # pylint: disable=unexpected-keyword-arg
-        max_steps=total_num_steps if cfg.max_steps else -1,
-        max_seq_length=cfg.sequence_len,
-        per_device_train_batch_size=cfg.micro_batch_size,
-        per_device_eval_batch_size=cfg.eval_batch_size
-        if cfg.eval_batch_size is not None
-        else cfg.micro_batch_size,
-        gradient_accumulation_steps=cfg.gradient_accumulation_steps,
-        eval_accumulation_steps=cfg.gradient_accumulation_steps,
-        num_train_epochs=cfg.num_epochs,
-        learning_rate=cfg.learning_rate,
-        output_dir=cfg.output_dir,
-        save_total_limit=cfg.save_total_limit if cfg.save_total_limit else 4,
-        load_best_model_at_end=(
-            (cfg.load_best_model_at_end is not False or cfg.early_stopping_patience)
-            and cfg.val_set_size > 0
-            and cfg.save_steps
-            and cfg.save_steps % cfg.eval_steps == 0
-        )
-        or False,
-        ddp_find_unused_parameters=False if cfg.ddp else None,
-        group_by_length=cfg.group_by_length,
-        report_to="wandb" if cfg.use_wandb else None,
-        run_name=cfg.wandb_run_id if cfg.use_wandb else None,
-        optim=cfg.optimizer if cfg.optimizer else "adamw_hf",
-        lr_scheduler_type=cfg.lr_scheduler
-        if cfg.lr_scheduler and cfg.lr_scheduler not in ("one_cycle", "log_sweep")
-        else "cosine",
-        weight_decay=cfg.weight_decay if cfg.weight_decay is not None else 0.0,
-        sample_packing=cfg.sample_packing if cfg.sample_packing else False,
-        eval_sample_packing=cfg.eval_sample_packing,
-        sample_packing_seq_len_multiplier=cfg.micro_batch_size,
-        relora_steps=cfg.relora_steps,
-        relora_warmup_steps=cfg.relora_warmup_steps,
-        **training_arguments_kwargs,
-    )
-
-    trainer_kwargs = {}
-
-    if cfg.optimizer == "adamw_anyprecision":
-        if Path(cfg.torchdistx_path).exists():
-            sys.path.append(cfg.torchdistx_path)
-            importlib.import_module("torchdistx")
-
-    callbacks = []
-    callbacks.append(GPUStatsCallback(cfg))
-
-    if cfg.relora_steps:
-        callbacks.append(ReLoRACallback(cfg))
-
-    if cfg.local_rank == 0 and cfg.adapter in [
-        "lora",
-        "qlora",
-    ]:  # only save in rank 0
-        callbacks.append(SavePeftModelCallback)
-
-    if hasattr(model, "use_bettertransformer") and model.use_bettertransformer is True:
-        callbacks.append(SaveBetterTransformerModelCallback)
-
-    data_collator_kwargs = {
-        "padding": True,  # True/"longest" is the default
-    }
-    if cfg.pad_to_sequence_len:
-        data_collator_kwargs["pad_to_multiple_of"] = 64 * math.ceil(
-            cfg.sequence_len / 64
-        )
-    else:
-        # A100 is best at 64, while others at 8. Let's use the larger so we don't have to check
-        # https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html
-        data_collator_kwargs["pad_to_multiple_of"] = 64
-
-    if cfg.is_llama_derived_model and cfg.landmark_attention:
-        from axolotl.monkeypatch.llama_landmark_attn import (
-            add_mem_tokens,
-            get_mem_id,
-            set_model_mem_id,
-        )
-
-        set_model_mem_id(model, tokenizer)
-
-        LOG.info("Adding landmark attention tokens to dataset")
-
-        for dataset in [train_dataset, eval_dataset]:
-            dataset = dataset.map(
-                partial(add_mem_tokens, mem_freq=50, mem_id=get_mem_id(tokenizer)),
-                batched=False,
-                num_proc=32,
-            )
-
-    trainer_cls = AxolotlTrainer
-    if cfg.lr_scheduler == "one_cycle" and (cfg.fsdp or cfg.adapter == "qlora"):
-        trainer_cls = OneCycleLRSchedulerTrainer
-    elif cfg.relora_steps:
-        trainer_cls = ReLoRATrainer
-    trainer = trainer_cls(
-        model=model,
-        train_dataset=train_dataset,
-        eval_dataset=eval_dataset,
-        args=training_args,
-        data_collator=DataCollatorForSeq2Seq(
-            tokenizer,
-            return_tensors="pt",
-            **data_collator_kwargs,
-        ),
-        bench_data_collator=transformers.DataCollatorForSeq2Seq(
-            tokenizer,
-            return_tensors="pt",
-            **data_collator_kwargs,
-        ),
-        callbacks=callbacks,
-        **trainer_kwargs,
-    )
-
-    if cfg.use_wandb and cfg.eval_table_size > 0:
-        LogPredictionCallback = log_prediction_callback_factory(trainer, tokenizer)
-        trainer.add_callback(LogPredictionCallback(cfg))
-
-    if cfg.do_bench_eval:
-        trainer.add_callback(bench_eval_callback_factory(trainer, tokenizer))
-
-    # TODO on_save callback to sync checkpoints to GCP/AWS in background
-    if cfg.early_stopping_patience:
-        early_stop_cb = EarlyStoppingCallback(
-            cfg.early_stopping_patience,
-        )
-        trainer.add_callback(early_stop_cb)
-
-    return trainer
+    return trainer_builder.build(total_num_steps)

tests/e2e/test_fused_llama.py

@@ -0,0 +1,73 @@
+"""
+E2E tests for lora llama
+"""
+
+import logging
+import os
+import unittest
+from pathlib import Path
+
+from transformers.utils import is_torch_bf16_gpu_available
+
+from axolotl.cli import load_datasets
+from axolotl.common.cli import TrainerCliArgs
+from axolotl.train import train
+from axolotl.utils.config import normalize_config
+from axolotl.utils.dict import DictDefault
+
+from .utils import with_temp_dir
+
+LOG = logging.getLogger("axolotl.tests.e2e")
+os.environ["WANDB_DISABLED"] = "true"
+
+
+class TestFusedLlama(unittest.TestCase):
+    """
+    Test case for Llama models using Fused layers
+    """
+
+    @with_temp_dir
+    def test_fft_packing(self, temp_dir):
+        # pylint: disable=duplicate-code
+        cfg = DictDefault(
+            {
+                "base_model": "JackFram/llama-68m",
+                "flash_attention": True,
+                "flash_attn_fuse_qkv": True,
+                "flash_attn_fuse_mlp": True,
+                "sample_packing": True,
+                "sequence_len": 1024,
+                "val_set_size": 0.1,
+                "special_tokens": {
+                    "unk_token": "<unk>",
+                    "bos_token": "<s>",
+                    "eos_token": "</s>",
+                },
+                "datasets": [
+                    {
+                        "path": "mhenrichsen/alpaca_2k_test",
+                        "type": "alpaca",
+                    },
+                ],
+                "num_epochs": 2,
+                "micro_batch_size": 2,
+                "gradient_accumulation_steps": 1,
+                "output_dir": temp_dir,
+                "learning_rate": 0.00001,
+                "optimizer": "adamw_torch",
+                "lr_scheduler": "cosine",
+                "max_steps": 20,
+                "save_steps": 10,
+                "eval_steps": 10,
+            }
+        )
+        if is_torch_bf16_gpu_available():
+            cfg.bf16 = True
+        else:
+            cfg.fp16 = True
+        normalize_config(cfg)
+        cli_args = TrainerCliArgs()
+        dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
+
+        train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
+        assert (Path(temp_dir) / "pytorch_model.bin").exists()

tests/e2e/test_lora_llama.py

@@ -4,8 +4,8 @@ E2E tests for lora llama
 
 import logging
 import os
-import tempfile
 import unittest
+from pathlib import Path
 
 from axolotl.cli import load_datasets
 from axolotl.common.cli import TrainerCliArgs
@@ -13,6 +13,8 @@ from axolotl.train import train
 from axolotl.utils.config import normalize_config
 from axolotl.utils.dict import DictDefault
 
+from .utils import with_temp_dir
+
 LOG = logging.getLogger("axolotl.tests.e2e")
 os.environ["WANDB_DISABLED"] = "true"
 
@@ -22,12 +24,12 @@ class TestLoraLlama(unittest.TestCase):
     Test case for Llama models using LoRA
     """
 
-    def test_lora(self):
+    @with_temp_dir
+    def test_lora(self, temp_dir):
         # pylint: disable=duplicate-code
         cfg = DictDefault(
             {
                 "base_model": "JackFram/llama-68m",
-                "base_model_config": "JackFram/llama-68m",
                 "tokenizer_type": "LlamaTokenizer",
                 "sequence_len": 1024,
                 "load_in_8bit": True,
@@ -51,7 +53,7 @@ class TestLoraLlama(unittest.TestCase):
                 "num_epochs": 2,
                 "micro_batch_size": 8,
                 "gradient_accumulation_steps": 1,
-                "output_dir": tempfile.mkdtemp(),
+                "output_dir": temp_dir,
                 "learning_rate": 0.00001,
                 "optimizer": "adamw_torch",
                 "lr_scheduler": "cosine",
@@ -62,13 +64,14 @@ class TestLoraLlama(unittest.TestCase):
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
 
         train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
+        assert (Path(temp_dir) / "adapter_model.bin").exists()
 
-    def test_lora_packing(self):
+    @with_temp_dir
+    def test_lora_packing(self, temp_dir):
         # pylint: disable=duplicate-code
         cfg = DictDefault(
             {
                 "base_model": "JackFram/llama-68m",
-                "base_model_config": "JackFram/llama-68m",
                 "tokenizer_type": "LlamaTokenizer",
                 "sequence_len": 1024,
                 "sample_packing": True,
@@ -94,7 +97,7 @@ class TestLoraLlama(unittest.TestCase):
                 "num_epochs": 2,
                 "micro_batch_size": 8,
                 "gradient_accumulation_steps": 1,
-                "output_dir": tempfile.mkdtemp(),
+                "output_dir": temp_dir,
                 "learning_rate": 0.00001,
                 "optimizer": "adamw_torch",
                 "lr_scheduler": "cosine",
@@ -105,3 +108,52 @@ class TestLoraLlama(unittest.TestCase):
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
 
         train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
+        assert (Path(temp_dir) / "adapter_model.bin").exists()
+
+    @with_temp_dir
+    def test_lora_gptq(self, temp_dir):
+        # pylint: disable=duplicate-code
+        cfg = DictDefault(
+            {
+                "base_model": "TheBlokeAI/jackfram_llama-68m-GPTQ",
+                "model_type": "AutoModelForCausalLM",
+                "tokenizer_type": "LlamaTokenizer",
+                "sequence_len": 1024,
+                "sample_packing": True,
+                "flash_attention": True,
+                "load_in_8bit": True,
+                "adapter": "lora",
+                "gptq": True,
+                "gptq_disable_exllama": True,
+                "lora_r": 32,
+                "lora_alpha": 64,
+                "lora_dropout": 0.05,
+                "lora_target_linear": True,
+                "val_set_size": 0.1,
+                "special_tokens": {
+                    "unk_token": "<unk>",
+                    "bos_token": "<s>",
+                    "eos_token": "</s>",
+                },
+                "datasets": [
+                    {
+                        "path": "mhenrichsen/alpaca_2k_test",
+                        "type": "alpaca",
+                    },
+                ],
+                "num_epochs": 2,
+                "save_steps": 0.5,
+                "micro_batch_size": 8,
+                "gradient_accumulation_steps": 1,
+                "output_dir": temp_dir,
+                "learning_rate": 0.00001,
+                "optimizer": "adamw_torch",
+                "lr_scheduler": "cosine",
+            }
+        )
+        normalize_config(cfg)
+        cli_args = TrainerCliArgs()
+        dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
+
+        train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
+        assert (Path(temp_dir) / "adapter_model.bin").exists()

tests/e2e/test_mistral.py

@@ -0,0 +1,115 @@
+"""
+E2E tests for lora llama
+"""
+
+import logging
+import os
+import unittest
+from pathlib import Path
+
+from transformers.utils import is_torch_bf16_gpu_available
+
+from axolotl.cli import load_datasets
+from axolotl.common.cli import TrainerCliArgs
+from axolotl.train import train
+from axolotl.utils.config import normalize_config
+from axolotl.utils.dict import DictDefault
+
+from .utils import with_temp_dir
+
+LOG = logging.getLogger("axolotl.tests.e2e")
+os.environ["WANDB_DISABLED"] = "true"
+
+
+class TestMistral(unittest.TestCase):
+    """
+    Test case for Llama models using LoRA
+    """
+
+    @with_temp_dir
+    def test_lora(self, temp_dir):
+        # pylint: disable=duplicate-code
+        cfg = DictDefault(
+            {
+                "base_model": "openaccess-ai-collective/tiny-mistral",
+                "flash_attention": True,
+                "sequence_len": 1024,
+                "load_in_8bit": True,
+                "adapter": "lora",
+                "lora_r": 32,
+                "lora_alpha": 64,
+                "lora_dropout": 0.05,
+                "lora_target_linear": True,
+                "val_set_size": 0.1,
+                "special_tokens": {
+                    "unk_token": "<unk>",
+                    "bos_token": "<s>",
+                    "eos_token": "</s>",
+                },
+                "datasets": [
+                    {
+                        "path": "mhenrichsen/alpaca_2k_test",
+                        "type": "alpaca",
+                    },
+                ],
+                "num_epochs": 2,
+                "micro_batch_size": 2,
+                "gradient_accumulation_steps": 1,
+                "output_dir": temp_dir,
+                "learning_rate": 0.00001,
+                "optimizer": "adamw_torch",
+                "lr_scheduler": "cosine",
+                "max_steps": 20,
+                "save_steps": 10,
+                "eval_steps": 10,
+            }
+        )
+        normalize_config(cfg)
+        cli_args = TrainerCliArgs()
+        dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
+
+        train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
+        assert (Path(temp_dir) / "adapter_model.bin").exists()
+
+    @with_temp_dir
+    def test_ft(self, temp_dir):
+        # pylint: disable=duplicate-code
+        cfg = DictDefault(
+            {
+                "base_model": "openaccess-ai-collective/tiny-mistral",
+                "flash_attention": True,
+                "sequence_len": 1024,
+                "val_set_size": 0.1,
+                "special_tokens": {
+                    "unk_token": "<unk>",
+                    "bos_token": "<s>",
+                    "eos_token": "</s>",
+                },
+                "datasets": [
+                    {
+                        "path": "mhenrichsen/alpaca_2k_test",
+                        "type": "alpaca",
+                    },
+                ],
+                "num_epochs": 2,
+                "micro_batch_size": 2,
+                "gradient_accumulation_steps": 1,
+                "output_dir": temp_dir,
+                "learning_rate": 0.00001,
+                "optimizer": "adamw_torch",
+                "lr_scheduler": "cosine",
+                "max_steps": 20,
+                "save_steps": 10,
+                "eval_steps": 10,
+            }
+        )
+        if is_torch_bf16_gpu_available():
+            cfg.bf16 = True
+        else:
+            cfg.fp16 = True
+        normalize_config(cfg)
+        cli_args = TrainerCliArgs()
+        dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
+
+        train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
+        assert (Path(temp_dir) / "pytorch_model.bin").exists()

tests/e2e/test_mistral_samplepack.py

@@ -0,0 +1,117 @@
+"""
+E2E tests for lora llama
+"""
+
+import logging
+import os
+import unittest
+from pathlib import Path
+
+from transformers.utils import is_torch_bf16_gpu_available
+
+from axolotl.cli import load_datasets
+from axolotl.common.cli import TrainerCliArgs
+from axolotl.train import train
+from axolotl.utils.config import normalize_config
+from axolotl.utils.dict import DictDefault
+
+from .utils import with_temp_dir
+
+LOG = logging.getLogger("axolotl.tests.e2e")
+os.environ["WANDB_DISABLED"] = "true"
+
+
+class TestMistral(unittest.TestCase):
+    """
+    Test case for Llama models using LoRA
+    """
+
+    @with_temp_dir
+    def test_lora_packing(self, temp_dir):
+        # pylint: disable=duplicate-code
+        cfg = DictDefault(
+            {
+                "base_model": "openaccess-ai-collective/tiny-mistral",
+                "flash_attention": True,
+                "sample_packing": True,
+                "sequence_len": 1024,
+                "load_in_8bit": True,
+                "adapter": "lora",
+                "lora_r": 32,
+                "lora_alpha": 64,
+                "lora_dropout": 0.05,
+                "lora_target_linear": True,
+                "val_set_size": 0.1,
+                "special_tokens": {
+                    "unk_token": "<unk>",
+                    "bos_token": "<s>",
+                    "eos_token": "</s>",
+                },
+                "datasets": [
+                    {
+                        "path": "mhenrichsen/alpaca_2k_test",
+                        "type": "alpaca",
+                    },
+                ],
+                "num_epochs": 2,
+                "micro_batch_size": 2,
+                "gradient_accumulation_steps": 1,
+                "output_dir": temp_dir,
+                "learning_rate": 0.00001,
+                "optimizer": "adamw_torch",
+                "lr_scheduler": "cosine",
+                "max_steps": 20,
+                "save_steps": 10,
+                "eval_steps": 10,
+            }
+        )
+        normalize_config(cfg)
+        cli_args = TrainerCliArgs()
+        dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
+
+        train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
+        assert (Path(temp_dir) / "adapter_model.bin").exists()
+
+    @with_temp_dir
+    def test_ft_packing(self, temp_dir):
+        # pylint: disable=duplicate-code
+        cfg = DictDefault(
+            {
+                "base_model": "openaccess-ai-collective/tiny-mistral",
+                "flash_attention": True,
+                "sample_packing": True,
+                "sequence_len": 1024,
+                "val_set_size": 0.1,
+                "special_tokens": {
+                    "unk_token": "<unk>",
+                    "bos_token": "<s>",
+                    "eos_token": "</s>",
+                },
+                "datasets": [
+                    {
+                        "path": "mhenrichsen/alpaca_2k_test",
+                        "type": "alpaca",
+                    },
+                ],
+                "num_epochs": 2,
+                "micro_batch_size": 2,
+                "gradient_accumulation_steps": 1,
+                "output_dir": temp_dir,
+                "learning_rate": 0.00001,
+                "optimizer": "adamw_torch",
+                "lr_scheduler": "cosine",
+                "max_steps": 20,
+                "save_steps": 10,
+                "eval_steps": 10,
+            }
+        )
+        if is_torch_bf16_gpu_available():
+            cfg.bf16 = True
+        else:
+            cfg.fp16 = True
+        normalize_config(cfg)
+        cli_args = TrainerCliArgs()
+        dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
+
+        train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
+        assert (Path(temp_dir) / "pytorch_model.bin").exists()

tests/e2e/test_phi.py

@@ -4,8 +4,8 @@ E2E tests for lora llama
 
 import logging
 import os
-import tempfile
 import unittest
+from pathlib import Path
 
 from axolotl.cli import load_datasets
 from axolotl.common.cli import TrainerCliArgs
@@ -13,6 +13,8 @@ from axolotl.train import train
 from axolotl.utils.config import normalize_config
 from axolotl.utils.dict import DictDefault
 
+from .utils import with_temp_dir
+
 LOG = logging.getLogger("axolotl.tests.e2e")
 os.environ["WANDB_DISABLED"] = "true"
 
@@ -22,18 +24,18 @@ class TestPhi(unittest.TestCase):
     Test case for Llama models using LoRA
     """
 
-    def test_ft(self):
+    @with_temp_dir
+    def test_ft(self, temp_dir):
         # pylint: disable=duplicate-code
         cfg = DictDefault(
             {
                 "base_model": "microsoft/phi-1_5",
-                "base_model_config": "microsoft/phi-1_5",
                 "trust_remote_code": True,
                 "model_type": "MixFormerSequentialForCausalLM",
                 "tokenizer_type": "AutoTokenizer",
-                "sequence_len": 2048,
+                "sequence_len": 512,
                 "sample_packing": False,
-                "load_in_8bit": True,
+                "load_in_8bit": False,
                 "adapter": None,
                 "val_set_size": 0.1,
                 "special_tokens": {
@@ -53,10 +55,11 @@ class TestPhi(unittest.TestCase):
                 "num_epochs": 1,
                 "micro_batch_size": 1,
                 "gradient_accumulation_steps": 1,
-                "output_dir": tempfile.mkdtemp(),
+                "output_dir": temp_dir,
                 "learning_rate": 0.00001,
-                "optimizer": "adamw_torch",
+                "optimizer": "adamw_bnb_8bit",
                 "lr_scheduler": "cosine",
+                "bf16": True,
             }
         )
         normalize_config(cfg)
@@ -64,19 +67,20 @@ class TestPhi(unittest.TestCase):
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
 
         train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
+        assert (Path(temp_dir) / "pytorch_model.bin").exists()
 
-    def test_ft_packed(self):
+    @with_temp_dir
+    def test_ft_packed(self, temp_dir):
         # pylint: disable=duplicate-code
         cfg = DictDefault(
             {
                 "base_model": "microsoft/phi-1_5",
-                "base_model_config": "microsoft/phi-1_5",
                 "trust_remote_code": True,
                 "model_type": "MixFormerSequentialForCausalLM",
                 "tokenizer_type": "AutoTokenizer",
-                "sequence_len": 2048,
+                "sequence_len": 512,
                 "sample_packing": True,
-                "load_in_8bit": True,
+                "load_in_8bit": False,
                 "adapter": None,
                 "val_set_size": 0.1,
                 "special_tokens": {
@@ -96,10 +100,11 @@ class TestPhi(unittest.TestCase):
                 "num_epochs": 1,
                 "micro_batch_size": 1,
                 "gradient_accumulation_steps": 1,
-                "output_dir": tempfile.mkdtemp(),
+                "output_dir": temp_dir,
                 "learning_rate": 0.00001,
-                "optimizer": "adamw_torch",
+                "optimizer": "adamw_bnb_8bit",
                 "lr_scheduler": "cosine",
+                "bf16": True,
             }
         )
         normalize_config(cfg)
@@ -107,3 +112,4 @@ class TestPhi(unittest.TestCase):
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
 
         train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
+        assert (Path(temp_dir) / "pytorch_model.bin").exists()

tests/e2e/utils.py

@@ -0,0 +1,22 @@
+"""
+helper utils for tests
+"""
+
+import shutil
+import tempfile
+from functools import wraps
+
+
+def with_temp_dir(test_func):
+    @wraps(test_func)
+    def wrapper(*args, **kwargs):
+        # Create a temporary directory
+        temp_dir = tempfile.mkdtemp()
+        try:
+            # Pass the temporary directory to the test function
+            test_func(*args, temp_dir=temp_dir, **kwargs)
+        finally:
+            # Clean up the directory after the test
+            shutil.rmtree(temp_dir)
+
+    return wrapper

tests/fixtures/conversation.missingturns.json

@@ -0,0 +1,52 @@
+{
+  "conversations": [
+    {
+      "from": "human",
+      "value": "how could i implement a minesweeper algorithm that utilises algebraic topology to solve boards?"
+    },
+    {
+      "from": "gpt",
+      "value": ""
+    },
+    {
+      "from": "human",
+      "value": "could you give me the gist of how it could though?"
+    },
+    {
+      "from": "gpt",
+      "value": "Sure. The basic idea would be to use algebraic topology to analyze the structure of the Minesweeper game board and determine the locations of mines based on certain topological invariants. This could involve representing the game board as a topological space and using topological invariants such as homotopy groups and Betti numbers to infer information about the location of mines on the board. However, as I mentioned before, implementing this would require a deep understanding of both algebraic topology and the game of Minesweeper, so it's not something that can be explained in a simple answer."
+    },
+    {
+      "from": "human",
+      "value": "give me a specific example"
+    },
+    {
+      "from": "gpt",
+      "value": "As a specific example, suppose we have a Minesweeper game board with the following configuration:\n```\n 1  1  1  1  1\n 1  ?  ?  ?  1\n 1  ?  ?  ?  1\n 1  ?  ?  ?  1\n 1  1  1  1  1\n```\nIn this case, the game board can be represented as a topological space, where each square on the board corresponds to a point in the space. The mines on the board can be thought of as \"holes\" in the space, which are points that are not part of the space. Using algebraic topology, we can analyze the structure of this space and determine the locations of the mines based on certain topological invariants.\n\nFor instance, one way to do this is to compute the homotopy groups of the space. Homotopy groups are a way of describing the holes in a space, and they can be used to infer information about the locations of the mines on the Minesweeper board. In the case of the board above, we can compute the first homotopy group of the space, which describes the number and \"shape\" of the holes in the space. In this case, the first homotopy group would tell us that there are four holes in the space, arranged in a square shape.\n\nBased on this information, we can infer that the mines in the game are located in the four corners of the board, as shown below:\n```\n 1  1  1  1  1\n 1  M  ?  ?  1\n 1  ?  ?  ?  1\n 1  ?  ?  M  1\n 1  1  1  1  1\n```\nOf course, this is just one example of how algebraic topology could be used to solve Minesweeper. There are many other possible approaches, and the specific details of how to implement them would depend on the specific game board and the topological invariants being used. As I mentioned before, implementing a solution like this would require a deep understanding of both algebraic topology and the game of Minesweeper."
+    },
+    {
+      "from": "human",
+      "value": "assuming i was someone with a deep understanding of both, could you explain the same concept to me with that assumption?"
+    },
+    {
+      "from": "gpt",
+      "value": "Sure. Assuming you have a deep understanding of both algebraic topology and the game of Minesweeper, here is how you might use algebraic topology to solve a Minesweeper game board:\n\n1. Represent the Minesweeper game board as a topological space. Each square on the board corresponds to a point in the space, and the mines on the board can be thought of as \"holes\" in the space, which are points that are not part of the space.\n2. Use topological invariants to infer information about the locations of the mines on the board. There are many possible topological invariants that could be used for this, such as homotopy groups, Betti numbers, or persistent homology.\n3. Use the information inferred from the topological invariants to determine the locations of the mines on the board. For example, if the first homotopy group of the space indicates that there are four holes in the space, arranged in a square shape, then we can infer that the mines are located in the four corners of the board.\n4. Use this information to solve the Minesweeper game. Once you know the locations of the mines, you can use this information to avoid them and clear the board.\n\nOf course, this is just one possible approach to using algebraic topology to solve Minesweeper. There may be other ways to do it, depending on the specific game board and the topological invariants being used. However, this should give you a general idea of how algebraic topology can be used to solve Minesweeper."
+    },
+    {
+      "from": "human",
+      "value": "how would i do it with persistent homology or any of the other topological invariants?"
+    },
+    {
+      "from": "gpt",
+      "value": "To use persistent homology or any of the other topological invariants to solve a Minesweeper game board, you would follow the same basic steps as described above, but with some differences in the specific details.\n\nFor instance, here is how you might use persistent homology to solve a Minesweeper game board:\n\n1. Represent the Minesweeper game board as a topological space. Each square on the board corresponds to a point in the space, and the mines on the board can be thought of as \"holes\" in the space, which are points that are not part of the space.\n2. Compute the persistent homology of the space. Persistent homology is a way of describing the holes in a space over time, and it can be used to infer information about the locations of the mines on the Minesweeper board.\n3. Use the information from the persistent homology to determine the locations of the mines on the board. For example, if the persistent homology indicates that there are four holes in the space, arranged in a square shape, then we can infer that the mines are located in the four corners of the board.\n4. Use this information to solve the Minesweeper game. Once you know the locations of the mines, you can use this information to avoid them and clear the board.\n\nOf course, this is just one possible approach to using persistent homology to solve Minesweeper. There may be other ways to do it, depending on the specific game board and the specific details of the persistent homology computation. However, this should give you a general idea of how persistent homology can be used to solve Minesweeper."
+    },
+    {
+      "from": "human",
+      "value": "how could i implement this in python?"
+    },
+    {
+      "from": "gpt",
+      "value": "To implement a Minesweeper solver that uses algebraic topology in Python, you would need to do the following:\n\n1. Install a Python package for algebraic topology. There are several options available, such as Gudhi, Dionysus, or scikit-tda. These packages provide algorithms and data structures for computing topological invariants, such as homotopy groups, Betti numbers, or persistent homology.\n2. Represent the Minesweeper game board as a topological space. This could be done using a data structure provided by the algebraic topology package, such as a simplicial complex or a Vietoris-Rips complex. Each square on the board would correspond to a point in the space, and the mines on the board would be represented as \"holes\" in the space.\n3. Compute the topological invariants of the space. This could be done using algorithms provided by the algebraic topology package. For example, you could compute the persistent homology of the space using the persistent homology algorithm.\n4. Use the information from the topological invariants to determine the locations of the mines on the board. This could be done using a combination of mathematical reasoning and programming. For example, if the persistent homology indicates that there are four holes in the space, arranged in a square shape, then you could use a piece of code to infer that the mines are located in the four corners of the board.\n5. Use this information to solve the Minesweeper game. Once you know the locations of the mines, you could use a piece of code to automatically click on the safe squares and clear the board.\n\nOf course, this is just one possible approach to implementing a Minesweeper solver that uses algebraic topology in Python. There may be other ways to do it, depending on the specific details of the implementation. However, this should give you a general idea of how to get started with such a project."
+    }
+  ]
+}