Update __init__.py

* fix: add multiple tips about eos_token masking

* fix: format dataset preprocessing doc

* Update docs/dataset-formats/conversation.qmd

Co-authored-by: salman <salman.mohammadi@outlook.com>

---------

Co-authored-by: salman <salman.mohammadi@outlook.com>

.github/workflows/main.yml

@@ -24,7 +24,7 @@ jobs:
             cuda_version: 12.4.1
             python_version: "3.11"
             pytorch: 2.5.1
-            axolotl_extras:
+            axolotl_extras: vllm
             is_latest: true
           - cuda: 124
             cuda_version: 12.4.1

.github/workflows/multi-gpu-e2e.yml

@@ -4,6 +4,10 @@ on:
   pull_request:
     paths:
       - 'tests/e2e/multigpu/*.py'
+      - 'requirements.txt'
+      - 'setup.py'
+      - 'pyproject.toml'
+      - '.github/workflows/multi-gpu-e2e.yml'
   workflow_dispatch:
   schedule:
     - cron: '0 0 * * 1,4'  # Runs at 00:00 UTC every monday & thursday
@@ -24,20 +28,21 @@ jobs:
             cuda_version: 12.4.1
             python_version: "3.11"
             pytorch: 2.4.1
-            axolotl_extras:
+            axolotl_extras:  # no vllm support for 2.4.1
             num_gpus: 2
             nightly_build: "true"
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"
             pytorch: 2.5.1
-            axolotl_extras:
+            axolotl_extras: vllm
             num_gpus: 2
             nightly_build: "true"
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"
             pytorch: 2.6.0
+            # awaiting vllm#12721
             axolotl_extras:
             num_gpus: 2
             nightly_build: "true"

.github/workflows/tests.yml

@@ -204,7 +204,7 @@ jobs:
             python_version: "3.11"
             pytorch: 2.5.1
             num_gpus: 1
-            axolotl_extras:
+            axolotl_extras: vllm
     steps:
       - name: Checkout
         uses: actions/checkout@v4

README.md

@@ -50,13 +50,14 @@ Features:
 ## 🚀 Quick Start
 
 **Requirements**:
+
 - NVIDIA GPU (Ampere or newer for `bf16` and Flash Attention) or AMD GPU
 - Python 3.11
 - PyTorch ≥2.4.1
 
 ### Installation
 
-```shell
+```bash
 pip3 install --no-build-isolation axolotl[flash-attn,deepspeed]
 
 # Download example axolotl configs, deepspeed configs
@@ -68,7 +69,7 @@ Other installation approaches are described [here](https://axolotl-ai-cloud.gith
 
 ### Your First Fine-tune
 
-```shell
+```bash
 # Fetch axolotl examples
 axolotl fetch examples
 

_quarto.yml

@@ -3,10 +3,12 @@ project:
 
 website:
   title: "Axolotl"
-  description: "Fine-tuning"
+  description: "We make fine-tuning accessible, scalable, and fun"
   favicon: favicon.jpg
+
   navbar:
-    title: Axolotl
+    logo: image/axolotl_logo_digital_white.svg
+    title: false
     background: dark
     pinned: false
     collapse: false
@@ -25,33 +27,58 @@ website:
       contents:
         - text: Home
           href: index.qmd
-        - section: "How-To Guides"
+
+        - section: "Getting Started"
           contents:
-          # TODO Edit folder structure after we have more docs.
             - docs/getting-started.qmd
             - docs/installation.qmd
-            - docs/debugging.qmd
+            - docs/cli.qmd
             - docs/inference.qmd
-            - docs/multipack.qmd
-            - docs/fsdp_qlora.qmd
-            - docs/input_output.qmd
-            - docs/rlhf.qmd
-            - docs/nccl.qmd
-            - docs/mac.qmd
-            - docs/multi-gpu.qmd
-            - docs/multi-node.qmd
-            - docs/unsloth.qmd
-            - docs/amd_hpc.qmd
-            - docs/ray-integration.qmd
+
         - section: "Dataset Formats"
           contents: docs/dataset-formats/*
+
+        - section: "Deployments"
+          contents:
+            - docs/multi-gpu.qmd
+            - docs/multi-node.qmd
+            - docs/ray-integration.qmd
+            - docs/amd_hpc.qmd
+            - docs/mac.qmd
+
+        - section: "How To Guides"
+          contents:
+            - docs/multimodal.qmd
+            - docs/rlhf.qmd
+            - docs/reward_modelling.qmd
+            - docs/lr_groups.qmd
+            - docs/lora_optims.qmd
+
+        - section: "Core Concepts"
+          contents:
+            - docs/batch_vs_grad.qmd
+            - docs/dataset_preprocessing.qmd
+            - docs/multipack.qmd
+
+        - section: "Advanced Features"
+          contents:
+            - docs/fsdp_qlora.qmd
+            - docs/unsloth.qmd
+            - docs/torchao.qmd
+            - docs/custom_integrations.qmd
+
+        - section: "Troubleshooting"
+          contents:
+            - docs/faq.qmd
+            - docs/debugging.qmd
+            - docs/nccl.qmd
+
         - section: "Reference"
           contents:
             - docs/config.qmd
-        - docs/faq.qmd
 
 format:
   html:
-    theme: materia
+    theme: darkly
     css: styles.css
     toc: true

cicd/cicd.sh

@@ -4,8 +4,8 @@ set -e
 python -c "import torch; assert '$PYTORCH_VERSION' in torch.__version__"
 
 pytest -v --durations=10 -n8 --ignore=tests/e2e/ --ignore=tests/patched/ /workspace/axolotl/tests/
-# pytest -v --durations=10 -n8 --dist loadfile /workspace/axolotl/tests/patched/
-pytest -v --durations=10 /workspace/axolotl/tests/e2e/patched/
+pytest -v --durations=10 /workspace/axolotl/tests/e2e/patched/lora_kernels  # running these with the other patches causes a failure
+pytest -v --durations=10 --ignore=tests/e2e/patched/lora_kernels /workspace/axolotl/tests/e2e/patched
 pytest -v --durations=10 -n1 /workspace/axolotl/tests/e2e/solo/
 pytest -v --durations=10 /workspace/axolotl/tests/e2e/integrations/
 pytest -v --durations=10 --ignore=tests/e2e/solo/ --ignore=tests/e2e/patched/ --ignore=tests/e2e/multigpu/ --ignore=tests/e2e/integrations/ /workspace/axolotl/tests/e2e/

cicd/multigpu.py

@@ -37,15 +37,11 @@ temp_dir = tempfile.mkdtemp()
 with open(pathlib.Path(temp_dir) / "Dockerfile", "w", encoding="utf-8") as f:
     f.write(dockerfile_contents)
 
-cicd_image = (
-    Image.from_dockerfile(
-        pathlib.Path(temp_dir) / "Dockerfile",
-        force_build=True,
-        gpu="A10G",
-    )
-    .env(df_args)
-    .pip_install("fastapi==0.110.0", "pydantic==2.6.3")
-)
+cicd_image = Image.from_dockerfile(
+    pathlib.Path(temp_dir) / "Dockerfile",
+    force_build=True,
+    gpu="A10G",
+).env(df_args)
 
 app = App("Axolotl CI/CD", secrets=[])
 

cicd/tests.py

@@ -1,6 +1,4 @@
-"""
- modal application to run axolotl gpu tests in Modal
- """
+"""Modal app to run axolotl GPU tests"""
 # pylint: disable=duplicate-code
 
 import os

docs/amd_hpc.qmd

@@ -1,5 +1,5 @@
 ---
-title: Training with AMD GPUs on HPC Systems
+title: AMD GPUs on HPC Systems
 description: A comprehensive guide for using Axolotl on distributed systems with AMD GPUs
 ---
 

docs/cli.qmd

@@ -1,28 +1,19 @@
-# Axolotl CLI Documentation
+---
+title: "CLI Reference"
+format:
+  html:
+    toc: true
+    toc-expand: 2
+    toc-depth: 3
+execute:
+  enabled: false
+---
 
 The Axolotl CLI provides a streamlined interface for training and fine-tuning large language models. This guide covers
 the CLI commands, their usage, and common examples.
 
-### Table of Contents
 
-- Basic Commands
-- Command Reference
-  - fetch
-  - preprocess
-  - train
-  - inference
-  - merge-lora
-  - merge-sharded-fsdp-weights
-  - evaluate
-  - lm-eval
-- Legacy CLI Usage
-- Remote Compute with Modal Cloud
-  - Cloud Configuration
-  - Running on Modal Cloud
-  - Cloud Configuration Options
-
-
-### Basic Commands
+## Basic Commands
 
 All Axolotl commands follow this general structure:
 
@@ -32,9 +23,9 @@ axolotl <command> [config.yml] [options]
 
 The config file can be local or a URL to a raw YAML file.
 
-### Command Reference
+## Command Reference
 
-#### fetch
+### fetch
 
 Downloads example configurations and deepspeed configs to your local machine.
 
@@ -49,7 +40,7 @@ axolotl fetch deepspeed_configs
 axolotl fetch examples --dest path/to/folder
 ```
 
-#### preprocess
+### preprocess
 
 Preprocesses and tokenizes your dataset before training. This is recommended for large datasets.
 
@@ -74,7 +65,7 @@ dataset_prepared_path: Local folder for saving preprocessed data
 push_dataset_to_hub: HuggingFace repo to push preprocessed data (optional)
 ```
 
-#### train
+### train
 
 Trains or fine-tunes a model using the configuration specified in your YAML file.
 
@@ -95,7 +86,38 @@ axolotl train config.yml --no-accelerate
 axolotl train config.yml --resume-from-checkpoint path/to/checkpoint
 ```
 
-#### inference
+It is possible to run sweeps over multiple hyperparameters by passing in a sweeps config.
+
+```bash
+# Basic training with sweeps
+axolotl train config.yml --sweep path/to/sweep.yaml
+```
+
+Example sweep config:
+```yaml
+_:
+  # This section is for dependent variables we need to fix
+  - load_in_8bit: false
+    load_in_4bit: false
+    adapter: lora
+  - load_in_8bit: true
+    load_in_4bit: false
+    adapter: lora
+
+# These are independent variables
+learning_rate: [0.0003, 0.0006]
+lora_r:
+  - 16
+  - 32
+lora_alpha:
+  - 16
+  - 32
+  - 64
+```
+
+
+
+### inference
 
 Runs inference using your trained model in either CLI or Gradio interface mode.
 
@@ -115,7 +137,7 @@ cat prompt.txt | axolotl inference config.yml \
     --base-model="./completed-model"
 ```
 
-#### merge-lora
+### merge-lora
 
 Merges trained LoRA adapters into the base model.
 
@@ -137,7 +159,7 @@ gpu_memory_limit: Limit GPU memory usage
 lora_on_cpu: Load LoRA weights on CPU
 ```
 
-#### merge-sharded-fsdp-weights
+### merge-sharded-fsdp-weights
 
 Merges sharded FSDP model checkpoints into a single combined checkpoint.
 
@@ -146,7 +168,7 @@ Merges sharded FSDP model checkpoints into a single combined checkpoint.
 axolotl merge-sharded-fsdp-weights config.yml
 ```
 
-#### evaluate
+### evaluate
 
 Evaluates a model's performance using metrics specified in the config.
 
@@ -155,27 +177,27 @@ Evaluates a model's performance using metrics specified in the config.
 axolotl evaluate config.yml
 ```
 
-#### lm-eval
+### lm-eval
 
 Runs LM Evaluation Harness on your model.
 
 ```bash
 # Basic evaluation
 axolotl lm-eval config.yml
-
-# Evaluate specific tasks
-axolotl lm-eval config.yml --tasks arc_challenge,hellaswag
 ```
 
 Configuration options:
 
 ```yaml
-lm_eval_tasks: List of tasks to evaluate
-lm_eval_batch_size: Batch size for evaluation
-output_dir: Directory to save evaluation results
+# List of tasks to evaluate
+lm_eval_tasks:
+  - arc_challenge
+  - hellaswag
+lm_eval_batch_size: # Batch size for evaluation
+output_dir: # Directory to save evaluation results
 ```
 
-### Legacy CLI Usage
+## Legacy CLI Usage
 
 While the new Click-based CLI is preferred, Axolotl still supports the legacy module-based CLI:
 
@@ -195,12 +217,18 @@ accelerate launch -m axolotl.cli.inference config.yml \
     --lora_model_dir="./outputs/lora-out" --gradio
 ```
 
-### Remote Compute with Modal Cloud
+::: {.callout-important}
+When overriding CLI parameters in the legacy CLI, use same notation as in yaml file (e.g., `--lora_model_dir`).
+
+**Note:** This differs from the new Click-based CLI, which uses dash notation (e.g., `--lora-model-dir`). Keep this in mind if you're referencing newer documentation or switching between CLI versions.
+:::
+
+## Remote Compute with Modal Cloud
 
 Axolotl supports running training and inference workloads on Modal cloud infrastructure. This is configured using a
 cloud YAML file alongside your regular Axolotl config.
 
-#### Cloud Configuration
+### Cloud Configuration
 
 Create a cloud config YAML with your Modal settings:
 
@@ -215,13 +243,17 @@ branch: main    # Git branch to use (optional)
 volumes:        # Persistent storage volumes
   - name: axolotl-cache
     mount: /workspace/cache
+  - name: axolotl-data
+    mount: /workspace/data
+  - name: axolotl-artifacts
+    mount: /workspace/artifacts
 
 env:            # Environment variables
   - WANDB_API_KEY
   - HF_TOKEN
 ```
 
-#### Running on Modal Cloud
+### Running on Modal Cloud
 
 Commands that support the --cloud flag:
 
@@ -239,18 +271,18 @@ axolotl train config.yml --cloud cloud_config.yml --no-accelerate
 axolotl lm-eval config.yml --cloud cloud_config.yml
 ```
 
-#### Cloud Configuration Options
+### Cloud Configuration Options
 
 ```yaml
-provider: compute provider, currently only `modal` is supported
-gpu: GPU type to use
-gpu_count: Number of GPUs (default: 1)
-memory: RAM in GB (default: 128)
-timeout: Maximum runtime in seconds
-timeout_preprocess: Preprocessing timeout
-branch: Git branch to use
-docker_tag: Custom Docker image tag
-volumes: List of persistent storage volumes
-env: Environment variables to pass
-secrets: Secrets to inject
+provider: # compute provider, currently only `modal` is supported
+gpu: # GPU type to use
+gpu_count: # Number of GPUs (default: 1)
+memory: # RAM in GB (default: 128)
+timeout: # Maximum runtime in seconds
+timeout_preprocess: # Preprocessing timeout
+branch: # Git branch to use
+docker_tag: # Custom Docker image tag
+volumes: # List of persistent storage volumes
+env: # Environment variables to pass
+secrets: # Secrets to inject
 ```

docs/config.qmd

@@ -78,6 +78,9 @@ tf32: true # require >=ampere
 bfloat16: true # require >=ampere
 float16: true
 
+# Use Tensor parallel
+tensor_parallel: true # require multi-gGPU
+
 # Limit the memory for all available GPUs to this amount (if an integer, expressed in gigabytes); default: unset
 gpu_memory_limit: 20GiB
 # Do the LoRA/PEFT loading on CPU -- this is required if the base model is so large it takes up most or all of the available GPU VRAM, e.g. during a model and LoRA merge
@@ -91,7 +94,12 @@ datasets:
     type: alpaca # format | format:<prompt_style> (chat/instruct) | <prompt_strategies>.load_<load_fn>
     ds_type: # Optional[str] (json|arrow|parquet|text|csv) defines the datatype when path is a file
     data_files: # Optional[str] path to source data files
-    shards: # Optional[int] number of shards to split data into
+
+    shards: # Optional[int] split dataset into N pieces (use with shards_idx)
+    shards_idx: # Optional[int] = 0 the index of sharded dataset to use
+
+    preprocess_shards: # Optional[int] process dataset in N sequential chunks for memory efficiency (exclusive with `shards`)
+
     name: # Optional[str] name of dataset configuration to load
     train_on_split: train # Optional[str] name of dataset split to load from
     revision: # Optional[str] The specific revision of the dataset to use when loading from the Hugging Face Hub. This can be a commit hash, tag, or branch name. If not specified, the latest version will be used. This parameter is ignored for local datasets.
@@ -137,10 +145,19 @@ datasets:
 
     # Key containing the messages (default: "messages")
     field_messages: messages
-    # Key for role in each message (default: "role")
-    message_field_role: role
-    # Key for content in each message (default:  "content")
-    message_field_content: content
+
+    # Mapping of properties from the input dataset to the chat template.
+    # (default: message_property_mappings={'role':'role', 'content':'content'})
+    # If a property exists in the template but not in this mapping, the system will attempt
+    # to load it directly from the message using the property name as the key.
+    # Example: In the mapping below, 'from' is loaded from input dataset and used as 'role',
+    # while 'value' is loaded and used as 'content' in the chat template.
+    message_property_mappings:
+      role: from
+      content: value
+      # ...
+
+    message_property_mappings:
 
     # Optional[Dict[str, List]]. Roles mapping in the messages. The default is:
     roles:
@@ -152,7 +169,7 @@ datasets:
     # IMPORTANT: The following fields determine which parts of the conversation to train on.
     # Priority order: message_field_training > message_field_training_detail > train_on_inputs or role in roles_to_train
     # See examples at `docs/dataset-formats/conversation.qmd`
-    # Note: If the below 4 fields are empty, defaults to training only on the last message.
+    # Note: If the below 4 fields are set to empty, defaults to training only on the last message.
 
     # Optional[List[str]]. Roles to train on. The tokens from these roles will be considered for the loss.
     roles_to_train: ["assistant"]  # default
@@ -160,6 +177,7 @@ datasets:
     # - all: train on all EOS tokens
     # - turn (default): train on the EOS token at the end of each trainable turn
     # - last: train on the last EOS token in the conversation
+    # TIP: Please make sure that your `tokenizer.eos_token` is same as EOS/EOT token in template. Otherwise, set `eos_token` under `special_tokens`.
     train_on_eos: last
     # The key in the message turn that indicates via boolean whether tokens of a turn should be considered for training. Useful to selectively train on certain turns besides the `roles_to_train`.
     message_field_training: training
@@ -300,6 +318,13 @@ lora_modules_to_save:
 
 lora_fan_in_fan_out: false
 
+# Apply custom LoRA autograd functions and activation function Triton kernels for
+# speed and memory savings
+# See: https://axolotl-ai-cloud.github.io/axolotl/docs/lora_optims.html
+lora_mlp_kernel: true
+lora_qkv_kernel: true
+lora_o_kernel: true
+
 # LoRA+ hyperparameters
 # For more details about the following options, see:
 # https://arxiv.org/abs/2402.12354  and `src/axolotl/core/train_builder.py`
@@ -348,6 +373,9 @@ comet_mode: # Create a new experiment ("create") or log to an existing one ("get
 comet_online: # Set to True to log data to Comet server, or False for offline storage. Default is True.
 comet_experiment_config: # Dictionary for additional configuration settings, see the doc for more details.
 
+# Tensorboard
+use_tensorboard: # Optional[bool]
+
 # Where to save the full-finetuned model to
 output_dir: ./completed-model
 
@@ -382,6 +410,12 @@ save_total_limit: # Checkpoints saved at a time
 # e.g., when 1 epoch is 1000 steps => `num_epochs: 2` and `max_steps: 100` will train for 100 steps
 max_steps:
 
+# bool of whether to include tokens trainer per second in the training metrics. This iterates over the entire dataset once, so it takes some time.
+include_tokens_per_second: # Optional[bool]
+
+# whether to find batch size that fits in memory. Passed to underlying transformers Trainer
+auto_find_batch_size: # Optional[bool]
+
 eval_table_size: # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0
 eval_max_new_tokens: # Total number of tokens generated for predictions sent to wandb. Default is 128
 eval_causal_lm_metrics: # HF evaluate metrics used during evaluation. Default is ["sacrebleu", "comet", "ter", "chrf", "perplexity"]

docs/custom_integrations.qmd

@@ -0,0 +1,57 @@
+---
+title: Custom Integrations
+toc: true
+toc-depth: 3
+---
+
+```{python}
+#| echo: false
+
+import re
+
+def process_readme(integration_name):
+    try:
+        path = f'../src/axolotl/integrations/{integration_name}/README.md'
+        with open(path, 'r') as f:
+            txt = f.read()
+            # Remove h1 headings
+            txt = re.sub(r'^# .*\n?', '', txt, flags=re.MULTILINE)
+            # Convert h2 to h3
+            txt = re.sub(r'^## ', '### ', txt, flags=re.MULTILINE)
+            return txt
+    except FileNotFoundError:
+        return None
+
+def print_section(name, folder_name):
+    output = f"\n## {name}\n"
+    content = process_readme(folder_name)
+    if content:
+        output += content
+    output += f"\nPlease see reference [here](https://github.com/axolotl-ai-cloud/axolotl/tree/main/src/axolotl/integrations/{folder_name})\n"
+    return output
+```
+
+```{python}
+#| output: asis
+#| echo: false
+
+# Introduction text
+print("""
+Axolotl adds custom features through `integrations`. They are located within the `src/axolotl/integrations` directory.
+
+To enable them, please check the respective documentations.
+""")
+
+# Sections
+sections = [
+    ("Cut Cross Entropy", "cut_cross_entropy"),
+    ("Grokfast", "grokfast"),
+    ("Knowledge Distillation (KD)", "kd"),
+    ("Liger Kernels", "liger"),
+    ("Language Model Evaluation Harness (LM Eval)", "lm_eval"),
+    ("Spectrum", "spectrum")
+]
+
+for section_name, folder_name in sections:
+    print(print_section(section_name, folder_name))
+```

docs/dataset-formats/conversation.qmd

@@ -6,7 +6,9 @@ order: 3
 
 ## sharegpt
 
-IMPORTANT: ShareGPT is deprecated!. Please see `chat_template` section below.
+::: {.callout-important}
+ShareGPT is deprecated!. Please see [chat_template](#chat_template) section below.
+:::
 
 ## pygmalion
 
@@ -22,7 +24,7 @@ Chat Template strategy uses a jinja2 template that converts a list of messages i
 {"conversations": [{"role": "...", "content": "..."}]}
 ```
 
-See `config.qmd` for full configs and supported templates.
+See [configs](../config.qmd) for full configs and supported templates.
 
 ### Migrating from sharegpt
 
@@ -42,8 +44,9 @@ datasets:
     type: chat_template
 
     field_messages: conversations
-    message_field_role: from
-    message_field_content: value
+    message_property_mappings:
+      role: from
+      content: value
 
 # new (if setting a new chat_template like chatml, gemma, etc)
 chat_template: chatml
@@ -52,8 +55,9 @@ datasets:
     type: chat_template
 
     field_messages: conversations
-    message_field_role: from
-    message_field_content: value
+    message_property_mappings:
+      role: from
+      content: value
 ```
 
 We recommend checking the below examples for other usecases.
@@ -100,6 +104,10 @@ datasets:
     type: chat_template
 ```
 
+::: {.callout-important}
+Please make sure that your `tokenizer.eos_token` is same as EOS/EOT token in template. Otherwise, set `eos_token` under `special_tokens`.
+:::
+
 5. (Advanced) Using fine-grained control over tokens and turns to train in a conversation
 
 For a data sample that looks like:
@@ -138,12 +146,15 @@ datasets:
     type: chat_template
     chat_template: tokenizer_default
     field_messages: conversations
-    message_field_role: from
-    message_field_content: value
+    message_property_mappings:
+      role: from
+      content: value
     roles_to_train: []
     train_on_eos: turn
     message_field_training: train
     message_field_training_detail: train_detail
 ```
 
-Tip: It is not necessary to use both `message_field_training` and `message_field_training_detail` at a time.
+::: {.callout-tip}
+It is not necessary to set both `message_field_training` and `message_field_training_detail` at once.
+:::

docs/dataset-formats/index.qmd

@@ -1,14 +1,491 @@
 ---
 title: Dataset Formats
-description: Supported dataset formats.
-listing:
-  fields: [title, description]
-  type: table
-  sort-ui: false
-  filter-ui: false
-  max-description-length: 250
+description: Guide to Dataset Formats in Axolotl
+back-to-top-navigation: true
+toc: true
+toc-depth: 5
 ---
 
-Axolotl supports a variety of dataset formats.  It is recommended to use a JSONL format.  The schema of the JSONL depends upon the task and the prompt template you wish to use. Instead of a JSONL, you can also use a HuggingFace dataset with columns for each JSONL field.
 
-Below are these various formats organized by task:
+Axolotl is a training framework that aims to make the process convenient yet flexible to users by simply passing a config yaml file.
+
+As there are a lot of available options in Axolotl, this guide aims to provide an simplify the user experience to choosing the proper choice.
+
+Axolotl supports 3 kinds of training methods: pre-training, supervised fine-tuning, and preference-based post-training (e.g. DPO, ORPO, PRMs). Each method has their own dataset format which are described below.
+
+## Pre-training
+
+When aiming to train on large corpora of text datasets, pre-training is your go-to choice. Due to the size of these datasets, downloading the entire-datasets before beginning training would be prohibitively time-consuming. Axolotl supports [streaming](https://huggingface.co/docs/datasets/en/stream) to only load batches into memory at a time.
+
+A sample format for a pre-training dataset is as follows:
+
+```json
+{"text": "first row"}
+{"text": "second row"}
+...
+```
+
+It is typically recommended to save your dataset as `.jsonl` due to its flexibility and simplicity.
+
+Axolotl supports loading from a Hugging Face hub repo or from local files.
+
+::: {.callout-important}
+For pre-training only, Axolotl would split texts if it exceeds the context length into multiple smaller prompts.
+:::
+
+### Pre-training from Hugging Face hub datasets
+
+As an example, to train using a Hugging Face dataset `hf_org/name`, you can pass the following config:
+
+```yaml
+pretraining_dataset: hf_org/name
+```
+
+### Pre-training from local dataset files
+
+Given a few corpus files: `A.jsonl`, `B.jsonl`, and `C.jsonl`, your config will look like the below:
+
+```yaml
+pretraining_dataset:
+  - path: json
+    data_files:
+      - A.jsonl
+      - B.jsonl
+      - C.jsonl
+```
+
+While we recommend `.jsonl`, you can also use the other formats (`csv`, `parquet`, `arrow`, `SQL`, `Webdataset`) that are supported by [`Dataset.load_dataset`](https://huggingface.co/docs/datasets/loading#local-and-remote-files)
+
+### Pre-training without streaming
+
+On the rare case that the dataset is small and can be loaded entirely into memory, another approach to running pre-training is to use the `completion` format. This would mean that the entire dataset is pre-tokenized instead of on-demand in streaming.
+
+One benefit of this is that the tokenization can be performed separately on a CPU-only machine, and then transferred to a GPU machine for training to save costs.
+
+From Hugging Face:
+
+```yaml
+datasets:
+  - path: hf_org/name
+    type: completion
+```
+
+From local files (either example works):
+
+```yaml
+datasets:
+  - path: A.jsonl
+    type: completion
+
+  - path: json
+    data_files: ["A.jsonl", "B.jsonl", "C.jsonl"]
+    type: completion
+```
+
+### Pre-training dataset configuration tips
+
+#### Setting max_steps
+
+When using streaming for large datasets, Axolotl does not know in advance how large the dataset is and does not know when to stop.
+
+Therefore, it is necessary to set `max_steps: int` in your config for pre-training to run, so that Axolotl knows when to stop training.
+
+One step is equal to `sequence_len * micro_batch_size * gradient_accumulation_steps * total_num_gpus` tokens.
+
+#### Group_by_length
+
+It is recommended to leave this off if downloading from Hugging Face hub as it would download the entire dataset which can be very large.
+
+### Reference
+
+Please see docs [here](pretraining.qmd).
+
+## Supervised fine-tuning (SFT)
+
+Supervised fine-tuning is the process of training models to respond to an instruction or chat input.
+
+As there are a wide variety of dataset formats, Axolotl tries to support a majority of the formats available in public datasets.
+
+Axolotl provides four approaches for loading datasets, however, it's easier to work backwards from the dataset you have available to figure out which approach to use.
+
+A flow chart is as follows:
+
+1. Do you already have the dataset tokenized? If yes, check [Pre-Tokenized Dataset](#pre-tokenized-dataset).
+
+2. Do you want to format the dataset yourself and manually choose each section to mask? If yes, check [Template Free Dataset](#template-free-dataset)
+
+3. Is your dataset in a "conversation" format, containing a `list[messages]`? If yes, check [Conversation Dataset](#conversation-dataset)
+
+4. Is your dataset in an "instruct" format, containing `{ instruction, response }`? If yes, check [Instruction Dataset](#instruction-dataset)
+
+If you went through the flow chart and did not find one that matches, it is recommended to preprocess your dataset into one of the above or create a thread on Github Discussion.
+
+::: {.callout-tip}
+You can mix and match within each approach or across approaches to train a model on a variety of datasets.
+:::
+
+### Pre-Tokenized Dataset
+
+We suggest this approach when you want to bring your own tokenized dataset.
+
+Axolotl expects the dataset to have three keys:
+- `input_ids`: from tokenizing formatted prompt
+- `attention_mask`: for masking padding. If you don't add padding, it would be equal to `len(input_ids) * [1]`
+- `labels`: this is the same as `input_ids`, however, if you want to mask certain tokens, you would set those indices to `-100`.
+
+::: {.callout-tip}
+Make sure to add BOS/EOS tokens to your prompt and mask it appropriately.
+:::
+
+A config for this would look like:
+
+```yaml
+datasets:
+  - path: A.jsonl
+    type:
+```
+
+::: {.callout-note}
+`type: ` is empty!
+:::
+
+Reference: [Pre-Tokenized Dataset Documentation](tokenized.qmd).
+
+### Template Free Dataset
+
+We reccomend this approach when you want granular control over the prompt formatting, special tokens, and masking, whilst letting Axolotl handle the tokenization. This is very useful if your dataset has unique prompts that differ across samples and where one single general template wouldn't suffice.
+
+In the example below, you could see that there is no proper structure. At the same time, it's very flexible as there are no constraints on how your prompt can look.
+
+```json
+{
+    "segments": [
+        {
+            "label": true,
+            "text": "<s>Hello\n"
+        },
+        {
+            "label": true,
+            "text": "hi there!. "
+        },
+        {
+            "label": false,
+            "text": "goodbye "
+        },
+        {
+            "label": true,
+            "text": "farewell</s>"
+        }
+    ]
+}
+```
+
+Each prompt must be have a key called `segments` which is a list of `{ text, label }`.
+
+```yaml
+datasets:
+  - path: A.jsonl
+    type: input_output
+```
+
+Reference: [Template Free Documentation](template_free.qmd).
+
+### Conversation Dataset
+
+`conversation` messages are a list of messages which usually contain a `role` and `content` key.
+
+::: {.callout-tip}
+Fun fact: Axolotl synonymously refers to "chat" messages as `conversation` messages due to how FastChat initially used this term to build a widely used [fastchat conversation](https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py) method for formatting chat messages prior to the creation of `chat_templates`.
+:::
+
+#### What are `chat_templates`?
+
+The current most popular and convenient method for inference is to use `chat_templates` for formatting prompts. Axolotl supports using `chat_templates` for training to ensure that the model performs in the same environment as in inference.
+
+Here's a quick rundown on `chat_template`: A `chat_template` is a Jinja2 template which formats a list of messages into a prompt.
+
+An example of a prompt formatted into a popular template called ChatML can be seen below:
+
+Single prompt (pretty-printed):
+```json
+{
+    "messages": [
+        {
+            "role": "user",
+            "content": "Hi"
+        },
+        {
+            "role": "assistant",
+            "content": "How can I help you?"
+        },
+        {
+            "role": "user",
+            "content": "Can you add 3+5?"
+        },
+        {
+            "role": "assistant",
+            "content": "The answer is 8."
+        }
+    ]
+}
+```
+
+The ChatML template is as follows:
+```jinja2
+{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}
+```
+
+The above prompt formatted into this template will result in:
+
+```
+<|im_start|>user
+Hi<|im_end|>
+<|im_start|>assistant
+How can I help you?<|im_end|>
+<|im_start|>user
+Can you add 3+5?<|im_end|>
+<|im_start|>assistant
+The answer is 8.<|im_end|>
+```
+
+By using delimiters (`<|im_start|>` and `<|im_end|>`), a prompt separates different speakers which helps the model identify which portion belongs to whom.
+
+#### Common Conversation Dataset formats
+
+Older conversation datasets with the following format are colloquially called `sharegpt` datasets.
+
+```json
+{"conversations": [{"from": "...", "value": "..."}]}
+```
+
+Newer conversation datasets usually follow the OpenAI format.
+
+```json
+{"messages": [{"role": "...", "content": "..."}]}
+```
+
+Axolotl supports both as well as allowing customization of any kind of key.
+
+#### Chat Template Usage
+
+To properly use this method, it is important to identify three things:
+
+1. Which `chat_template` would you use?
+
+2. What are the keys in your dataset, and what are the possible roles? For example, in OpenAI format, the keys would be `messages`, `role`, and `content`, respectively, whereas the possible roles are `system`, `user`, and `assistant`.
+
+3. What do you want to mask? For instance, only assistant messages, only last message, or nothing.
+
+##### Choosing a `chat_template`
+
+There are a lot of `chat_templates` out there. Axolotl supports the common ones: [supported chat templates](https://github.com/axolotl-ai-cloud/axolotl/blob/860609392184cf62a7e0ca676658b170e059ce6c/src/axolotl/utils/chat_templates.py#L17). For example, to use ChatML, it would be `chat_template: chatml`.
+
+However, it is also possible to use the already configured template within the tokenizer by specifying `chat_template: tokenizer_default`. If you want a fallback (in case some tokenizer does not have it pre-configured), you can do `chat_template: tokenizer_default_fallback_chatml` to fallback to the ChatML template if a tokenizer template was not found.
+
+One last but powerful approach is to bring your own template. This can be set via:
+
+```yaml
+chat_template_jinja: # your template
+```
+
+##### Setting `chat_template` dataset keys
+
+We currently default to OpenAI format for dataset keys, so if that's your current dataset format, there's nothing to do here.
+
+If your dataset format is different, here are the keys you should check (with their defaults):
+
+```yaml
+datasets:
+    ...
+    field_messages: messages  # this should point to the key containing the list of conversations
+    message_property_mappings:  # this is a mapping from keys in your dataset to keys in chat_template
+      role: role
+      content: content
+```
+
+In some `chat_templates` (e.g. [Gemma](https://huggingface.co/google/gemma-2b-it/blob/main/tokenizer_config.json#L1507)), the roles are hardcoded to `user` and `assistant`. Consequently, you may find it necessary to map the roles in your dataset to these above. We currently have some defaults that should work for common datasets, but if you get a `KeyError`, it would be necessary to add mapping for your roles. Here is an example of how it would look like:
+
+```yaml
+datasets:
+    ...
+    roles:
+      assistant:
+        - gpt
+        - model
+      user:
+        - human
+```
+
+In the example above, all `gpt` and `model` values are converted to `assistant`. All `human` values are converted to `user.`
+
+##### Handling masking
+
+The common use case for `chat_template` is for chat messages, therefore, it is common to mask all non-assistant messages. Assistant messages refer to the bot messages that you want the model to learn on.
+
+To train on all `assistant` messages, you would set the following configs.
+
+```yaml
+datasets:
+    ...
+    roles_to_train: ["assistant"]
+    train_on_eos: "turn"
+```
+
+The `train_on_eos` config means that it would mask all EOS tokens for turns that aren't assistant-turns. The other options are: `all` and `last` to choose which EOS to train on.
+
+Perhaps, you want to train on `assistant` and `narrator` roles, you can simply add `narrator` to the list of `roles_to_train`. You would also need to add it to the mapping of `roles` above.
+
+```yaml
+datasets:
+    ...
+    roles_to_train: ["assistant", "narrator"]
+    roles:
+      assistant:
+        - gpt
+        - model
+      user:
+        - human
+      narrator: ["narrator"]
+```
+
+::: {.callout-tip}
+As chat_templates may use hardcoded EOS/EOT tokens that are different from the tokenizer's EOS, it is highly recommended to set them. For example, `ChatML` uses `<|im_end|>` to end turns.
+
+```yaml
+special_tokens:
+  eos_token: <|im_end|>
+```
+
+:::
+
+##### Applying `chat_template`
+
+Once all the above steps are completed, you could combine all these configs together to form a bespoke configuration for your custom dataset.
+
+```yaml
+datasets:
+  - path: A.jsonl
+    type: chat_template
+
+    # step 1
+    chat_template: chatml
+
+    # step 2
+    field_messages: messages
+    message_property_mappings:
+      role: role
+      content: content
+
+    roles:
+      assistant:
+        - gpt
+        - model
+        - assistant
+      user:
+        - human
+        - user
+
+    # step 3
+    roles_to_train: ["assistant"]
+    train_on_eos: "turn"
+
+special_tokens:
+  eos_token: <|im_end|>
+```
+
+If this config were to be applied to the sample dataset above, the output would look as such (which can be retrieved via `axolotl preprocess config.yaml --debug`):
+
+```
+<|im_start|>(-100, 128256) user(-100, 882)
+(-100, 198) Hi(-100, 13347) <|im_end|>(-100, 128257)
+(-100, 198) <|im_start|>(-100, 128256) assistant(-100, 78191)
+(-100, 198) How(4438, 4438)  can(649, 649)  I(358, 358)  help(1520, 1520)  you(499, 499) ?(30, 30) <|im_end|>(128257, 128257)
+(-100, 198) <|im_start|>(-100, 128256) user(-100, 882)
+(-100, 198) Can(-100, 6854)  you(-100, 499)  add(-100, 923)  (-100, 220) 3(-100, 18) +(-100, 10) 5(-100, 20) ?(-100, 30) <|im_end|>(-100, 128257)
+(-100, 198) <|im_start|>(-100, 128256) assistant(-100, 78191)
+(-100, 198) The(791, 791)  answer(4320, 4320)  is(374, 374)  (220, 220) 8(23, 23) .(13, 13) <|im_end|>(128257, 128257)
+(-100, 198)
+```
+
+The first number refers to the label, the second refers to the `token_id`. For example, `-100` labels appear on non-assistant portions, meaning that they are masked during. For assistant portions, the label is the same as the `token_id`.
+
+::: {.callout-note}
+
+If during `preprocess`, there are a lot of warnings of `Could not find content __ boundary`, please check the FAQ section for [chat_templates](../faq.qmd#chat-templates).
+
+:::
+
+#### Reference
+
+Please see docs [here](conversation.qmd).
+
+### Instruction Dataset
+
+Instruction datasets are used to train instruction-following models and comprise a prompt, containing an instruction, and a single response. In contrast to chat datasets which may be multi-turn, instruct datasets are typically single-turn.
+
+An example is of a common format called Alpaca:
+```json
+{"instruction": "...", "input": "...", "output": "..."}
+```
+
+Using those keys, a prompt can be built based on it.
+```
+Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
+
+### Instruction:
+{instruction}
+
+### Input:
+{input}
+
+### Response:
+{output}
+```
+
+This can be configured as such:
+```yaml
+datasets:
+  - path: A.jsonl
+    type: alpaca
+```
+
+Axolotl supports many kinds of instruction dataset. All of them can be found here (https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/inst_tune.html) with their respective type and sample row format.
+
+
+Reference: [Instruction Dataset Documentation](inst_tune.qmd).
+
+#### Custom Instruct Prompt Format
+
+Due to the myriad possibilities of instruction formats, Axolotl allows customizing your own instruction format without having to dive into the code directly.
+
+In the example below, a sample row is used to output in `mistral_v1` format.
+```json
+{"input": "...", "output": "..."}
+```
+
+```yaml
+datasets:
+  - path: repo
+    type:
+      system_prompt: ""
+
+      field_system:
+      field_instruction: input
+      field_input:
+      field_output: output
+
+      # multi-line example with input
+      format: |-
+        [INST] {instruction} {input} [/INST]
+
+      # single-line example without input
+      no_input_format: "[INST] {instruction} [/INST]"
+```
+
+The config sets that the `field_instruction` is actually named `input`, and the `field_input` is empty as we don't have an `input` in this sample. Generally, `instruction` can be thought as the question to the model, and `input` as the additional information with `output` being the response. It is not necessary to have an `input` nor `system`. In the end, the most important part is to understand what format you want it to look like and how you can customize this to your use case.
+
+Reference: [Custom Instruct Prompt Format Documentation](inst_tune.qmd#how-to-add-custom-prompt-format).
+
+## Reinforcement Learning from Human Feedback (RLHF)
+
+As there are multiple RLHF methods with their own dataset requirements. Please see [RLHF documentation](../rlhf.qmd) for more detail.

docs/dataset-formats/pretraining.qmd

@@ -27,7 +27,6 @@ pretraining_dataset:
     type: pretrain
     trust_remote_code:
     skip: # number of rows of data to skip over from the beginning
-...
 ```
 
 :::

docs/dataset-formats/template_free.qmd

@@ -1,7 +1,239 @@
 ---
 title: Template-Free
 description: Construct prompts without a template.
+toc: true
+toc-depth: 3
 order: 4
 ---
 
-See [these docs](../input_output.qmd).
+## Background {#sec-background}
+
+### Masking Inputs {#masking-inputs}
+
+One of the most popular features of
+[axolotl](https://github.com/axolotl-ai-cloud/axolotl) is
+setting the following configuration value:
+
+
+```yaml
+train_on_inputs: false
+```
+
+If you declare a [dataset formats](https://github.com/axolotl-ai-cloud/axolotl?tab=readme-ov-file#dataset)
+such as `alpaca` or `chatml`, axolotl knows what is an input
+(i.e. human) vs. an output (i.e. the assistant) and masks the input
+labels so that your model can focus on predicting the outputs only.
+
+### You may not want prompt templates {#sec-you-may-not-want-prompt-templates}
+
+However, there are many situations where you don't want to use one of
+these formats or templates. This is because they can:
+
+-   Add unnecessary boilerplate to your prompts.
+-   Create artifacts like special delimiters `<|im_start|>` that can
+    quickly become footguns if you don't include them correctly at
+    inference time.
+-   Enforce a *chat* interface when you do not want one. Sometimes you
+    just want to fine-tune a model to a very specific task and do NOT
+    want multi-turn conversations, roles, etc.
+-   Limit you to only certain roles that the template allows.
+
+### The `input_output` format {#sec-the-inputoutput-format}
+
+You can construct your prompts without a template by using the
+`input_output` format, by setting `type: input_output` in your
+configuration file like this:
+
+**config.yml**
+
+```yaml
+train_on_inputs: false # Mask segments of your data
+datasets:
+  - path: output.jsonl
+    type: input_output  # use template free prompt construction
+```
+
+Unlike `type: completion`, which is also template-free,
+`type: input_output` allows you to mask segments of your text. More
+details on how this works are described below.
+
+## Usage {#sec-usage}
+
+This is how you can use the `input_output` format:
+
+### 1. Prepare Data {#sec-1-prepare-data}
+
+To use the `input_output` format, collect your data in the following
+format into a jsonl file (below is the first row from the file
+`output`.jsonl` pretty printed):
+
+```bash
+$ head -n1 output.jsonl | python -m json.tool
+```
+
+:::{.cell-output .cell-output-stdout}
+    {
+        "segments": [
+            {
+                "label": true,
+                "text": "<s>Hello\n"
+            },
+            {
+                "label": true,
+                "text": "hi there!. "
+            },
+            {
+                "label": false,
+                "text": "goodbye "
+            },
+            {
+                "label": true,
+                "text": "farewell</s>"
+            }
+        ]
+    }
+:::
+
+Set `label:false` when you want to mask a segment of text so that the
+model isn't trained on it. Some things to keep in mind:
+
+> [!IMPORTANT]
+> 1.  **EOS, BOS, spaces, newlines etc. are entirely up to you. Axolotl
+    concatenates all the segments as-is.** The tokenizer doesn't add
+    anything additional. Notice how I added spaces, newlines, `<s>`
+    (BOS), and `</s>` (EOS) myself.
+> 2.  Make sure you check the materialized output to validate that the
+    prompt is getting assembled how you like.
+
+### 2. Use `type: input_output` {#sec-2-use-type-inputoutput}
+
+Let's materialize data with our `output.jsonl` file by setting
+`type: input_output` in our axolotl config:
+
+```yaml
+# training_config.yaml
+base_model: mistralai/Mistral-7B-v0.1
+data_seed: 49
+seed: 49
+
+datasets:
+  - path: output.jsonl
+    type: input_output
+val_set_size: 0.1
+
+sequence_len: 896
+sample_packing: false
+
+micro_batch_size: 2
+gradient_accumulation_steps: 3
+eval_batch_size: 2
+num_epochs: 1
+learning_rate: 0.0002
+
+train_on_inputs: false
+special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"
+```
+
+You can use the following command to materialize your data. The
+`--debug` flag will print the tokens, along with the labels so you can
+verify that the correct items are being ignored:
+
+```bash
+axolotl preprocess training_config.yaml --debug
+
+...
+[2024-03-05 23:36:46,969] [INFO] [axolotl.check_example_labels:35] [PID:607731] [RANK:0] <s>(1, 1) Hello(22557, 22557)
+(13, 13) hi(12014, 12014) there(736, 736) !(28808, 28808) .(28723, 28723) (28705, 28705) good(-100, 1179) bye(-100, 17664) (-100, 28705) fare(19111, 19111) well(5458, 5458) </s>(2, 2)
+
+```
+
+The format is `decoded_token`(`label`, `token_id`), for example,
+`<s>(1, 1)` means that the token is `<s>`, the label is `1` and the
+token_id is `1`. When the label is `-100` then that token is ignored for
+training.
+
+### 3. Check the prompts {#sec-3-check-the-prompts}
+
+Here is another way to check the materialized output:
+
+```python
+from transformers import AutoTokenizer
+from datasets import load_from_disk
+import yaml
+
+directory = !ls last_run_prepared/
+with open('training_config.yaml', 'r') as f:
+    cfg = yaml.safe_load(f)
+model_id = cfg['base_model']
+tok = AutoTokenizer.from_pretrained(model_id)
+ds = load_from_disk(f'last_run_prepared/{directory[0]}/')
+```
+
+```python
+>>> row = ds[0]
+>>> print(tok.decode(row['input_ids']))
+<s> Hello
+    hi there!.  goodbye  farewell</s>
+```
+
+We can check that the right tokens are ignored by comparing the labels
+to each token:
+
+```python
+import pandas as pd
+pd.DataFrame([{'token': tok.decode(i), 'label': l, 'id':i} for i,l in
+              zip(row['input_ids'], row['labels'])])
+```
+
+| token | label | id    |
+|-------|-------|-------|
+| 0     | \<s\> | 1     |
+| 1     | Hello | 22557 |
+| 2     | \\n   | 13    |
+| 3     | hi    | 12014 |
+| 4     | there | 736   |
+| 5     | !     | 28808 |
+| 6     | .     | 28723 |
+| 7     |       | 28705 |
+| 8     | good  | -100  |
+| 9     | bye   | -100  |
+| 10    |       | -100  |
+| 11    | fare  | 19111 |
+| 12    | well  | 5458  |
+| 13    | \</s\>| 2     |
+
+
+
+If we look at the input data, the above table seems correct! (The jsonl
+version is repeated below for reference):
+
+
+```bash
+$ head -n1 output.jsonl | python -m json.tool
+```
+
+:::{.cell-output .cell-output-stdout}
+    {
+        "segments": [
+            {
+                "label": true,
+                "text": "<s>Hello\n"
+            },
+            {
+                "label": true,
+                "text": "hi there!. "
+            },
+            {
+                "label": false,
+                "text": "goodbye "
+            },
+            {
+                "label": true,
+                "text": "farewell</s>"
+            }
+        ]
+    }
+:::

docs/dataset_preprocessing.qmd

@@ -3,8 +3,11 @@ title: Dataset Preprocessing
 description: How datasets are processed
 ---
 
+## Overview
+
 Dataset pre-processing is the step where Axolotl takes each dataset you've configured alongside
-the (dataset format)[../dataset-formats/] and prompt strategies to:
+the [dataset format](docs/dataset-formats) and prompt strategies to:
+
  - parse the dataset based on the *dataset format*
  - transform the dataset to how you would interact with the model based on the *prompt strategy*
  - tokenize the dataset based on the configured model & tokenizer
@@ -12,10 +15,12 @@ the (dataset format)[../dataset-formats/] and prompt strategies to:
 
 The processing of the datasets can happen one of two ways:
 
-1. Before kicking off training by calling `python -m axolotl.cli.preprocess /path/to/your.yaml --debug`
+1. Before kicking off training by calling `axolotl preprocess config.yaml --debug`
 2. When training is started
 
-What are the benefits of pre-processing? When training interactively or for sweeps
+### What are the benefits of pre-processing?
+
+When training interactively or for sweeps
 (e.g. you are restarting the trainer often), processing the datasets can oftentimes be frustratingly
 slow. Pre-processing will cache the tokenized/formatted datasets according to a hash of dependent
 training parameters so that it will intelligently pull from its cache when possible.
@@ -28,8 +33,12 @@ default path of `./last_run_prepared/`, but will ignore anything already cached
 setting `dataset_prepared_path: ./last_run_prepared`, the trainer will use whatever pre-processed
 data is in the cache.
 
-What are the edge cases? Let's say you are writing a custom prompt strategy or using a user-defined
+### What are the edge cases?
+
+Let's say you are writing a custom prompt strategy or using a user-defined
 prompt template. Because the trainer cannot readily detect these changes, we cannot change the
-calculated hash value for the pre-processed dataset. If you have `dataset_prepared_path: ...` set
+calculated hash value for the pre-processed dataset.
+
+If you have `dataset_prepared_path: ...` set
 and change your prompt templating logic, it may not pick up the changes you made and you will be
 training over the old prompt.

docs/debugging.qmd

@@ -31,11 +31,13 @@ While debugging it's helpful to simplify your test scenario as much as possible.
     - Set `CUDA_VISIBLE_DEVICES` to a single GPU, ex: `export CUDA_VISIBLE_DEVICES=0`.
     - Set `dataset_processes: 1` in your axolotl config or run the training command with `--dataset_processes=1`.
 2. **Use a small dataset**: Construct or use a small dataset from HF Hub. When using a small dataset, you will often have to make sure `sample_packing: False` and `eval_sample_packing: False` to avoid errors.  If you are in a pinch and don't have time to construct a small dataset but want to use from the HF Hub, you can shard the data (this will still tokenize the entire dataset, but will only use a fraction of the data for training.  For example, to shard the dataset into 20 pieces, add the following to your axolotl config):
+
     ```yaml
-    dataset:
+    datasets:
         ...
         shards: 20
     ```
+
 3. **Use a small model**: A good example of a small model is [TinyLlama/TinyLlama-1.1B-Chat-v1.0](https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v1.0).
 4. **Minimize iteration time**: Make sure the training loop finishes as fast as possible, with these settings.
     - `micro_batch_size: 1`
@@ -85,7 +87,7 @@ The easiest way to get started is to modify the [.vscode/launch.json](../.vscode
 
 For example, to mimic the command `cd devtools && CUDA_VISIBLE_DEVICES=0 accelerate launch -m axolotl.cli.train dev_chat_template.yml`, you would use the below configuration[^1].  Note that we add additional flags that override the axolotl config and incorporate the tips above (see the comments). We also set the working directory to `devtools` and set the `env` variable `HF_HOME` to a temporary folder that is later partially deleted.  This is because we want to delete the HF dataset cache before each run in order to ensure that the data preprocessing code is run from scratch.
 
-```jsonc
+```json
 // .vscode/launch.json
 {
     "version": "0.2.0",
@@ -132,7 +134,7 @@ For example, to mimic the command `cd devtools && CUDA_VISIBLE_DEVICES=0 acceler
 
 Below is the [./vscode/tasks.json](../.vscode/tasks.json) file that defines the `cleanup-for-dataprep` task.  This task is run before each debugging session when you use the above configuration.  Note how there are two tasks that delete the two folders mentioned above.  The third task `cleanup-for-dataprep` is a composite task that combines the two tasks.  A composite task is necessary because VSCode does not allow you to specify multiple tasks in the `preLaunchTask` argument of the `launch.json` file.
 
-```jsonc
+```json
 // .vscode/tasks.json
 // this file is used by launch.json
 {

docs/faq.qmd

@@ -3,6 +3,7 @@ title: FAQ
 description: Frequently asked questions
 ---
 
+### General
 
 **Q: The trainer stopped and hasn't progressed in several minutes.**
 
@@ -23,3 +24,29 @@ description: Frequently asked questions
 **Q: The codes is stuck on saving preprocessed datasets.**
 
 > A: This is usually an issue with the GPU. This can be resolved through setting the os environment variable `CUDA_VISIBLE_DEVICES=0`. If you are on runpod, this is usually a pod issue. Starting a new pod should take care of it.
+
+### Chat templates
+
+**Q: `jinja2.exceptions.UndefinedError: 'dict object' has no attribute 'content' / 'role' / ____`**
+
+> A: This means that the property mapping for the stated attribute does not exist when building `chat_template` prompt. For example, if `no attribute 'content'`, please check you have added the correct mapping for `content` under `message_property_mappings`.
+
+**Q: `Empty template generated for turn ___`**
+
+> A: The `content` is empty for that turn.
+
+**Q: `Could not find content start/end boundary for turn __`**
+
+> A: The specific turn's start/end could not be detected. Please ensure you have set the `eos_token` following your `chat_template`. Otherwise, this could be a `chat_template` which doesn't use proper boundaries for each turn (like system). On the rare occurrence, make sure your content is not `[[dummy_message]]`. Please let us know about this.
+
+**Q: `Content end boundary is before start boundary for turn ___`**
+
+> A: This is an edge case which should not occur. Please create an Issue if this happens.
+
+**Q: `Content end boundary is the same as start boundary for turn ___. This is likely an empty turn.`**
+
+> A: This is likely an empty turn.
+
+**Q: The EOS/EOT token is incorrectly being masked or not being masked.**
+
+> A: This is because of the mismatch between `tokenizer.eos_token` and EOS/EOT token in template. Please make sure to set `eos_token` under `special_tokens` to the same EOS/EOT token as in template.

docs/getting-started.qmd

@@ -1,5 +1,5 @@
 ---
-title: "Getting Started with Axolotl"
+title: "Quickstart"
 format:
   html:
     toc: true
@@ -17,12 +17,12 @@ Let's start by fine-tuning a small language model using LoRA. This example uses
 Assuming `axolotl` is installed (if not, see our [Installation Guide](installation.qmd))
 
 1. Download example configs:
-```shell
+```bash
 axolotl fetch examples
 ```
 
 2. Run the training:
-```shell
+```bash
 axolotl train examples/llama-3/lora-1b.yml
 ```
 
@@ -108,7 +108,7 @@ Please consult the supported [Dataset Formats](dataset-formats/) for more detail
 
 3. Run the training:
 
-```shell
+```bash
 axolotl train my_training.yml
 ```
 
@@ -118,7 +118,7 @@ axolotl train my_training.yml
 
 After training, test your model:
 
-```shell
+```bash
 axolotl inference my_training.yml --lora-model-dir="./outputs/lora-out"
 ```
 
@@ -126,7 +126,7 @@ axolotl inference my_training.yml --lora-model-dir="./outputs/lora-out"
 
 For large datasets, preprocess first:
 
-```shell
+```bash
 axolotl preprocess my_training.yml
 ```
 
@@ -134,7 +134,7 @@ axolotl preprocess my_training.yml
 
 Launch a Gradio interface:
 
-```shell
+```bash
 axolotl inference my_training.yml --lora-model-dir="./outputs/lora-out" --gradio
 ```
 

docs/inference.qmd

@@ -1,11 +1,10 @@
 ---
-title: "Inference Guide"
+title: "Inference"
 format:
   html:
     toc: true
     toc-depth: 3
     number-sections: true
-    code-tools: true
 execute:
   enabled: false
 ---

docs/input_output.qmd

@@ -3,263 +3,4 @@ title: Template-free prompt construction
 description: "Template-free prompt construction with the `input_output` format"
 ---
 
-<!-- TOC -->
-
-- [Background](#background)
-    - [Masking Inputs](#masking-inputs)
-    - [You may not want prompt templates](#you-may-not-want-prompt-templates)
-    - [The `input_output` format](#the-input_output-format)
-- [Usage](#usage)
-    - [1. Prepare Data](#1-prepare-data)
-    - [2. Use `type: input_output`](#2-use-type-input_output)
-    - [3. Check the prompts](#3-check-the-prompts)
-
-<!-- /TOC -->
-
-<a id="markdown-background" name="background"></a>
-
-## Background
-
-<a id="markdown-masking-inputs" name="masking-inputs"></a>
-
-### Masking Inputs
-
-One of the most popular features of
-[axolotl](https://github.com/axolotl-ai-cloud/axolotl) is
-setting the following configuration value:
-
-
-```yaml
-train_on_inputs: false
-```
-
-If you declare a [dataset formats](https://github.com/axolotl-ai-cloud/axolotl?tab=readme-ov-file#dataset)
-such as `alpaca` or `chatml`, axolotl knows what is an input
-(i.e. human) vs. an output (i.e. the assistant) and masks the input
-labels so that your model can focus on predicting the outputs only.
-
-<a id="markdown-you-may-not-want-prompt-templates" name="you-may-not-want-prompt-templates"></a>
-
-### You may not want prompt templates
-
-However, there are many situations where you don't want to use one of
-these formats or templates. This is because they can:
-
--   Add unnecessary boilerplate to your prompts.
--   Create artifacts like special delimiters `<|im_start|>` that can
-    quickly become footguns if you don't include them correctly at
-    inference time.
--   Enforce a *chat* interface when you do not want one. Sometimes you
-    just want to fine-tune a model to a very specific task and do NOT
-    want multi-turn conversations, roles, etc.
--   Limit you to only certain roles that the template allows.
-
-<a id="markdown-the-inputoutput-format" name="the-inputoutput-format"></a>
-
-### The `input_output` format
-
-You can construct your prompts without a template by using the
-`input_output` format, by setting `type: input_output` in your
-configuration file like this:
-
-**config.yml**
-
-```yaml
-train_on_inputs: false # Mask segments of your data
-datasets:
-  - path: output.jsonl
-    type: input_output  # use template free prompt construction
-```
-
-Unlike `type: completion`, which is also template-free,
-`type: input_output` allows you to mask segments of your text. More
-details on how this works are described below.
-
-<a id="markdown-usage" name="usage"></a>
-
-## Usage
-
-This is how you can use the `input_output` format:
-
-<a id="markdown-1-prepare-data" name="1-prepare-data"></a>
-
-### 1. Prepare Data
-
-To use the `input_output` format, collect your data in the following
-format into a jsonl file (below is the first row from the file
-`output`.jsonl` pretty printed):
-
-```bash
-$ head -n1 output.jsonl | python -m json.tool
-```
-
-:::{.cell-output .cell-output-stdout}
-    {
-        "segments": [
-            {
-                "label": true,
-                "text": "<s>Hello\n"
-            },
-            {
-                "label": true,
-                "text": "hi there!. "
-            },
-            {
-                "label": false,
-                "text": "goodbye "
-            },
-            {
-                "label": true,
-                "text": "farewell</s>"
-            }
-        ]
-    }
-:::
-
-Set `label:false` when you want to mask a segment of text so that the
-model isn't trained on it. Some things to keep in mind:
-
-> [!IMPORTANT]
-> 1.  **EOS, BOS, spaces, newlines etc. are entirely up to you. Axolotl
-    concatenates all the segments as-is.** The tokenizer doesn't add
-    anything additional. Notice how I added spaces, newlines, `<s>`
-    (BOS), and `</s>` (EOS) myself.
-> 2.  Make sure you check the materialized output to validate that the
-    prompt is getting assembled how you like.
-
-<a id="markdown-2-use-type-inputoutput" name="2-use-type-inputoutput"></a>
-
-### 2. Use `type: input_output`
-
-Let's materialize data with our `output.jsonl` file by setting
-`type: input_output` in our axolotl config:
-
-```yaml
-# training_config.yaml
-base_model: mistralai/Mistral-7B-v0.1
-data_seed: 49
-seed: 49
-
-datasets:
-  - path: output.jsonl
-    type: input_output
-val_set_size: 0.1
-
-sequence_len: 896
-sample_packing: false
-
-micro_batch_size: 2
-gradient_accumulation_steps: 3
-eval_batch_size: 2
-num_epochs: 1
-learning_rate: 0.0002
-
-train_on_inputs: false
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"
-```
-
-You can use the following command to materialize your data. The
-`--debug` flag will print the tokens, along with the labels so you can
-verify that the correct items are being ignored:
-
-```bash
-$ python -m axolotl.cli.preprocess training_config.yaml --debug
-
-...
-[2024-03-05 23:36:46,969] [INFO] [axolotl.check_example_labels:35] [PID:607731] [RANK:0] <s>(1, 1) Hello(22557, 22557)
-(13, 13) hi(12014, 12014) there(736, 736) !(28808, 28808) .(28723, 28723) (28705, 28705) good(-100, 1179) bye(-100, 17664) (-100, 28705) fare(19111, 19111) well(5458, 5458) </s>(2, 2)
-
-```
-
-The format is `decoded_token`(`label`, `token_id`), for example,
-`<s>(1, 1)` means that the token is `<s>`, the label is `1` and the
-token_id is `1`. When the label is `-100` then that token is ignored for
-training.
-
-<a id="markdown-3-check-the-prompts" name="3-check-the-prompts"></a>
-
-### 3. Check the prompts
-
-Here is another way to check the materialized output:
-
-```python
-from transformers import AutoTokenizer
-from datasets import load_from_disk
-import yaml
-
-directory = !ls last_run_prepared/
-with open('training_config.yaml', 'r') as f:
-    cfg = yaml.safe_load(f)
-model_id = cfg['base_model']
-tok = AutoTokenizer.from_pretrained(model_id)
-ds = load_from_disk(f'last_run_prepared/{directory[0]}/')
-```
-
-```python
->>> row = ds[0]
->>> print(tok.decode(row['input_ids']))
-<s> Hello
-    hi there!.  goodbye  farewell</s>
-```
-
-We can check that the right tokens are ignored by comparing the labels
-to each token:
-
-```python
-import pandas as pd
-pd.DataFrame([{'token': tok.decode(i), 'label': l, 'id':i} for i,l in
-              zip(row['input_ids'], row['labels'])])
-```
-
-| token | label | id    |
-|-------|-------|-------|
-| 0     | \<s\> | 1     |
-| 1     | Hello | 22557 |
-| 2     | \\n   | 13    |
-| 3     | hi    | 12014 |
-| 4     | there | 736   |
-| 5     | !     | 28808 |
-| 6     | .     | 28723 |
-| 7     |       | 28705 |
-| 8     | good  | -100  |
-| 9     | bye   | -100  |
-| 10    |       | -100  |
-| 11    | fare  | 19111 |
-| 12    | well  | 5458  |
-| 13    | \</s\>| 2     |
-
-
-
-If we look at the input data, the above table seems correct! (The jsonl
-version is repeated below for reference):
-
-
-```bash
-$ head -n1 output.jsonl | python -m json.tool
-```
-
-:::{.cell-output .cell-output-stdout}
-    {
-        "segments": [
-            {
-                "label": true,
-                "text": "<s>Hello\n"
-            },
-            {
-                "label": true,
-                "text": "hi there!. "
-            },
-            {
-                "label": false,
-                "text": "goodbye "
-            },
-            {
-                "label": true,
-                "text": "farewell</s>"
-            }
-        ]
-    }
-:::
+The documentation moved to [here](dataset-formats/template_free.qmd).

docs/installation.qmd

@@ -1,11 +1,10 @@
 ---
-title: "Installation Guide"
+title: "Installation"
 format:
   html:
     toc: true
     toc-depth: 3
     number-sections: true
-    code-tools: true
 execute:
   enabled: false
 ---

docs/lora_optims.qmd

@@ -0,0 +1,127 @@
+---
+title: "LoRA Optimizations"
+description: "Custom autograd functions and Triton kernels in Axolotl for optimized LoRA fine-tuning"
+---
+
+Inspired by [Unsloth](https://github.com/unslothai/unsloth), we've implemented two
+optimizations for LoRA and QLoRA fine-tuning, supporting both single GPU and multi-GPU
+(in the DDP and DeepSpeed settings) training. These include (1) SwiGLU and GEGLU activation function
+Triton kernels, and (2) LoRA MLP and attention custom autograd functions. Our goal was
+to leverage operator fusion and tensor re-use in order to improve speed and reduce
+memory usage during the forward and backward passes of these calculations.
+
+We currently support several common model architectures, including (but not limited to):
+
+- `llama`
+- `mistral`
+- `qwen2`
+- `gemma`
+- `gemma2`
+
+<details>
+
+The set of models we support is currently limited by our attention patching strategy,
+which assumes (and replaces) specific code blocks for query / key / value and output
+projections:
+
+```python
+ORIGINAL_QKV_CODE = """
+    query_states = self.q_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+    key_states = self.k_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+    value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+""".lstrip(
+    "\n"
+)
+
+ORIGINAL_O_CODE = """
+    attn_output = self.o_proj(attn_output)
+""".lstrip(
+    "\n"
+)
+```
+
+Is replaced with:
+
+```python
+PATCHED_QKV_CODE = """
+    query_states, key_states, value_states = self.apply_qkv(hidden_states)
+    query_states = query_states.view(hidden_shape).transpose(1, 2)
+    key_states = key_states.view(hidden_shape).transpose(1, 2)
+    value_states = value_states.view(hidden_shape).transpose(1, 2)
+""".lstrip(
+    "\n"
+)
+
+PATCHED_O_CODE = """
+    attn_output = self.apply_o(attn_output)
+""".lstrip(
+    "\n"
+)
+```
+
+Where `apply_qkv` and `apply_o` are defined in the `axolotl.kernels.lora` module.
+
+We welcome testing of other model architectures and / or PRs to expand our patching
+logic to be compatible with more of them.
+
+</details>
+
+## Usage
+
+These optimizations can be enabled in your Axolotl config YAML file. The
+`lora_mlp_kernel` option enables the optimized MLP path, while `lora_qkv_kernel` and
+`lora_o_kernel` enable the fused query-key-value projection and optimized output
+projection, respectively.
+
+```yaml
+lora_mlp_kernel: true
+lora_qkv_kernel: true
+lora_o_kernel: true
+```
+
+## Requirements
+
+- One or more NVIDIA or AMD GPUs (in order to use the Triton kernels)
+    - Note: Set `TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1` to enable [memory-efficient attention on AMD GPUs](https://github.com/ROCm/aotriton/issues/16#issuecomment-2346675491)
+- Targeted LoRA adapters cannot use Dropout
+    - This may limit model expressivity / cause overfitting
+- Targeted LoRA adapters cannot have bias terms
+    - This may limit model expressivity
+
+Models with pre-existing LoRA adapters that use Dropout or have bias terms may need to
+be re-finetuned without these features in order to be useful.
+
+## Implementation details
+
+### Custom autograd functions
+
+The LoRA MLP autograd function optimizes the entire MLP computation path. It fuses the
+LoRA and base weight computations together and provides a single, efficient backward
+pass for the entire MLP block.
+
+For attention components, similar optimizations are provided through a function that
+handles the query, key, and value projections, and a function that handles the output
+projection. They are designed to work with the existing `transformers` attention
+implementation via some monkey-patching logic.
+
+### Triton kernels
+
+Two activation functions (SwiGLU and GeGLU) are implemented with Triton kernels for
+improved speed and memory performance. These kernels handle both the forward and
+backward passes.
+
+### Integration
+
+The custom autograd functions and Triton kernels are designed to work together. The
+autograd function manages the high-level computation flow and gradient tracking, while
+calling the Triton kernels for the activation function computation. During the backward
+pass, the kernel computes both the activation output and the required gradients, which
+the autograd function then uses to compute the final gradients for the entire
+computation path.
+
+## Future Work
+
+- Support for additional model architectures
+- Support for the FSDP setting
+- Support for dropout and bias
+- Additional operator fusions

docs/mac.qmd

@@ -19,4 +19,5 @@ Current support:
 - [ ] DeepSpeed
 
 Untested:
+
 - FSDP

docs/multi-gpu.qmd

@@ -1,5 +1,5 @@
 ---
-title: "Multi-GPU Training Guide"
+title: "Multi-GPU"
 format:
   html:
     toc: true
@@ -35,7 +35,11 @@ deepspeed: deepspeed_configs/zero1.json
 ### Usage {#sec-deepspeed-usage}
 
 ```{.bash}
-accelerate launch -m axolotl.cli.train examples/llama-2/config.yml --deepspeed deepspeed_configs/zero1.json
+# Passing arg via config
+axolotl train config.yml
+
+# Passing arg via cli
+axolotl train config.yml --deepspeed deepspeed_configs/zero1.json
 ```
 
 ### ZeRO Stages {#sec-zero-stages}
@@ -70,25 +74,7 @@ For combining FSDP with QLoRA, see our [dedicated guide](fsdp_qlora.qmd).
 
 ### Liger Kernel Integration {#sec-liger}
 
-::: {.callout-note}
-Liger Kernel provides efficient Triton kernels for LLM training, offering:
-
-- 20% increase in multi-GPU training throughput
-- 60% reduction in memory usage
-- Compatibility with both FSDP and DeepSpeed
-:::
-
-Configuration:
-
-```{.yaml}
-plugins:
-  - axolotl.integrations.liger.LigerPlugin
-liger_rope: true
-liger_rms_norm: true
-liger_glu_activation: true
-liger_layer_norm: true
-liger_fused_linear_cross_entropy: true
-```
+Please see [docs](custom_integrations.qmd#liger) for more info.
 
 ## Troubleshooting {#sec-troubleshooting}
 

docs/multi-node.qmd

@@ -13,7 +13,7 @@ You will also need to have the same configuration file for your model on each ma
 Make sure the main machine is reachable by other machines.
 :::
 
-# Accelerate
+## Accelerate
 
 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:
 
@@ -51,17 +51,17 @@ fsdp_config:
 
 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.
 
-# Raytrain
+## Raytrain
 
 Please see ray train doc [here](ray-integration.qmd).
 
-# Torchrun
+## Torchrun
 
 If you are using Infiniband, we recommend torchrun to utilize the full bandwidth.
 
 Set the following env (change buffersize/socketname depending on your system):
 
-```yaml
+```bash
 export NCCL_IB_DISABLE=0
 export NCCL_SOCKET_IFNAME="eth0,en,eth,em,bond"
 export NCCL_BUFFSIZE=2097152

docs/nccl.qmd

@@ -13,13 +13,13 @@ Often, this timeout will happen after 30 minutes (the default setting) and is ac
 
 Forcing cross-GPU communication via [NVLink](https://en.wikipedia.org/wiki/NVLink) may help without increasing timeouts. To verify that your configuration is leveraging NVLink run the following command:
 
-```shell
+```bash
 nvidia-smi nvlink --status
 ```
 
 To force NCCL to use NVLink, simply set this in the environment:
 
-```shell
+```bash
 export NCCL_P2P_LEVEL=NVL
 ```
 
@@ -33,13 +33,13 @@ If NVLink is not available in your environment there are other options for ``NCC
 
 To validate that acceptable data transfer speeds exist for your training job, running [NCCL Tests](https://github.com/NVIDIA/nccl-tests/blob/master/README.md) can help pinpoint bottlenecks, for example:
 
-```shell
+```bash
 ./build/all_reduce_perf -b 8 -e 128M -f 2 -g 3
 ```
 
 It can be useful when debugging NCCL communication timeouts to activate additional logging in both PyTorch and NCCL:
 
-```shell
+```bash
 export NCCL_DEBUG=INFO
 export NCCL_DEBUG_SUBSYS=ALL
 export TORCH_DISTRIBUTED_DEBUG=INFO

docs/ray-integration.qmd

@@ -1,5 +1,5 @@
 ---
-title: Ray Train integration
+title: Ray Train
 description: How to use Axolotl with Ray Train
 ---
 
@@ -9,7 +9,7 @@ With the `--use-ray` CLI flag, Axolotl will use Ray Train's [`TorchTrainer`](htt
 
 ## Ray cluster setup
 
-A prerequisite using the Ray Train integration is to setup a Ray cluster on your desired node(s). For a detailed guide on how you can get started with ray clusters, check the official Ray docs here: https://docs.ray.io/en/latest/cluster/getting-started.html
+A prerequisite using the Ray Train integration is to setup a Ray cluster on your desired node(s). For a detailed guide on how you can get started with ray clusters, check the official Ray docs [here](https://docs.ray.io/en/latest/cluster/getting-started.html).
 
 Every Ray cluster has one _head_ node and a set of worker nodes. The head node is just like any other worker node, but it also runs certain special processes related to scheduling and orchestration. Ray-enabled scripts are run on the head node and depending on the resources (number of CPUs, GPUs, etc) they request, will be scheduled to run certain tasks on the worker nodes. For more on key concepts behind a Ray cluster, you can refer this [doc](https://docs.ray.io/en/latest/cluster/key-concepts.html#cluster-key-concepts).
 
@@ -58,13 +58,11 @@ You can find an example configuration at `configs/llama-3/lora-1b-ray.yaml`.
 The key parameters to note here are:
 
 ```yaml
-...
 use_ray: true
 ray_num_workers: 4
 # optional
 resources_per_worker:
     GPU: 1
-...
 ```
 
 - `use_ray`: This is the flag that enables the Ray Train integration. You can either use the corresponding `--use-ray` flag in the CLI or set `use_ray` in the config file.

docs/rlhf.qmd

@@ -1,26 +1,39 @@
 ---
 title: "RLHF (Beta)"
 description: "Reinforcement Learning from Human Feedback is a method whereby a language model is optimized from data using human feedback."
+back-to-top-navigation: true
+toc: true
+toc-depth: 4
 ---
 
-### Overview
+## Overview
 
 Reinforcement Learning from Human Feedback is a method whereby a language model is optimized from data using human
 feedback. Various methods include, but not limited to:
 
+- [Direct Preference Optimization (DPO)](#dpo)
+- [Identity Preference Optimization (IPO)](#ipo)
+- [Kahneman-Tversky Optimization (KTO)](#kto)
+- [Odds Ratio Preference Optimization (ORPO)](#orpo)
 - Proximal Policy Optimization (PPO) (not yet supported in axolotl)
-- Direct Preference Optimization (DPO)
-- Identity Preference Optimization (IPO)
 
 
-### RLHF using Axolotl
+## RLHF using Axolotl
 
->[!IMPORTANT]
->This is a BETA feature and many features are not fully implemented. You are encouraged to open new PRs to improve the integration and functionality.
+::: {.callout-important}
+This is a BETA feature and many features are not fully implemented. You are encouraged to open new PRs to improve the integration and functionality.
+:::
 
-The various RL training methods are implemented in trl and wrapped via axolotl. Below are various examples with how you can use various preference datasets to train models that use ChatML
+We rely on the [TRL](https://github.com/huggingface/trl) library for implementations of various RL training methods, which we wrap around to expose in axolotl. Each method has their own supported ways of loading datasets and prompt formats.
+
+::: {.callout-tip}
+You can find what each method supports by going into `src/axolotl/prompt_strategies/{method}` where `{method}` is one of our supported methods. The `type: ` can be retrieved from `{method}.{function_name}`.
+:::
+
+### DPO
+
+Example config:
 
-#### DPO
 ```yaml
 rl: dpo
 datasets:
@@ -32,12 +45,265 @@ datasets:
     type: chatml
 ```
 
-#### IPO
+DPO supports the following types with the following dataset format:
+
+#### chatml.argilla
+
+```json
+{
+    "system": "...", // optional
+    "instruction": "...",
+    "chosen_response": "...",
+    "rejected_response": "..."
+}
+```
+
+#### chatml.argilla_chat
+
+```json
+{
+    "chosen": [
+        {"role": "user", "content": "..."},
+        {"role": "assistant", "content": "..."}
+    ],
+    "rejected": [
+        {"role": "user", "content": "..."},
+        {"role": "assistant", "content": "..."}
+    ]
+}
+```
+
+#### chatml.icr
+
+```json
+{
+    "system": "...", // optional
+    "input": "...",
+    "chosen": "...",
+    "rejected": "..."
+}
+```
+
+#### chatml.intel
+
+```json
+{
+    "system": "...", // optional
+    "question": "...",
+    "chosen": "...",
+    "rejected": "..."
+}
+```
+
+#### chatml.prompt_pairs
+
+```json
+{
+    "system": "...", // optional
+    "prompt": "...",
+    "chosen": "...",
+    "rejected": "..."
+}
+```
+
+#### chatml.ultra
+
+```json
+{
+    "system": "...", // optional
+    "prompt": "...",
+    "chosen": [
+        {"role": "user", "content": "..."},
+        {"role": "assistant", "content": "..."}
+    ],
+    "rejected": [
+        {"role": "user", "content": "..."},
+        {"role": "assistant", "content": "..."}
+    ]
+}
+```
+
+#### llama3.argilla
+
+```json
+{
+    "system": "...", // optional
+    "instruction": "...",
+    "chosen_response": "...",
+    "rejected_response": "..."
+}
+```
+
+#### llama3.argilla_chat
+
+```json
+{
+    "chosen": [
+        {"role": "user", "content": "..."},
+        {"role": "assistant", "content": "..."}
+    ],
+    "rejected": [
+        {"role": "user", "content": "..."},
+        {"role": "assistant", "content": "..."}
+    ]
+}
+```
+
+#### llama3.icr
+
+```json
+{
+    "system": "...", // optional
+    "input": "...",
+    "chosen": "...",
+    "rejected": "..."
+}
+```
+
+#### llama3.intel
+
+```json
+{
+    "system": "...", // optional
+    "question": "...",
+    "chosen": "...",
+    "rejected": "..."
+}
+```
+
+#### llama3.prompt_pairs
+
+```json
+{
+    "system": "...", // optional
+    "prompt": "...",
+    "chosen": "...",
+    "rejected": "..."
+}
+```
+
+#### llama3.ultra
+
+```json
+{
+    "system": "...", // optional
+    "prompt": "...",
+    "chosen": [
+        {"role": "user", "content": "..."},
+        {"role": "assistant", "content": "..."}
+    ],
+    "rejected": [
+        {"role": "user", "content": "..."},
+        {"role": "assistant", "content": "..."}
+    ]
+}
+```
+
+#### zephyr.nectar
+
+```json
+{
+    "prompt": "...",
+    "answers": [
+        {
+            "answer": "...",
+            "rank": 1
+        },
+        {
+            "answer": "...",
+            "rank": 2
+        }
+        // ... more answers with ranks
+    ]
+}
+```
+
+#### chat_template.default
+
+```yaml
+rl: dpo
+datasets:
+  - path: ...
+    split: train
+    type: chat_template.default
+    field_messages: "messages"
+    field_chosen: "chosen"
+    field_rejected: "rejected"
+    message_property_mappings:
+      role: role
+      content: content
+    roles:
+      user: ["user"]
+      assistant: ["assistant"]
+      system: ["system"]
+```
+
+Sample input format:
+
+```json
+{
+    "messages": [
+        {
+            "role": "system",
+            "content": "..."
+        },
+        {
+            "role": "user",
+            "content": "..."
+        },
+        // ... more messages
+    ],
+    "chosen": {
+        "role": "assistant",
+        "content": "..."
+    },
+    "rejected": {
+        "role": "assistant",
+        "content": "..."
+    }
+}
+```
+
+#### user_defined.default
+
+For custom behaviors,
+
+```yaml
+rl: dpo
+datasets:
+  - path: ...
+    split: train
+    type: user_defined.default
+
+    field_prompt: "prompt"
+    field_system: "system"
+    field_chosen: "chosen"
+    field_rejected: "rejected"
+    prompt_format: "{prompt}"
+    chosen_format: "{chosen}"
+    rejected_format: "{rejected}"
+```
+
+The input format is a simple JSON input with customizable fields based on the above config.
+
+```json
+{
+    "system": "...",  // optional
+    "prompt": "...",
+    "chosen": "...",
+    "rejected": "..."
+}
+```
+
+### IPO
+
+As IPO is just DPO with a different loss function, all supported options for DPO works here.
+
 ```yaml
 rl: ipo
 ```
 
-#### ORPO
+### ORPO
 
 Paper: https://arxiv.org/abs/2403.07691
 
@@ -52,8 +318,28 @@ datasets:
     type: chat_template.argilla
 ```
 
+ORPO supports the following types with the following dataset format:
 
-#### KTO
+#### chat_template.argilla
+
+```json
+{
+    "system": "...",  // optional
+    "prompt": "...",  // if available, will be taken as user message for single-turn instead of from list below
+
+    // chosen/rejected should be same till last content and only even-number of alternating user/assistant turns
+    "chosen": [
+        {"role": "user", "content": "..."},
+        {"role": "assistant", "content": "..."}
+    ],
+    "rejected": [
+        {"role": "user", "content": "..."},
+        {"role": "assistant", "content": "..."}
+    ]
+}
+```
+
+### KTO
 
 ```yaml
 rl: kto
@@ -72,7 +358,186 @@ gradient_checkpointing_kwargs:
   use_reentrant: true
 ```
 
-#### Using local dataset files
+KTO supports the following types with the following dataset format:
+
+#### chatml.argilla
+
+```json
+{
+    "system": "...", // optional
+    "instruction": "...",
+    "completion": "..."
+}
+```
+
+#### chatml.argilla_chat
+
+```json
+{
+    "chosen": [
+        {"role": "user", "content": "..."}
+    ],
+    "completion": [
+        {"role": "assistant", "content": "..."}
+    ]
+}
+```
+
+#### chatml.intel
+
+```json
+{
+    "system": "...", // optional
+    "question": "...",
+    "completion": "..."
+}
+```
+
+#### chatml.prompt_pairs
+
+```json
+{
+    "system": "...", // optional
+    "prompt": "...",
+    "completion": "..."
+}
+```
+
+#### chatml.ultra
+
+```json
+{
+    "system": "...", // optional
+    "prompt": "...",
+    "completion": "..."
+}
+```
+
+#### llama3.argilla
+
+```json
+{
+    "system": "...", // optional
+    "instruction": "...",
+    "completion": "..."
+}
+```
+
+#### llama3.argilla_chat
+
+```json
+{
+    "completion": [
+        {"role": "user", "content": "..."},
+        {"role": "assistant", "content": "..."}
+    ]
+}
+```
+
+#### llama3.intel
+
+```json
+{
+    "system": "...", // optional
+    "question": "...",
+    "completion": "..."
+}
+```
+
+#### llama3.prompt_pairs
+
+```json
+{
+    "system": "...", // optional
+    "prompt": "...",
+    "completion": "..."
+}
+```
+
+#### llama3.ultra
+
+```json
+{
+    "system": "...", // optional
+    "prompt": "...",
+    "completion": "..."
+}
+```
+
+#### user_defined.default
+
+For custom behaviors,
+
+```yaml
+rl: kto
+datasets:
+  - path: ...
+    split: train
+    type: user_defined.default
+
+    field_prompt: "prompt"
+    field_system: "system"
+    field_completion: "completion"
+    field_label: "label"
+    prompt_format: "{prompt}"
+    completion_format: "{completion}"
+```
+
+The input format is a simple JSON input with customizable fields based on the above config.
+
+```json
+{
+    "system": "...",  // optional
+    "prompt": "...",
+    "completion": "...",
+    "label": "..."
+}
+```
+
+### GRPO
+
+GRPO uses custom reward functions and transformations. Please have them ready locally.
+
+For ex, to load OpenAI's GSM8K and use a random reward for completions:
+
+```python
+# rewards.py
+import random
+
+def rand_reward_func(completions, **kwargs) -> list[float]:
+    return [random.uniform(0, 1) for _ in completions]
+
+def oai_gsm8k_transform(cfg, *args, **kwargs):
+    def transform_fn(example, tokenizer=None):
+        label = example["answer"].split("####")[-1].strip().replace(",", "")
+        return {
+            "prompt": [{"role": "user", "content": example["question"]},],
+            "answer": label,
+        }
+    return transform_fn, {"remove_columns": ["question"]}
+```
+
+```yaml
+rl: grpo
+
+trl:
+    beta: 0.001
+    max_completion_length: 256
+    use_vllm: True
+    vllm_device: auto
+    vllm_gpu_memory_utilization: 0.15
+    num_generations: 4
+    reward_funcs: ["rewards.rand_reward_func"]    # format: '{file_name}.{fn_name}'
+datasets:
+  - path: openai/gsm8k
+    name: main
+    type: rewards.oai_gsm8k_transform  # format: '{file_name}.{fn_name}'
+```
+
+To see other examples of custom reward functions, please see [TRL GRPO Docs](https://github.com/huggingface/trl/blob/main/docs/source/grpo_trainer.md#using-a-custom-reward-function).
+
+### Using local dataset files
+
 ```yaml
 datasets:
   - ds_type: json
@@ -82,9 +547,9 @@ datasets:
     type: chatml.intel
 ```
 
-#### Trl autounwrap for peft
+### TRL auto-unwrapping for PEFT
 
-Trl supports autounwrapping peft models, so that a ref model does not need to be additionally loaded, leading to less VRAM needed. This is on by default. To turn it off, pass the following config.
+TRL supports auto-unwrapping PEFT models for RL training paradigms which rely on a reference model. This significantly reduces memory pressure as an additional refreference model does not need to be loaded, and reference model log-probabilities can be obtained by disabling PEFT adapters. This is enabled by default. To turn it off, pass the following config:
 
 ```yaml
 # load ref model when adapter training.

docs/torchao.qmd

@@ -3,6 +3,12 @@ title: "PyTorch ao"
 description: "Custom data types and layouts for training and inference"
 ---
 
+To use experimental optimizers (`AdamWFp8`, `AdamW4bit`, `AdamW8bit`) from Pytorch Ao, please install the package as shown below.
+
+::: {.callout-tip}
+Some experimental optimizers are already present in regular Pytorch, so please re-check if you actually need this package!
+:::
+
 ### Installation
 
 Stable Release from the PyTorch index

docs/unsloth.qmd

@@ -8,6 +8,12 @@ description: "Hyper-optimized QLoRA finetuning for single GPUs"
 Unsloth provides hand-written optimized kernels for LLM finetuning that slightly improve speed and VRAM over
 standard industry baselines.
 
+::: {.callout-important}
+Due to breaking changes in transformers `v4.48.0`, users will need to downgrade to `<=v4.47.1` to use this patch.
+
+This will later be deprecated in favor of [LoRA Optimizations](lora_optims.qmd).
+:::
+
 
 ### Installation
 
@@ -17,7 +23,7 @@ The following will install the correct unsloth and extras from source.
 python scripts/unsloth_install.py | sh
 ```
 
-### Using unsloth w Axolotl
+### Usage
 
 Axolotl exposes a few configuration options to try out unsloth and get most of the performance gains.
 

examples/deepseek-v2/qlora-fsdp-2_5.yaml

@@ -21,8 +21,9 @@ datasets:
     type: chat_template
     split: train[:20%]
     field_messages: conversations
-    message_field_role: from
-    message_field_content: value
+    message_property_mappings:
+      role: from
+      content: value
 
 dataset_prepared_path: last_run_prepared
 val_set_size: 0.0

examples/gemma2/qlora.yml

@@ -16,8 +16,9 @@ datasets:
     type: chat_template
     drop_system_message: true
     field_messages: conversations
-    message_field_role: from
-    message_field_content: value
+    message_property_mappings:
+      role: from
+      content: value
 
 val_set_size: 0.0
 output_dir: ./outputs/out

examples/jamba/qlora_fsdp_large.yaml

@@ -13,8 +13,9 @@ datasets:
     type: chat_template
     drop_system_message: true
     field_messages: conversations
-    message_field_role: from
-    message_field_content: value
+    message_property_mappings:
+      role: from
+      content: value
 
 dataset_prepared_path: last_run_prepared
 val_set_size: 0.0

examples/llama-3/fft-8b-liger-fsdp.yaml

@@ -17,8 +17,9 @@ datasets:
     type: chat_template
     split: train[:20%]
     field_messages: conversations
-    message_field_role: from
-    message_field_content: value
+    message_property_mappings:
+      role: from
+      content: value
 
 dataset_prepared_path: last_run_prepared
 val_set_size: 0.02

examples/llama-3/instruct-dpo-lora-8b.yml

@@ -17,8 +17,9 @@ datasets:
     field_messages: conversation
     field_chosen: chosen
     field_rejected: rejected
-    message_field_role: role
-    message_field_content: content
+    message_property_mappings:
+      role: role
+      content: content
     roles:
       system:
         - system

examples/llama-3/instruct-lora-8b.yml

@@ -14,8 +14,9 @@ datasets:
   - path: fozziethebeat/alpaca_messages_2k_test
     type: chat_template
     field_messages: messages
-    message_field_role: role
-    message_field_content: content
+    message_property_mappings:
+      role: role
+      content: content
     roles:
       user:
         - user

examples/llama-3/lora-1b-deduplicate-dpo.yml

@@ -17,8 +17,9 @@ datasets:
     field_messages: conversation
     field_chosen: chosen
     field_rejected: rejected
-    message_field_role: role
-    message_field_content: content
+    message_property_mappings:
+      role: role
+      content: content
     roles:
       system:
         - system
@@ -31,8 +32,9 @@ datasets:
     field_messages: conversation
     field_chosen: chosen
     field_rejected: rejected
-    message_field_role: role
-    message_field_content: content
+    message_property_mappings:
+      role: role
+      content: content
     roles:
       system:
         - system

examples/llama-3/lora-1b-kernels.yml

@@ -0,0 +1,82 @@
+base_model: NousResearch/Llama-3.2-1B
+# Automatically upload checkpoint and final model to HF
+# hub_model_id: username/custom_model_name
+
+load_in_8bit: false
+load_in_4bit: false
+strict: false
+
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.1
+output_dir: ./outputs/lora-out
+
+adapter: lora
+lora_model_dir:
+
+sequence_len: 2048
+sample_packing: true
+pad_to_sequence_len: true
+
+lora_r: 16
+lora_alpha: 32
+# Currently, we don't support dropout with our custom Triton kernels
+# lora_dropout: 0.05
+lora_fan_in_fan_out:
+lora_target_modules:
+  - gate_proj
+  - down_proj
+  - up_proj
+  - q_proj
+  - v_proj
+  - k_proj
+  - o_proj
+
+# These options enable our custom Triton kernels / autograd
+# functions for MLP and attention calculations
+lora_mlp_kernel: true
+lora_qkv_kernel: true
+lora_o_kernel: true
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 2
+micro_batch_size: 2
+num_epochs: 1
+optimizer: adamw_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: auto
+fp16:
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+loss_watchdog_threshold: 5.0
+loss_watchdog_patience: 3
+
+warmup_steps: 10
+evals_per_epoch: 4
+saves_per_epoch: 1
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:
+  pad_token: "<|end_of_text|>"

examples/mistral/mistral-dpo-qlora.yml

@@ -22,8 +22,9 @@ datasets:
     field_messages: conversation
     field_chosen: chosen
     field_rejected: rejected
-    message_field_role: role
-    message_field_content: content
+    message_property_mappings:
+      role: role
+      content: content
 
 dataset_prepared_path:
 val_set_size: 0.05

examples/phi/lora-3.5.yaml

@@ -14,8 +14,9 @@ datasets:
   - path: fozziethebeat/alpaca_messages_2k_test
     type: chat_template
     field_messages: messages
-    message_field_role: role
-    message_field_content: content
+    message_property_mappings:
+      role: role
+      content: content
     roles:
       user:
         - user

examples/qwen2/dpo.yaml

@@ -12,8 +12,9 @@ datasets:
     field_messages: conversation
     field_chosen: chosen
     field_rejected: rejected
-    message_field_role: role
-    message_field_content: content
+    message_property_mappings:
+      role: role
+      content: content
     roles:
       system:
         - system

index.qmd

@@ -1,7 +1,7 @@
 ---
-toc-location: right-body
-toc-title: Table Of Contents
-toc-expand: 2
+# toc-location: right-body
+# toc-title: Table Of Contents
+# toc-expand: 2
 ---
 
 ```{python}

requirements.txt

@@ -7,18 +7,18 @@ mamba-ssm==1.2.0.post1
 flash-attn==2.7.4.post1
 xformers>=0.0.23.post1
 autoawq==0.2.7.post3
-liger-kernel==0.5.2
+liger-kernel==0.5.3
 # END section
 
 packaging==23.2
 
 peft==0.14.0
-transformers==4.48.3
+transformers==4.49.0
 tokenizers>=0.21.0
 accelerate==1.3.0
 datasets==3.2.0
 deepspeed==0.16.1
-trl==0.13.0
+trl==0.15.1
 
 optimum==1.16.2
 hf_transfer
@@ -26,7 +26,7 @@ sentencepiece
 gradio==3.50.2
 
 modal==0.70.5
-pydantic==2.6.3
+pydantic==2.10.6
 addict
 fire
 PyYAML>=6.0

scripts/chat_datasets.py

@@ -31,27 +31,26 @@ def parse_dataset(dataset=None, split="train"):
     ds_cfg["field_messages"] = field_messages
 
     message_fields = features[field_messages][0].keys()
-    message_field_role = None
+
+    message_property_mappings = {"role": None, "content": None}
     for key in ["from", "role"]:
         if key in message_fields:
-            message_field_role = key
+            message_property_mappings["role"] = key
             break
-    if not message_field_role:
+    if not message_property_mappings["role"]:
         raise ValueError(
             f'No role field found in messages: {", ".join(message_fields)}'
         )
-    ds_cfg["message_field_role"] = message_field_role
 
-    message_field_content = None
     for key in ["content", "text", "value"]:
         if key in message_fields:
-            message_field_content = key
+            message_property_mappings["content"] = key
             break
-    if not message_field_content:
+    if not message_property_mappings["content"]:
         raise ValueError(
             f'No content field found in messages: {", ".join(message_fields)}'
         )
-    ds_cfg["message_field_content"] = message_field_content
+    ds_cfg["message_property_mappings"] = message_property_mappings
 
     print(yaml.dump({"datasets": [ds_cfg]}))
 

setup.py

@@ -79,7 +79,7 @@ def parse_requirements():
                 if patch == 0:
                     _install_requires.append("xformers==0.0.28.post2")
                 else:
-                    _install_requires.append("xformers==0.0.29")
+                    _install_requires.append("xformers>=0.0.28.post3")
                 _install_requires.pop(_install_requires.index(autoawq_version))
             elif (major, minor) >= (2, 4):
                 if patch == 0:
@@ -125,7 +125,7 @@ setup(
     },
     extras_require={
         "flash-attn": [
-            "flash-attn==2.7.0.post2",
+            "flash-attn==2.7.4.post1",
         ],
         "deepspeed": [
             "deepspeed==0.16.1",
@@ -156,5 +156,8 @@ setup(
         "ray": [
             "ray[train]",
         ],
+        "vllm": [
+            "vllm==0.7.2",
+        ],
     },
 )

src/axolotl/__init__.py

@@ -4,4 +4,4 @@ import pkgutil
 
 __path__ = pkgutil.extend_path(__path__, __name__)  # Make this a namespace package
 
-__version__ = "0.6.0"
+__version__ = "0.8.0.dev0"

src/axolotl/cli/cloud/__init__.py

@@ -35,13 +35,18 @@ def do_cli_train(
     cloud_config: Union[Path, str],
     config: Union[Path, str],
     accelerate: bool = True,
+    cwd=None,
+    **kwargs,
 ) -> None:
     print_axolotl_text_art()
     cloud_cfg = load_cloud_cfg(cloud_config)
     cloud = ModalCloud(cloud_cfg)
     with open(config, "r", encoding="utf-8") as file:
         config_yaml = file.read()
-    cloud.train(config_yaml, accelerate=accelerate)
+    local_dirs = {}
+    if cwd and not Path(cwd).joinpath("src", "axolotl").exists():
+        local_dirs = {"/workspace/mounts": cwd}
+    cloud.train(config_yaml, accelerate=accelerate, local_dirs=local_dirs, **kwargs)
 
 
 def do_cli_lm_eval(

src/axolotl/cli/cloud/modal_.py

@@ -7,6 +7,7 @@ import os
 import subprocess  # nosec B404
 from pathlib import Path
 from random import randint
+from typing import Optional
 
 import modal
 
@@ -22,8 +23,18 @@ def run_cmd(cmd: str, run_folder: str, volumes=None):
 
     # modal workaround so it doesn't use the automounted axolotl
     new_env = copy.deepcopy(os.environ)
+
     if "PYTHONPATH" in new_env:
-        del new_env["PYTHONPATH"]
+        paths = ["/workspace/mounts"]
+        for sub_python_path_str in new_env["PYTHONPATH"].split(":"):
+            sub_python_path = Path(sub_python_path_str)
+            if not sub_python_path.joinpath("src", "axolotl").exists():
+                # we don't want to use the automounted axolotl or unexpected behavior happens
+                paths.append(str(sub_python_path))
+        if paths:
+            new_env["PYTHONPATH"] = ":".join(paths)
+        else:
+            del new_env["PYTHONPATH"]
 
     # Propagate errors from subprocess.
     if exit_code := subprocess.call(  # nosec B603
@@ -112,8 +123,6 @@ class ModalCloud(Cloud):
         if env := self.get_env():
             image = image.env(env)
 
-        image = image.pip_install("fastapi==0.110.0", "pydantic==2.6.3")
-
         return image
 
     def get_secrets(self):
@@ -203,9 +212,12 @@ class ModalCloud(Cloud):
             memory = int(self.config.memory)
         return 1024 * memory
 
-    def get_train_env(self):
+    def get_train_env(self, local_dirs=None):
+        image = self.get_image()
+        for mount, local_dir in (local_dirs or {}).items():
+            image = image.add_local_dir(local_dir, mount)
         return self.app.function(
-            image=self.get_image(),
+            image=image,
             volumes={k: v[0] for k, v in self.volumes.items()},
             cpu=16.0,
             gpu=self.get_train_gpu(),
@@ -214,14 +226,21 @@ class ModalCloud(Cloud):
             secrets=self.get_secrets(),
         )
 
-    def train(self, config_yaml: str, accelerate: bool = True):
-        modal_fn = self.get_train_env()(_train)
+    def train(
+        self,
+        config_yaml: str,
+        accelerate: bool = True,
+        local_dirs: Optional[dict[str, str]] = None,
+        **kwargs,
+    ):
+        modal_fn = self.get_train_env(local_dirs)(_train)
         with modal.enable_output():
             with self.app.run(detach=True):
                 modal_fn.remote(
                     config_yaml,
                     accelerate=accelerate,
                     volumes={k: v[0] for k, v in self.volumes.items()},
+                    **kwargs,
                 )
 
     def lm_eval(self, config_yaml: str):
@@ -239,44 +258,41 @@ class ModalCloud(Cloud):
 
 
 def _preprocess(config_yaml: str, volumes=None):
-    Path("/workspace/artifacts/axolotl").mkdir(parents=True, exist_ok=True)
-    with open(
-        "/workspace/artifacts/axolotl/config.yaml", "w", encoding="utf-8"
-    ) as f_out:
+    Path("/workspace/mounts").mkdir(parents=True, exist_ok=True)
+    with open("/workspace/mounts/config.yaml", "w", encoding="utf-8") as f_out:
         f_out.write(config_yaml)
-    run_folder = "/workspace/artifacts/axolotl"
+    run_folder = "/workspace/mounts"
     run_cmd(
-        "axolotl preprocess /workspace/artifacts/axolotl/config.yaml --dataset-processes=8",
+        "axolotl preprocess /workspace/mounts/config.yaml --dataset-processes=8",
         run_folder,
         volumes,
     )
 
 
-def _train(config_yaml: str, accelerate: bool = True, volumes=None):
-    with open(
-        "/workspace/artifacts/axolotl/config.yaml", "w", encoding="utf-8"
-    ) as f_out:
+def _train(config_yaml: str, accelerate: bool = True, volumes=None, **kwargs):
+    with open("/workspace/mounts/config.yaml", "w", encoding="utf-8") as f_out:
         f_out.write(config_yaml)
-    run_folder = "/workspace/artifacts/axolotl"
+    run_folder = "/workspace/mounts"
     if accelerate:
         accelerate_args = "--accelerate"
     else:
         accelerate_args = "--no-accelerate"
+    num_processes_args = ""
+    if num_processes := kwargs.pop("num_processes", None):
+        num_processes_args = f"--num-processes {num_processes}"
     run_cmd(
-        f"axolotl train {accelerate_args} /workspace/artifacts/axolotl/config.yaml",
+        f"axolotl train {accelerate_args} {num_processes_args} /workspace/mounts/config.yaml",
         run_folder,
         volumes,
     )
 
 
 def _lm_eval(config_yaml: str, volumes=None):
-    with open(
-        "/workspace/artifacts/axolotl/config.yaml", "w", encoding="utf-8"
-    ) as f_out:
+    with open("/workspace/mounts/config.yaml", "w", encoding="utf-8") as f_out:
         f_out.write(config_yaml)
-    run_folder = "/workspace/artifacts/axolotl"
+    run_folder = "/workspace/mounts"
     run_cmd(
-        "axolotl lm-eval /workspace/artifacts/axolotl/config.yaml",
+        "axolotl lm-eval /workspace/mounts/config.yaml",
         run_folder,
         volumes,
     )

src/axolotl/cli/main.py

@@ -2,19 +2,19 @@
 # pylint: disable=redefined-outer-name
 
 import logging
-import random
+import os
 import subprocess  # nosec B404
 import tempfile
-from copy import deepcopy
-from itertools import product
 from pathlib import Path
 from typing import Optional
 
 import click
 import yaml
+from dotenv import load_dotenv
 
 import axolotl
 from axolotl.cli.args import EvaluateCliArgs, PreprocessCliArgs, TrainerCliArgs
+from axolotl.cli.sweeps import generate_sweep_configs
 from axolotl.cli.utils import (
     add_options_from_config,
     add_options_from_dataclass,
@@ -27,76 +27,6 @@ from axolotl.utils import set_pytorch_cuda_alloc_conf
 from axolotl.utils.config.models.input.v0_4_1 import AxolotlInputConfig
 
 
-def generate_sweep_configs(base_config, sweeps_config):
-    """
-    Recursively generates all possible configurations by applying sweeps to the base config.
-
-    Args:
-        base_config (dict): The original configuration dictionary
-        sweeps_config (dict): Dictionary where keys are parameters and values are either:
-            - lists of values to sweep independently
-            - or for paired values, a list of dicts under the '_' key
-
-    Returns:
-        list: List of all possible configuration dictionaries
-
-    Example:
-        sweeps_config = {
-            'learning_rate': [0.1, 0.01],
-            '_': [
-                {'load_in_8bit': True, 'adapter': 'lora'},
-                {'load_in_4bit': True, 'adapter': 'qlora'}
-            ]
-        }
-    """
-    # Separate paired values from regular sweeps
-    paired_values = sweeps_config.get("_", [])
-    regular_sweeps = {k: v for k, v in sweeps_config.items() if k != "_"}
-
-    # Process regular sweeps
-    param_names = list(regular_sweeps.keys())
-    param_values = list(regular_sweeps.values())
-
-    # Generate combinations for regular sweeps
-    regular_combinations = list(product(*param_values)) if param_values else [()]
-
-    # Combine regular sweeps with paired values
-    all_combinations = []
-    for reg_combo in regular_combinations:
-        if paired_values:
-            for paired_set in paired_values:
-                new_config = {}
-                # new_config = deepcopy(base_config)
-                # Combine regular parameters with paired parameters
-                full_combo = {**dict(zip(param_names, reg_combo)), **paired_set}
-                for param_name, param_value in full_combo.items():
-                    new_config[param_name] = param_value
-                print(new_config)
-                all_combinations.append(new_config)
-        else:
-            # If no paired values, just use regular combinations
-            # new_config = deepcopy(base_config)
-            new_config = {}
-            for param_name, param_value in zip(param_names, reg_combo):
-                new_config[param_name] = param_value
-            print(new_config)
-            all_combinations.append(new_config)
-
-    # randomize the order of trials
-    random.seed(42)
-    random.shuffle(all_combinations)
-
-    # Generate a new config for each combination
-    result_configs = []
-    for combination in all_combinations:
-        new_config = deepcopy(base_config)
-        for param_name, param_value in combination.items():
-            new_config[param_name] = param_value
-        result_configs.append(new_config)
-
-    return result_configs
-
-
 @click.group()
 @click.version_option(version=axolotl.__version__, prog_name="axolotl")
 def cli():
@@ -165,7 +95,6 @@ def train(
     """
     # Enable expandable segments for cuda allocation to improve VRAM usage
     set_pytorch_cuda_alloc_conf()
-    from axolotl.cli.cloud import do_cli_train
 
     if "use_ray" in kwargs and kwargs["use_ray"]:
         accelerate = False
@@ -199,7 +128,16 @@ def train(
         try:
             if accelerate:
                 if cloud:
-                    do_cli_train(cloud_config=cloud, config=config, accelerate=True)
+                    from axolotl.cli.cloud import do_cli_train
+
+                    cwd = os.getcwd()
+                    do_cli_train(
+                        cloud_config=cloud,
+                        config=config,
+                        accelerate=True,
+                        cwd=cwd,
+                        **kwargs,
+                    )
                 else:
                     accelerate_args = []
                     if "main_process_port" in kwargs:
@@ -208,7 +146,7 @@ def train(
                         accelerate_args.append(str(main_process_port))
                     if "num_processes" in kwargs:
                         num_processes = kwargs.pop("num_processes", None)
-                        accelerate_args.append("--num-processes")
+                        accelerate_args.append("--num_processes")
                         accelerate_args.append(str(num_processes))
 
                     base_cmd = ["accelerate", "launch"]
@@ -220,7 +158,11 @@ def train(
                     subprocess.run(cmd, check=True)  # nosec B603
             else:
                 if cloud:
-                    do_cli_train(cloud_config=cloud, config=config, accelerate=False)
+                    from axolotl.cli.cloud import do_cli_train
+
+                    do_cli_train(
+                        cloud_config=cloud, config=config, accelerate=False, **kwargs
+                    )
                 else:
                     from axolotl.cli.train import do_cli
 
@@ -381,4 +323,5 @@ def main():
 
 
 if __name__ == "__main__":
+    load_dotenv()
     main()

src/axolotl/cli/sweeps.py

@@ -0,0 +1,77 @@
+"""Utilities for handling sweeps over configs for axolotl train CLI command"""
+
+import random
+from copy import deepcopy
+from itertools import product
+
+
+def generate_sweep_configs(
+    base_config: dict[str, list], sweeps_config: dict[str, list]
+) -> list[dict[str, list]]:
+    """
+    Recursively generates all possible configurations by applying sweeps to the base config.
+
+    Args:
+        base_config (dict): The original configuration dictionary
+        sweeps_config (dict): Dictionary where keys are parameters and values are either:
+            - lists of values to sweep independently
+            - or for paired values, a list of dicts under the '_' key
+
+    Returns:
+        list: List of all possible configuration dictionaries
+
+    Example:
+        sweeps_config = {
+            'learning_rate': [0.1, 0.01],
+            '_': [
+                {'load_in_8bit': True, 'adapter': 'lora'},
+                {'load_in_4bit': True, 'adapter': 'qlora'}
+            ]
+        }
+    """
+    # Separate paired values from regular sweeps
+    paired_values = sweeps_config.get("_", [])
+    regular_sweeps = {k: v for k, v in sweeps_config.items() if k != "_"}
+
+    # Process regular sweeps
+    param_names = list(regular_sweeps.keys())
+    param_values = list(regular_sweeps.values())
+
+    # Generate combinations for regular sweeps
+    regular_combinations = list(product(*param_values)) if param_values else [()]
+
+    # Combine regular sweeps with paired values
+    all_combinations = []
+    for reg_combo in regular_combinations:
+        if paired_values:
+            for paired_set in paired_values:
+                new_config = {}
+                # new_config = deepcopy(base_config)
+                # Combine regular parameters with paired parameters
+                full_combo = {**dict(zip(param_names, reg_combo)), **paired_set}
+                for param_name, param_value in full_combo.items():
+                    new_config[param_name] = param_value
+                print(new_config)
+                all_combinations.append(new_config)
+        else:
+            # If no paired values, just use regular combinations
+            # new_config = deepcopy(base_config)
+            new_config = {}
+            for param_name, param_value in zip(param_names, reg_combo):
+                new_config[param_name] = param_value
+            print(new_config)
+            all_combinations.append(new_config)
+
+    # randomize the order of trials
+    random.seed(42)
+    random.shuffle(all_combinations)
+
+    # Generate a new config for each combination
+    result_configs = []
+    for combination in all_combinations:
+        new_config = deepcopy(base_config)
+        for param_name, param_value in combination.items():
+            new_config[param_name] = param_value
+        result_configs.append(new_config)
+
+    return result_configs

src/axolotl/common/datasets.py

@@ -122,9 +122,11 @@ def load_preference_datasets(
         `total_num_steps`.
     """
     train_dataset, eval_dataset = load_prepare_preference_datasets(cfg)
-    total_num_steps = int(
+    total_num_steps: Optional[int] = int(
         math.ceil(len(train_dataset) * cfg.num_epochs / cfg.batch_size)
     )
+    if cfg.rl == "grpo":
+        total_num_steps = None
 
     if cli_args.debug or cfg.debug:
         LOG.info("check_dataset_labels...")

src/axolotl/core/trainer_builder.py

@@ -39,7 +39,6 @@ from trl.trainer.utils import RewardDataCollatorWithPadding
 
 from axolotl.core.trainers.base import (
     AxolotlCPOTrainer,
-    AxolotlDPOTrainer,
     AxolotlKTOTrainer,
     AxolotlMambaTrainer,
     AxolotlORPOTrainer,
@@ -48,9 +47,11 @@ from axolotl.core.trainers.base import (
     AxolotlTrainer,
     ReLoRATrainer,
 )
+from axolotl.core.trainers.dpo import DPOStrategy
+from axolotl.core.trainers.dpo.args import AxolotlDPOConfig
+from axolotl.core.trainers.grpo import GRPOStrategy
 from axolotl.core.training_args import (
     AxolotlCPOConfig,
-    AxolotlDPOConfig,
     AxolotlKTOConfig,
     AxolotlORPOConfig,
     AxolotlPRMConfig,
@@ -329,6 +330,12 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
         )
 
         training_arguments_kwargs = {}
+
+        if self.cfg.include_tokens_per_second is not None:
+            training_arguments_kwargs[
+                "include_tokens_per_second"
+            ] = self.cfg.include_tokens_per_second
+
         if self.cfg.bf16 == "full":
             training_arguments_kwargs["bf16_full_eval"] = True
         else:
@@ -641,9 +648,6 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
                 tokenizer=self.tokenizer,
             )
 
-        if self.cfg.rl == "orpo":
-            training_arguments_kwargs["orpo_alpha"] = self.cfg.orpo_alpha
-
         if self.cfg.neftune_noise_alpha is not None:
             training_arguments_kwargs[
                 "neftune_noise_alpha"
@@ -652,7 +656,7 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
         trainer_kwargs = {}
 
         if self.cfg.reward_model:
-            trainer_kwargs["max_length"] = self.cfg.sequence_len
+            training_arguments_kwargs["max_length"] = self.cfg.sequence_len
 
         # pylint: disable=duplicate-code
         if self.cfg.optimizer in [
@@ -699,6 +703,9 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
                 "accelerator_config"
             ] = self.cfg.accelerator_config
 
+        if self.cfg.tensor_parallel:
+            training_arguments_kwargs["tp_size"] = torch.cuda.device_count()
+
         if self.cfg.kd_ce_alpha is not None:
             training_arguments_kwargs["kd_ce_alpha"] = self.cfg.kd_ce_alpha
         if self.cfg.kd_alpha is not None:
@@ -965,10 +972,11 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
             # default to saving each epoch if not defined
             training_args_kwargs["save_strategy"] = "epoch"
 
-        training_args_kwargs["dataset_num_proc"] = self.cfg.dataset_processes
+        if self.cfg.dataset_processes:
+            training_args_kwargs["dataset_num_proc"] = self.cfg.dataset_processes
 
-        if self.cfg.rl_beta:
-            training_args_kwargs["beta"] = self.cfg.rl_beta
+        if (self.cfg.trl and self.cfg.trl.beta) or self.cfg.rl_beta:
+            training_args_kwargs["beta"] = self.cfg.trl.beta or self.cfg.rl_beta
         if self.cfg.orpo_alpha:
             # trl does some odd mapping of alpha to beta to reuse the beta parameter ???
             training_args_kwargs["beta"] = self.cfg.orpo_alpha
@@ -977,6 +985,7 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
             training_args_kwargs["rpo_alpha"] = self.cfg.rpo_alpha
 
         training_args_cls = None
+        blocklist_args_kwargs = []
         if self.cfg.rl == "simpo":
             training_args_cls = AxolotlCPOConfig
             training_args_kwargs["loss_type"] = "simpo"
@@ -1001,11 +1010,15 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
                 self.cfg.kto_undesirable_weight or 1.0
             )
 
-            training_args_kwargs["dataset_num_proc"] = self.cfg.dataset_processes
             training_args_kwargs["max_length"] = self.cfg.sequence_len
             if self.cfg.max_prompt_len:
                 training_args_kwargs["max_prompt_length"] = self.cfg.max_prompt_len
 
+        elif self.cfg.rl == "grpo":
+            training_args_cls = GRPOStrategy.get_training_args_class()
+            training_args_kwargs.update(GRPOStrategy.set_training_args_kwargs(self.cfg))
+            blocklist_args_kwargs = GRPOStrategy.get_blocklist_args_kwargs()
+
         else:
             training_args_cls = AxolotlDPOConfig
             if self.cfg.rl == "ipo":
@@ -1016,11 +1029,21 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
             training_args_kwargs["generate_during_eval"] = self.cfg.use_wandb
             if self.cfg.dpo_use_weighting is not None:
                 training_args_kwargs["use_weighting"] = self.cfg.dpo_use_weighting
+            if self.cfg.dpo_use_logits_to_keep is not None:
+                training_args_kwargs[
+                    "use_logits_to_keep"
+                ] = self.cfg.dpo_use_logits_to_keep
 
+        for blocklist_key in blocklist_args_kwargs:
+            if blocklist_key in training_args_kwargs:
+                del training_args_kwargs[blocklist_key]
+
+        max_steps = self.cfg.max_steps or total_num_steps or -1
+        training_args_kwargs["num_train_epochs"] = self.cfg.num_epochs
         training_args = training_args_cls(  # pylint: disable=unexpected-keyword-arg
-            output_dir=self.cfg.output_dir,
+            self.cfg.output_dir,
             per_device_train_batch_size=self.cfg.micro_batch_size,
-            max_steps=self.cfg.max_steps or total_num_steps,
+            max_steps=max_steps,
             gradient_accumulation_steps=self.cfg.gradient_accumulation_steps,
             learning_rate=self.cfg.learning_rate,
             warmup_steps=self.cfg.warmup_steps,
@@ -1047,8 +1070,13 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
             dpo_trainer_kwargs[
                 "precompute_ref_log_probs"
             ] = self.cfg.precompute_ref_log_probs
-        if self.cfg.rl in ["dpo", "ipo"]:
-            trainer_cls = AxolotlDPOTrainer
+        if self.cfg.rl == "grpo":
+            trainer_cls = GRPOStrategy.get_trainer_class()
+            trainer_cls_args = [self.model]
+            trainer_cls_args.extend(GRPOStrategy.set_trainer_args(self.cfg))
+            dpo_trainer_kwargs.update(GRPOStrategy.set_trainer_kwargs(self.cfg))
+        elif self.cfg.rl in ["dpo", "ipo"]:
+            trainer_cls = DPOStrategy.get_trainer_class()
             trainer_cls_args = [self.model, self.model_ref]
         elif self.cfg.rl == "orpo":
             trainer_cls = AxolotlORPOTrainer
@@ -1063,12 +1091,14 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
             raise ValueError(f"Unsupported RL: {self.cfg.rl}")
 
         sig = inspect.signature(trainer_cls)
-        if "processing_class" in sig.parameters.keys():
-            dpo_trainer_kwargs["processing_class"] = self.tokenizer
-        else:
+        if "tokenizer" in sig.parameters.keys():
             dpo_trainer_kwargs["tokenizer"] = self.tokenizer
+        else:
+            dpo_trainer_kwargs["processing_class"] = self.tokenizer
 
-        if self.cfg.datasets is not None and (trainer_cls is AxolotlDPOTrainer):
+        if self.cfg.datasets is not None and (
+            trainer_cls is DPOStrategy.get_trainer_class()
+        ):
             dpo_trainer_kwargs["dataset_tags"] = [
                 d["path"] for d in self.cfg.datasets if not Path(d["path"]).is_dir()
             ]

src/axolotl/core/trainers/base.py

@@ -5,30 +5,21 @@ module for customized trainers
 from __future__ import annotations
 
 # pylint: disable=too-many-lines
-import gc
 import logging
 import os
 from collections import defaultdict
 from functools import wraps
-from typing import Any, Dict, Literal, Optional, Union
+from typing import Dict, Literal, Optional
 
 import torch
 from datasets import Dataset
 from peft.optimizers import create_loraplus_optimizer
-from torch import nn
 from torch.optim.lr_scheduler import OneCycleLR
 from torch.utils.data import BatchSampler, DataLoader, RandomSampler, SequentialSampler
 from transformers import Trainer
 from transformers.trainer_utils import PREFIX_CHECKPOINT_DIR, seed_worker
 from transformers.utils import is_sagemaker_mp_enabled
-from trl import (
-    CPOTrainer,
-    DPOTrainer,
-    KTOTrainer,
-    ORPOTrainer,
-    PRMTrainer,
-    RewardTrainer,
-)
+from trl import CPOTrainer, KTOTrainer, ORPOTrainer, PRMTrainer, RewardTrainer
 from trl.trainer.utils import pad_to_length
 
 from axolotl.monkeypatch.relora import ReLoRAScheduler
@@ -847,107 +838,6 @@ class ReLoRATrainer(AxolotlTrainer):
         return self.lr_scheduler
 
 
-class AxolotlDPOTrainer(SchedulerMixin, DPOTrainer):
-    """
-    Extend the base DPOTrainer for axolotl helpers
-    """
-
-    tag_names = ["axolotl", "dpo"]
-
-    def __init__(self, *args, dataset_tags=None, **kwargs):
-        super().__init__(*args, **kwargs)
-        self.dataset_tags = dataset_tags
-        self.optimizer = None
-        self.model_accepts_loss_kwargs = False
-
-    def create_optimizer(self):
-        if self.args.loraplus_lr_ratio is None:
-            return super().create_optimizer()
-
-        opt_model = self.model_wrapped if is_sagemaker_mp_enabled() else self.model
-        if self.optimizer is None:  # pylint: disable=access-member-before-definition
-            optimizer_cls, optimizer_kwargs = Trainer.get_optimizer_cls_and_kwargs(
-                self.args,
-                opt_model,
-            )
-
-            loraplus_lr_ratio = getattr(self.args, "loraplus_lr_ratio", None)
-            if loraplus_lr_ratio:
-                print("Using lora+")
-            loraplus_lr_embedding = getattr(self.args, "loraplus_lr_embedding", None)
-            self.optimizer = create_loraplus_optimizer(  # pylint: disable=attribute-defined-outside-init
-                opt_model,
-                optimizer_cls,
-                loraplus_lr_ratio=loraplus_lr_ratio,
-                loraplus_lr_embedding=loraplus_lr_embedding,
-                **optimizer_kwargs,
-            )
-
-        if is_sagemaker_mp_enabled():
-            self.optimizer = smp.DistributedOptimizer(  # pylint: disable=attribute-defined-outside-init
-                self.optimizer
-            )
-
-        return self.optimizer
-
-    @wraps(DPOTrainer.push_to_hub)
-    def push_to_hub(self, *args, **kwargs) -> str:
-        """
-        Overwrite the `push_to_hub` method in order to force-add the tags when pushing the
-        model on the Hub. Please refer to `~transformers.Trainer.push_to_hub` for more details.
-        """
-        kwargs = _sanitize_kwargs_for_ds_tagging(
-            dataset_tags=self.dataset_tags, kwargs=kwargs
-        )
-        kwargs = _sanitize_kwargs_for_tagging(tag_names=self.tag_names, kwargs=kwargs)
-
-        return super().push_to_hub(*args, **kwargs)
-
-    @staticmethod
-    def tokenize_row(
-        features,
-        processing_class,
-        max_prompt_length,
-        max_completion_length,
-        add_special_tokens,
-    ) -> Dict:
-        res = DPOTrainer.tokenize_row(
-            features,
-            processing_class,
-            max_prompt_length,
-            max_completion_length,
-            add_special_tokens,
-        )
-        # fix when the tokenizer doesn't have a bos_token_id, e.g. Qwen
-        if processing_class.bos_token is None and res["prompt_input_ids"][0] is None:
-            for key in res.keys():
-                res[key] = res[key][1:]
-
-        if processing_class.bos_token and processing_class.bos_token_id is not None:
-            # dpo trainer may incorrectly prepend the bos_token_id to the dpo outputs
-            if res["chosen_input_ids"][0] == processing_class.bos_token_id:
-                res["chosen_input_ids"] = res["chosen_input_ids"][1:]
-                res["chosen_labels"] = res["chosen_labels"][1:]
-                res["chosen_attention_mask"] = res["chosen_attention_mask"][1:]
-            if res["rejected_input_ids"][0] == processing_class.bos_token_id:
-                res["rejected_input_ids"] = res["rejected_input_ids"][1:]
-                res["rejected_labels"] = res["rejected_labels"][1:]
-                res["rejected_attention_mask"] = res["rejected_attention_mask"][1:]
-
-        return res
-
-    def training_step(
-        self,
-        model: nn.Module,
-        inputs: Dict[str, Union[torch.Tensor, Any]],
-        num_items_in_batch=None,
-    ) -> torch.Tensor:
-        loss: torch.Tensor = super().training_step(model, inputs, num_items_in_batch)
-        gc.collect()
-        torch.cuda.empty_cache()
-        return loss
-
-
 class AxolotlORPOTrainer(SchedulerMixin, ORPOTrainer):
     """
     Extend the base ORPOTrainer for axolotl helpers

src/axolotl/core/trainers/dpo/__init__.py

@@ -0,0 +1,33 @@
+"""
+DPO Specific Strategy for training
+"""
+from axolotl.core.trainers.dpo.trainer import AxolotlDPOTrainer
+
+
+class DPOStrategy:
+    """
+    Strategy for DPO training
+    """
+
+    @classmethod
+    def get_trainer_class(cls):
+        return AxolotlDPOTrainer
+
+    @classmethod
+    def get_training_args_class(cls):
+        from axolotl.core.trainers.dpo.args import AxolotlDPOConfig
+
+        return AxolotlDPOConfig
+
+    @classmethod
+    def set_training_args_kwargs(cls, cfg):
+        training_args_kwargs = {}
+        if cfg.rl == "ipo":
+            training_args_kwargs["loss_type"] = "ipo"
+        training_args_kwargs["max_length"] = cfg.sequence_len
+        training_args_kwargs["max_completion_length"] = None
+        training_args_kwargs["max_prompt_length"] = cfg.sequence_len
+        training_args_kwargs["generate_during_eval"] = cfg.use_wandb
+        if cfg.dpo_use_weighting is not None:
+            training_args_kwargs["use_weighting"] = cfg.dpo_use_weighting
+        return training_args_kwargs

src/axolotl/core/trainers/dpo/args.py

@@ -0,0 +1,15 @@
+"""
+Axolotl specific DPO args
+"""
+from dataclasses import dataclass
+
+from trl import DPOConfig
+
+from axolotl.core.training_args import AxolotlTrainingMixins
+
+
+@dataclass
+class AxolotlDPOConfig(AxolotlTrainingMixins, DPOConfig):
+    """
+    DPO config for DPO training
+    """

src/axolotl/core/trainers/dpo/trainer.py

@@ -0,0 +1,125 @@
+"""
+DPO trainer for axolotl
+"""
+import gc
+from functools import wraps
+from typing import Any, Dict, Union
+
+import torch
+from peft.optimizers import create_loraplus_optimizer
+from torch import nn
+from transformers import Trainer
+from transformers.utils import is_sagemaker_mp_enabled
+from trl import DPOTrainer
+
+from axolotl.core.trainers.base import (
+    SchedulerMixin,
+    _sanitize_kwargs_for_ds_tagging,
+    _sanitize_kwargs_for_tagging,
+)
+
+if is_sagemaker_mp_enabled():
+    import smdistributed.modelparallel.torch as smp
+
+
+class AxolotlDPOTrainer(SchedulerMixin, DPOTrainer):
+    """
+    Extend the base DPOTrainer for axolotl helpers
+    """
+
+    tag_names = ["axolotl", "dpo"]
+
+    def __init__(self, *args, dataset_tags=None, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.dataset_tags = dataset_tags
+        self.optimizer = None
+        self.model_accepts_loss_kwargs = False
+
+    def create_optimizer(self):
+        # pylint: disable=duplicate-code
+        if self.args.loraplus_lr_ratio is None:
+            return super().create_optimizer()
+
+        opt_model = self.model_wrapped if is_sagemaker_mp_enabled() else self.model
+        if self.optimizer is None:  # pylint: disable=access-member-before-definition
+            optimizer_cls, optimizer_kwargs = Trainer.get_optimizer_cls_and_kwargs(
+                self.args,
+                opt_model,
+            )
+
+            loraplus_lr_ratio = getattr(self.args, "loraplus_lr_ratio", None)
+            if loraplus_lr_ratio:
+                print("Using lora+")
+            loraplus_lr_embedding = getattr(self.args, "loraplus_lr_embedding", None)
+            # pylint: disable=duplicate-code
+            self.optimizer = create_loraplus_optimizer(  # pylint: disable=attribute-defined-outside-init
+                opt_model,
+                optimizer_cls,
+                loraplus_lr_ratio=loraplus_lr_ratio,
+                loraplus_lr_embedding=loraplus_lr_embedding,
+                **optimizer_kwargs,
+            )
+
+        if is_sagemaker_mp_enabled():
+            self.optimizer = smp.DistributedOptimizer(  # pylint: disable=attribute-defined-outside-init
+                self.optimizer
+            )
+
+        return self.optimizer
+
+    @wraps(DPOTrainer.push_to_hub)
+    def push_to_hub(self, *args, **kwargs) -> str:
+        """
+        Overwrite the `push_to_hub` method in order to force-add the tags when pushing the
+        model on the Hub. Please refer to `~transformers.Trainer.push_to_hub` for more details.
+        """
+        kwargs = _sanitize_kwargs_for_ds_tagging(
+            dataset_tags=self.dataset_tags, kwargs=kwargs
+        )
+        kwargs = _sanitize_kwargs_for_tagging(tag_names=self.tag_names, kwargs=kwargs)
+
+        return super().push_to_hub(*args, **kwargs)
+
+    @staticmethod
+    def tokenize_row(
+        features,
+        processing_class,
+        max_prompt_length,
+        max_completion_length,
+        add_special_tokens,
+    ) -> Dict:
+        res = DPOTrainer.tokenize_row(
+            features,
+            processing_class,
+            max_prompt_length,
+            max_completion_length,
+            add_special_tokens,
+        )
+        # fix when the tokenizer doesn't have a bos_token_id, e.g. Qwen
+        if processing_class.bos_token is None and res["prompt_input_ids"][0] is None:
+            for key in res.keys():
+                res[key] = res[key][1:]
+
+        if processing_class.bos_token and processing_class.bos_token_id is not None:
+            # dpo trainer may incorrectly prepend the bos_token_id to the dpo outputs
+            if res["chosen_input_ids"][0] == processing_class.bos_token_id:
+                res["chosen_input_ids"] = res["chosen_input_ids"][1:]
+                res["chosen_labels"] = res["chosen_labels"][1:]
+                res["chosen_attention_mask"] = res["chosen_attention_mask"][1:]
+            if res["rejected_input_ids"][0] == processing_class.bos_token_id:
+                res["rejected_input_ids"] = res["rejected_input_ids"][1:]
+                res["rejected_labels"] = res["rejected_labels"][1:]
+                res["rejected_attention_mask"] = res["rejected_attention_mask"][1:]
+
+        return res
+
+    def training_step(
+        self,
+        model: nn.Module,
+        inputs: Dict[str, Union[torch.Tensor, Any]],
+        num_items_in_batch=None,
+    ) -> torch.Tensor:
+        loss: torch.Tensor = super().training_step(model, inputs, num_items_in_batch)
+        gc.collect()
+        torch.cuda.empty_cache()
+        return loss

src/axolotl/core/trainers/grpo/__init__.py

@@ -0,0 +1,119 @@
+"""
+GRPO Specific Strategy for training
+"""
+
+import importlib
+import inspect
+import logging
+
+from trl.trainer.grpo_trainer import RewardFunc
+
+from axolotl.core.trainers.grpo.trainer import AxolotlGRPOTrainer
+
+LOG = logging.getLogger("axolotl")
+
+
+class GRPOStrategy:
+    """
+    Strategy for GRPO training
+    """
+
+    @classmethod
+    def get_trainer_class(cls):
+        return AxolotlGRPOTrainer
+
+    @classmethod
+    def get_training_args_class(cls):
+        from axolotl.core.trainers.grpo.args import AxolotlGRPOConfig
+
+        return AxolotlGRPOConfig
+
+    @classmethod
+    def set_training_args_kwargs(cls, cfg):
+        grpo_args_kwargs = {}
+        if cfg.trl and cfg.trl.use_vllm:
+            grpo_args_kwargs["use_vllm"] = cfg.trl.use_vllm
+            if cfg.trl and cfg.trl.vllm_device:
+                grpo_args_kwargs["vllm_device"] = cfg.trl.vllm_device
+            else:
+                grpo_args_kwargs["vllm_device"] = "auto"
+            if cfg.trl and cfg.trl.vllm_gpu_memory_utilization:
+                grpo_args_kwargs[
+                    "vllm_gpu_memory_utilization"
+                ] = cfg.trl.vllm_gpu_memory_utilization
+            if cfg.trl and cfg.trl.vllm_max_model_len:
+                grpo_args_kwargs["vllm_max_model_len"] = cfg.trl.vllm_max_model_len
+        if cfg.trl and cfg.trl.num_generations:
+            grpo_args_kwargs["num_generations"] = cfg.trl.num_generations
+        if cfg.trl and cfg.trl.sync_ref_model:
+            grpo_args_kwargs["sync_ref_model"] = cfg.trl.sync_ref_model
+            if cfg.trl and cfg.trl.ref_model_mixup_alpha:
+                grpo_args_kwargs[
+                    "ref_model_mixup_alpha"
+                ] = cfg.trl.ref_model_mixup_alpha
+            if cfg.trl and cfg.trl.ref_model_sync_steps:
+                grpo_args_kwargs["ref_model_sync_steps"] = cfg.trl.ref_model_sync_steps
+        grpo_args_kwargs["max_completion_length"] = cfg.trl.max_completion_length
+        grpo_args_kwargs["log_completions"] = cfg.trl.log_completions
+        return grpo_args_kwargs
+
+    @classmethod
+    def set_trainer_args(cls, cfg):
+        trainer_args = []
+        if cfg.trl and cfg.trl.reward_funcs:
+            reward_funcs = []
+            for reward_func_fqn in cfg.trl.reward_funcs:
+                reward_funcs.append(cls.get_reward_func(reward_func_fqn))
+            trainer_args.append(reward_funcs)
+        return trainer_args
+
+    @classmethod
+    def set_trainer_kwargs(cls, cfg):
+        trainer_kwargs = {}
+        if cfg.trl and cfg.trl.reward_processing_classes:
+            trainer_kwargs[
+                "reward_processing_classes"
+            ] = cfg.trl.reward_processing_classes
+        return trainer_kwargs
+
+    @classmethod
+    def get_collator(cls, *args, **kwargs):  # pylint: disable=unused-argument
+        # No data collation is needed in GRPO, handled by trl's trainer __init__
+        return None
+
+    @classmethod
+    def get_blocklist_args_kwargs(cls):
+        return ["dataset_num_proc"]
+
+    @classmethod
+    def get_reward_func(cls, reward_func_fqn: str) -> RewardFunc:
+        """
+        Returns the reward function from the given fully qualified name, or the path to the reward function model.
+
+        Args:
+            reward_func_fqn (str): Fully qualified name of the reward function (e.g. r1_grpo.gsm8k_transform),
+                or a HF hub path to the reward model.
+        Raises:
+            ValueError: If the reward function does not accept at least two arguments.
+
+        Returns:
+            RewardFunc: A callable that accepts prompts and completions and returns rewards,
+                or a path to a reward model.
+
+        """
+        try:
+            # use importlib to dynamically load the reward function from the module
+            reward_func_module_name = reward_func_fqn.split(".")[-1]
+            reward_func_module = importlib.import_module(reward_func_fqn.split(".")[-2])
+            reward_func = getattr(reward_func_module, reward_func_module_name)
+            if not len(inspect.signature(reward_func).parameters) >= 2:
+                raise ValueError(
+                    "Reward function must accept at least two arguments: prompts: list and completions: list"
+                )
+            return reward_func
+        except ModuleNotFoundError:
+            # the user has passed a string (ideally indicating the path of a reward model)
+            LOG.info(
+                f"Reward function {reward_func_fqn} is a pre-trained model path - if this is unexpected, please check the reward function path."
+            )
+            return reward_func

src/axolotl/core/trainers/grpo/args.py

@@ -0,0 +1,15 @@
+"""
+Axolotl Specific Training Args
+"""
+from dataclasses import dataclass
+
+from trl import GRPOConfig
+
+from axolotl.core.training_args import AxolotlTrainingMixins
+
+
+@dataclass
+class AxolotlGRPOConfig(AxolotlTrainingMixins, GRPOConfig):
+    """
+    Axolotl GRPO Config for GRPO training
+    """

src/axolotl/core/trainers/grpo/trainer.py

@@ -0,0 +1,108 @@
+"""
+Axolotl GRPO trainer
+"""
+from accelerate.utils import is_peft_model
+from accelerate.utils.other import is_compiled_module
+from transformers import PreTrainedModel
+from trl import GRPOConfig, GRPOTrainer
+from trl.models import unwrap_model_for_generation
+
+from axolotl.core.trainers.base import SchedulerMixin
+
+
+# mypy: ignore-errors
+class AxolotlGRPOTrainer(SchedulerMixin, GRPOTrainer):
+    """
+    Extend the base GRPOTrainer for axolotl helpers
+    """
+
+    _tag_names = ["trl", "grpo", "axolotl"]
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        # pylint: disable=access-member-before-definition
+        # Enable gradient checkpointing if requested
+        if kwargs["args"].gradient_checkpointing:
+            # Ensure use_cache is disabled
+            if hasattr(self.model, "config"):
+                self.model.config.use_cache = False
+
+            # Enable gradient checkpointing on the base model for PEFT
+            if is_peft_model(self.model) and hasattr(
+                self.model.base_model, "gradient_checkpointing_enable"
+            ):
+                self.model.base_model.gradient_checkpointing_enable()
+            # Enable gradient checkpointing for non-PEFT models
+            elif hasattr(self.model, "gradient_checkpointing_enable"):
+                self.model.gradient_checkpointing_enable()
+            self.model = self._enable_gradient_checkpointing(self.model, kwargs["args"])
+        # pylint: enable=access-member-before-definition
+
+    def _enable_gradient_checkpointing(
+        self, model: PreTrainedModel, args: GRPOConfig
+    ) -> PreTrainedModel:
+        """Enables gradient checkpointing for the model."""
+        # pylint: disable=unused-argument,redefined-builtin
+        gradient_checkpointing_kwargs = args.gradient_checkpointing_kwargs or {}
+        use_reentrant = (
+            "use_reentrant" not in gradient_checkpointing_kwargs
+            or gradient_checkpointing_kwargs["use_reentrant"]
+        )
+
+        if use_reentrant:
+            if hasattr(model, "enable_input_require_grads"):
+                model.enable_input_require_grads()
+            else:
+
+                def make_inputs_require_grad(module, input, output):
+                    output.requires_grad_(True)
+
+                model.get_input_embeddings().register_forward_hook(
+                    make_inputs_require_grad
+                )
+
+        return model
+        # pylint: enable=unused-argument,redefined-builtin
+
+    def _move_model_to_vllm(self):
+        with unwrap_model_for_generation(
+            self.model,
+            self.accelerator,
+            gather_deepspeed3_params=self.args.ds3_gather_for_generation,
+        ) as unwrapped_model:
+            if is_compiled_module(unwrapped_model):
+                unwrapped_model = (
+                    unwrapped_model._orig_mod  # pylint: disable=protected-access
+                )
+            if is_peft_model(unwrapped_model):
+                unwrapped_model.merge_adapter()
+                state_dict = unwrapped_model.state_dict()
+                # Remove base_model and base_layer prefixes
+                state_dict = {
+                    k.removeprefix("base_model.model.")
+                    .removeprefix("base_model.model.")
+                    .replace(".base_layer", ""): v
+                    for k, v in state_dict.items()
+                }
+                # Remove values with adapter prefix (example: "_lora")
+                state_dict = {
+                    k: v
+                    for k, v in state_dict.items()
+                    if unwrapped_model.prefix not in k
+                }
+                # When module to save, remove its prefix and discard the original module
+                state_dict = {
+                    k.replace("modules_to_save.default.", ""): v
+                    for k, v in state_dict.items()
+                    if "original_module" not in k
+                }
+            else:
+                state_dict = unwrapped_model.state_dict()
+            if self.accelerator.is_main_process:
+                llm_model = (
+                    self.llm.llm_engine.model_executor.driver_worker.model_runner.model
+                )
+                llm_model.load_weights(state_dict.items())
+            if is_peft_model(unwrapped_model):
+                unwrapped_model.unmerge_adapter()

src/axolotl/core/training_args.py

@@ -5,7 +5,7 @@ from dataclasses import dataclass, field
 from typing import Optional
 
 from transformers import TrainingArguments
-from trl import CPOConfig, DPOConfig, KTOConfig, ORPOConfig, PRMConfig, RewardConfig
+from trl import CPOConfig, KTOConfig, ORPOConfig, PRMConfig, RewardConfig
 
 
 @dataclass
@@ -217,13 +217,6 @@ class AxolotlTrainingArguments(AxolotlTrainingMixins, TrainingArguments):
     """
 
 
-@dataclass
-class AxolotlDPOConfig(AxolotlTrainingMixins, DPOConfig):
-    """
-    DPO config for DPO training
-    """
-
-
 @dataclass
 class AxolotlORPOConfig(AxolotlTrainingMixins, ORPOConfig):
     """

src/axolotl/integrations/cut_cross_entropy/README.md

@@ -1,6 +1,10 @@
 # Cut Cross Entropy
 
-### Usage
+Cut Cross Entropy reduces VRAM usage through optimization on the cross-entropy operation during loss calculation.
+
+See https://github.com/apple/ml-cross-entropy
+
+## Usage
 
 ```yaml
 plugins:
@@ -8,3 +12,19 @@ plugins:
 
 cut_cross_entropy: true
 ```
+
+## Citation
+
+```bib
+@article{wijmans2024cut,
+  author       = {Erik Wijmans and
+                  Brody Huval and
+                  Alexander Hertzberg and
+                  Vladlen Koltun and
+                  Philipp Kr\"ahenb\"uhl},
+  title        = {Cut Your Losses in Large-Vocabulary Language Models},
+  journal      = {arXiv},
+  year         = {2024},
+  url          = {https://arxiv.org/abs/2411.09009},
+}
+```

src/axolotl/integrations/grokfast/README.md

@@ -2,7 +2,7 @@
 
 See https://github.com/ironjr/grokfast
 
-### Usage
+## Usage
 
 ```yaml
 plugins:
@@ -11,3 +11,14 @@ plugins:
 grokfast_alpha: 2.0
 grokfast_lamb: 0.98
 ```
+
+## Citation
+
+```bib
+@article{lee2024grokfast,
+    title={{Grokfast}: Accelerated Grokking by Amplifying Slow Gradients},
+    author={Lee, Jaerin and Kang, Bong Gyun and Kim, Kihoon and Lee, Kyoung Mu},
+    journal={arXiv preprint arXiv:2405.20233},
+    year={2024}
+}
+```

src/axolotl/integrations/kd/README.md

@@ -0,0 +1,23 @@
+# Knowledge Distillation
+
+## Usage
+
+```yaml
+plugins:
+  - "axolotl.integrations.kd.KDPlugin"
+
+kd_trainer: True
+kd_ce_alpha: 0.1
+kd_alpha: 0.9
+kd_temperature: 1.0
+
+torch_compile: True  # torch>=2.5.1, recommended to reduce vram
+
+datasets:
+  - path: ...
+    type: "axolotl.integrations.kd.chat_template"
+    field_messages: "messages_combined"
+    logprobs_field: "llm_text_generation_vllm_logprobs"  # for kd only, field of logprobs
+```
+
+An example dataset can be found at [`axolotl-ai-co/evolkit-logprobs-pipeline-75k-v2-sample`](https://huggingface.co/datasets/axolotl-ai-co/evolkit-logprobs-pipeline-75k-v2-sample)

src/axolotl/integrations/kd/topk_logprob/LICENSE.md

@@ -1,58 +0,0 @@
-### AXOLOTL COMMUNITY LICENSE AGREEMENT
-
-This Axolotl Community License Agreement (“Agreement”) is entered into by and between Axolotl AI Corp. (“Axolotl”) and
-any individual or entity (“Licensee”) who wishes to use the Software (as defined below) in accordance with the terms
-and conditions set forth in this Agreement.
-
-1.  Definitions
-    1.1 “Licensee” refers to any individual or entity who has obtained a copy of the Software under this Agreement.
-    1.2 “Plugin Integration” means independent integration software modules which may or may not be offered by Axolotl,
-        which may be licensed separately by their respective  authors and/or licensors.
-    1.3 “Software” refers to the specific sub-directory of the Axolotl, Inc. software located at
-        https://github.com/axolotl-ai-cloud/axolotl/tree/main/src/axolotl/integrations and its subdirectories which
-        permits Plugin Integrations to integrate with the Axolotl service.
-2.  Grant of License
-    2.1	Axolotl hereby grants Licensee a worldwide, non-exclusive, royalty-free, license to use, copy, modify, merge,
-        publish, distribute, sublicense, and/or otherwise exploit the Software, subject to the following conditions:
-        - Licensee must comply with all the terms and conditions of this Agreement.
-        - Licensee must include the original copyright notice and disclaimer of warranty in all copies or substantial
-          portions of the Software.
-    2.2 Licensee may use the Software for any lawful purpose, except as restricted in Section 3.
-3.  Restrictions
-    3.1 Licensee shall not use the Software for any activity that constitutes a commercial activity of offering for
-        free or for sale any services, platform, or equivalent  to third parties for the purposes of allowing such
-        third parties to fine-tune artificial intelligence models.
-    3.2 Licensee shall not:
-        - Use the Software for any illegal or unauthorized purpose.
-        - Reverse engineer, decompile, or disassemble the Software.
-        - Remove or modify any copyright, trademark, or other proprietary notices contained in the Software.
-        - Use the Software in a way that could damage, disable, overburden, or impair the functionality of the
-          Software or interfere with any third-party use of the Software.
-    3.3 Axolotl reserves the right to restrict certain Plugin Integrations for use with the Software. To the extent Licensee integrates a permitted, applicable Plugin Integration with the Software, Licensee shall comply with any additional terms and conditions imposed by the licensors of such Plugin Integration for use of such Plugin Integrations. Licensee shall contact Axolotl if it has questions about whether its use of the Software falls beyond the scope of this Agreement.
-4.  Intellectual Property Rights
-    4.1 Axolotl and its contributors retain all intellectual property rights in and to the Software. Licensee
-        acknowledges that this Agreement does not transfer any ownership rights or intellectual property rights to
-        Licensee.
-5.  Disclaimer of Warranty
-    5.1 THE SOFTWARE IS PROVIDED “AS IS,” WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED
-        TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND NON-INFRINGEMENT. IN NO EVENT SHALL
-        THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES, OR OTHER LIABILITY, WHETHER IN AN ACTION OF
-        CONTRACT, TORT, OR OTHERWISE, ARISING FROM, OUT OF, OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
-        DEALINGS IN THE SOFTWARE.
-6.  Termination
-    6.1 Axolotl may terminate this Agreement at any time if Licensee fails to comply with any of the terms and
-        conditions set forth herein. Upon termination, Licensee shall cease all use of the Software and destroy any
-        copies in its possession.
-7.  Governing Law
-    7.1 This Agreement shall be governed by and construed in accordance with the laws of the State of California,
-        without regards to conflicts of laws provisions thereof.
-8.  Entire Agreement
-    8.1 This Agreement constitutes the entire agreement between Axolotl and Licensee with respect to the subject matter
-        hereof and supersedes all prior or contemporaneous understandings or agreements between the parties concerning
-        the Software, whether written or oral. Axolotl may update the terms of this Agreement from time to time, and
-        Licensee’s continued use of the Software after any such updates shall constitute acceptance of updated terms
-        on a go-forward basis.  Axolotl will use commercially reasonable efforts to provide Licensee notice of any
-        material updates. By using the Software, Licensee acknowledges that it has read, understood, and agrees to be
-        bound by the terms and conditions of this Agreement.
-
-This Agreement was last updated on August 23, 2024.

src/axolotl/integrations/kd/topk_logprob/forward_kl.py

@@ -1,14 +1,16 @@
 # Copyright 2024 Axolotl AI. All rights reserved.
 #
-# This software may be used and distributed according to
-# the terms of the Axolotl Community License Agreement (the "License");
+# 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.
+# 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.
 
 """
 loss for top_k KL divergence

src/axolotl/integrations/liger/README.md

@@ -0,0 +1,36 @@
+# Liger Kernel Integration
+
+Liger Kernel provides efficient Triton kernels for LLM training, offering:
+
+- 20% increase in multi-GPU training throughput
+- 60% reduction in memory usage
+- Compatibility with both FSDP and DeepSpeed
+
+See https://github.com/linkedin/Liger-Kernel
+
+## Usage
+
+```yaml
+plugins:
+  - axolotl.integrations.liger.LigerPlugin
+liger_rope: true
+liger_rms_norm: true
+liger_glu_activation: true
+liger_layer_norm: true
+liger_fused_linear_cross_entropy: true
+```
+
+## Citation
+
+```bib
+@article{hsu2024ligerkernelefficienttriton,
+      title={Liger Kernel: Efficient Triton Kernels for LLM Training},
+      author={Pin-Lun Hsu and Yun Dai and Vignesh Kothapalli and Qingquan Song and Shao Tang and Siyu Zhu and Steven Shimizu and Shivam Sahni and Haowen Ning and Yanning Chen},
+      year={2024},
+      eprint={2410.10989},
+      archivePrefix={arXiv},
+      primaryClass={cs.LG},
+      url={https://arxiv.org/abs/2410.10989},
+      journal={arXiv preprint arXiv:2410.10989},
+}
+```

src/axolotl/integrations/lm_eval/README.md

@@ -1,6 +1,10 @@
 # LM Eval Harness
 
-### Usage
+Run evaluation on model using the popular lm-evaluation-harness library.
+
+See https://github.com/EleutherAI/lm-evaluation-harness
+
+## Usage
 
 ```yaml
 plugins:
@@ -10,4 +14,22 @@ lm_eval_tasks:
   - gsm8k
   - hellaswag
   - arc_easy
+
+lm_eval_batch_size: # Batch size for evaluation
+output_dir: # Directory to save evaluation results
+```
+
+## Citation
+
+```bib
+@misc{eval-harness,
+  author       = {Gao, Leo and Tow, Jonathan and Abbasi, Baber and Biderman, Stella and Black, Sid and DiPofi, Anthony and Foster, Charles and Golding, Laurence and Hsu, Jeffrey and Le Noac'h, Alain and Li, Haonan and McDonell, Kyle and Muennighoff, Niklas and Ociepa, Chris and Phang, Jason and Reynolds, Laria and Schoelkopf, Hailey and Skowron, Aviya and Sutawika, Lintang and Tang, Eric and Thite, Anish and Wang, Ben and Wang, Kevin and Zou, Andy},
+  title        = {A framework for few-shot language model evaluation},
+  month        = 07,
+  year         = 2024,
+  publisher    = {Zenodo},
+  version      = {v0.4.3},
+  doi          = {10.5281/zenodo.12608602},
+  url          = {https://zenodo.org/records/12608602}
+}
 ```

src/axolotl/integrations/spectrum/README.md

@@ -1,15 +1,17 @@
-## Spectrum: Targeted Training on Signal to Noise Ratio
+# Spectrum: Targeted Training on Signal to Noise Ratio
 
 by Eric Hartford, Lucas Atkins, Fernando Fernandes, David Golchinfar
 
 This plugin contains code to freeze the bottom fraction of modules in a model, based on the Signal-to-Noise Ratio (SNR).
 
-### Overview
+See https://github.com/cognitivecomputations/spectrum
+
+## Overview
 
 Spectrum is a tool for scanning and evaluating the Signal-to-Noise Ratio (SNR) of layers in large language models.
 By identifying the top n% of layers with the highest SNR, you can optimize training efficiency.
 
-### Usage
+## Usage
 
 ```yaml
 plugins:
@@ -19,3 +21,17 @@ spectrum_top_fraction: 0.5
 # Optional if using a pre-scanned model as your base_model. Useful if using a model mirror
 spectrum_model_name: meta-llama/Meta-Llama-3.1-8B
 ```
+
+## Citation
+
+```bib
+@misc{hartford2024spectrumtargetedtrainingsignal,
+      title={Spectrum: Targeted Training on Signal to Noise Ratio},
+      author={Eric Hartford and Lucas Atkins and Fernando Fernandes Neto and David Golchinfar},
+      year={2024},
+      eprint={2406.06623},
+      archivePrefix={arXiv},
+      primaryClass={cs.LG},
+      url={https://arxiv.org/abs/2406.06623},
+}
+```

src/axolotl/kernels/geglu.py

@@ -0,0 +1,159 @@
+"""
+Module for definition of GEGLU Triton kernels.
+
+See "GLU Variants Improve Transformer" (https://arxiv.org/abs/2002.05202).
+
+Credit to `unsloth` (https://unsloth.ai/) for inspiration for this implementation.
+"""
+# pylint: disable=invalid-name,unnecessary-lambda-assignment,duplicate-code
+
+import torch
+import triton
+import triton.language as tl
+
+SQRT_2_PI: tl.constexpr = 0.7978845608028654  # sqrt(2/π)
+
+
+@triton.jit
+def _geglu_fwd_kernel(
+    gate_ptr,
+    up_ptr,
+    out_ptr,
+    n_elements,
+    BLOCK_SIZE: tl.constexpr,
+):
+    """GEGLU forward kernel.
+
+    Args:
+        gate_ptr: Pointer to gate tensor [*, hidden_dim].
+        up_ptr: Pointer to up-projection tensor [*, hidden_dim].
+        out_ptr: Pointer to output tensor [*, hidden_dim].
+        n_elements: Total number of elements in the input tensors.
+        BLOCK_SIZE: Size of thread blocks for parallel computation.
+    """
+    block_idx = tl.program_id(0)
+    offsets = block_idx * BLOCK_SIZE + tl.arange(0, BLOCK_SIZE)
+    mask = offsets < n_elements
+
+    gate = tl.load(gate_ptr + offsets, mask=mask, other=0).to(tl.float32)
+    up = tl.load(up_ptr + offsets, mask=mask, other=0)
+
+    # Compute activation in fp32 then convert back
+    gelu_gate = 0.5 * gate * (tl.math.erf(tl.math.rsqrt(2.0) * gate) + 1.0)
+    gelu_gate = gelu_gate.to(up.dtype)
+    result = gelu_gate * up
+
+    tl.store(out_ptr + offsets, result, mask=mask)
+
+
+def geglu_forward(gate: torch.Tensor, up: torch.Tensor) -> torch.Tensor:
+    """GEGLU forward pass.
+
+    Args:
+        gate: Input gate tensor of shape [batch, seq_len, hidden_dim].
+        up: Up-projection tensor of shape [batch, seq_len, hidden_dim].
+
+    Returns:
+        torch.Tensor: Output tensor of shape [batch, seq_len, hidden_dim].
+    """
+    batch, seq_len, hidden_dim = gate.shape
+    n_elements = gate.numel()
+    out = torch.empty((batch, seq_len, hidden_dim), dtype=gate.dtype, device="cuda")
+
+    grid = lambda meta: (triton.cdiv(n_elements, meta["BLOCK_SIZE"]),)  # noqa: E731
+    _geglu_fwd_kernel[grid](
+        gate_ptr=gate,
+        up_ptr=up,
+        out_ptr=out,
+        n_elements=n_elements,
+        BLOCK_SIZE=1024,
+    )
+    return out
+
+
+@triton.jit
+def _geglu_bwd_kernel(
+    grad_out_ptr,
+    gate_ptr,
+    up_ptr,
+    n_elements,
+    BLOCK_SIZE: tl.constexpr,
+):
+    """GEGLU backward kernel. Stores gradient results in-place.
+
+    Args:
+        grad_out_ptr: Pointer to gradient output tensor [*, hidden_dim].
+        gate_ptr: Pointer to gate tensor [*, hidden_dim].
+        up_ptr: Pointer to up-projection tensor [*, hidden_dim].
+        n_elements: Total number of elements in the input tensors.
+        BLOCK_SIZE: Size of thread blocks for parallel computation.
+
+    Note:
+        After kernel execution, tensors are modified in-place:
+        - `grad_out_ptr` contains GEGLU activation output (`h`)
+        - `gate_ptr` contains gradient w.r.t gate (`grad_gate`)
+        - `up_ptr` contains gradient w.r.t up (`grad_up`)
+    """
+    block_idx = tl.program_id(0)
+    offsets = block_idx * BLOCK_SIZE + tl.arange(0, BLOCK_SIZE)
+    mask = offsets < n_elements
+
+    grad_out = tl.load(grad_out_ptr + offsets, mask=mask, other=0)
+    gate = tl.load(gate_ptr + offsets, mask=mask, other=0).to(tl.float32)
+    up = tl.load(up_ptr + offsets, mask=mask, other=0)
+
+    # Forward pass
+    gelu_partial = 0.5 * (tl.math.erf(tl.math.rsqrt(2.0) * gate) + 1.0)
+    gelu_gate = gelu_partial * gate
+    gelu_gate = gelu_gate.to(grad_out.dtype)
+
+    # Forward output
+    h = gelu_gate * up
+
+    # Compute gradients
+    grad_up = grad_out * gelu_gate
+
+    # Compute gate gradient using GELU derivative
+    temp = grad_out * up
+    t = 0.3989422804014327  # 1/sqrt(2*pi)
+    dgelu_dgate = gelu_partial + t * gate * tl.exp(-0.5 * gate * gate)
+    grad_gate = temp.to(tl.float32) * dgelu_dgate
+    grad_gate = grad_gate.to(grad_out.dtype)
+
+    # Store results
+    tl.store(grad_out_ptr + offsets, h, mask=mask)
+    tl.store(gate_ptr + offsets, grad_gate, mask=mask)
+    tl.store(up_ptr + offsets, grad_up, mask=mask)
+
+
+def geglu_backward(
+    grad_output: torch.Tensor, gate: torch.Tensor, up: torch.Tensor
+) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+    """GEGLU backward pass using in-place operations.
+
+    Args:
+        grad_output: Gradient of loss with respect to output, shape `[batch, seq_len, hidden_dim]`.
+        gate: Gate tensor from forward pass, shape `[batch, seq_len, hidden_dim]`.
+        up: Up-projection tensor from forward pass, shape `[batch, seq_len, hidden_dim]`.
+
+    Returns:
+        Tuple containing:
+            - GEGLU activation output (`h`)
+            - Gradient with respect to gate (`grad_gate`)
+            - Gradient with respect to up (`grad_up`)
+
+    Note:
+        This function modifies its input tensors in-place to store results.
+    """
+    n_elements = grad_output.numel()
+
+    grid = lambda meta: (triton.cdiv(n_elements, meta["BLOCK_SIZE"]),)  # noqa: E731
+    _geglu_bwd_kernel[grid](
+        grad_out_ptr=grad_output,
+        gate_ptr=gate,
+        up_ptr=up,
+        n_elements=n_elements,
+        BLOCK_SIZE=1024,
+    )
+
+    return grad_output, gate, up

src/axolotl/kernels/lora.py

@@ -0,0 +1,779 @@
+"""
+Module for definition of Low-Rank Adaptation (LoRA) Triton kernels.
+
+See "LoRA: Low-Rank Adaptation of Large Language Models"
+(https://arxiv.org/abs/2106.09685).
+
+Credit to `unsloth` (https://unsloth.ai/) for inspiration for this implementation.
+"""
+# pylint: disable=invalid-name
+
+from typing import Callable
+
+import torch
+from bitsandbytes.functional import QuantState
+from torch import nn
+
+from .geglu import geglu_backward, geglu_forward
+from .quantize import dequantize
+from .swiglu import swiglu_backward, swiglu_forward
+from .utils import torch_amp_custom_bwd, torch_amp_custom_fwd
+
+
+def get_lora_parameters(
+    proj: nn.Module,
+) -> tuple[
+    torch.Tensor,
+    QuantState | None,
+    torch.Tensor | None,
+    torch.Tensor | None,
+    float | None,
+]:
+    """
+    Gets LoRA parameters from a projection module.
+
+    Args:
+        proj: The projection module to extract parameters from.
+
+    Returns:
+        A tuple containing the base weight matrix, quantization state, LoRA A matrix,
+        LoRA B matrix, and scaling factor. States and matrices may be None if not
+        available.
+    """
+    # For DPO or disabled adapters
+    base_layer = proj.base_layer if hasattr(proj, "base_layer") else proj
+    W = base_layer.weight
+
+    if not hasattr(proj, "disable_adapters") or proj.disable_adapters or proj.merged:
+        quant_state = getattr(W, "quant_state", None)
+        return W, quant_state, None, None, None
+
+    active_adapter = (
+        proj.active_adapters[0]
+        if hasattr(proj, "active_adapters")
+        else proj.active_adapter
+    )
+    A = proj.lora_A[active_adapter].weight
+    B = proj.lora_B[active_adapter].weight
+    s = proj.scaling[active_adapter]
+
+    quant_state = getattr(W, "quant_state", None)
+
+    return W, quant_state, A, B, s
+
+
+def matmul_lora(
+    X: torch.Tensor,
+    W: torch.Tensor,
+    W_quant: QuantState,
+    A: torch.Tensor,
+    B: torch.Tensor,
+    s: float,
+    out: torch.Tensor | None = None,
+) -> torch.Tensor:
+    """
+    Efficient fused matmul + LoRA computation.
+
+    Args:
+        X: Input tensor [*, in_features]
+        W: Base weight matrix [out_features, in_features]
+        W_quant: Quantization state for W
+        A: LoRA A matrix [rank, in_features]
+        B: LoRA B matrix [out_features, rank]
+        s: LoRA scaling factor
+        out: Optional output tensor for inplace operations
+
+    Returns:
+        Result of X @ W + X @ A @ B
+    """
+    dtype = X.dtype
+    W = dequantize(W.t(), W_quant)
+
+    if X.dim() == 3:
+        batch, seq_len, _ = X.shape
+        X = X.view(-1, X.shape[-1])
+        reshape = True
+    else:
+        reshape = False
+
+    out = torch.matmul(X, W, out=out)
+    if W_quant is not None:
+        del W
+
+    if A is not None:
+        A, B = A.t(), B.t()
+        out += (X @ A.to(dtype)) @ (s * B.to(dtype))
+
+    return out.view(batch, seq_len, -1) if reshape else out
+
+
+class LoRA_MLP(torch.autograd.Function):
+    """Optimized LoRA MLP implementation."""
+
+    @staticmethod
+    @torch_amp_custom_fwd
+    def forward(
+        ctx,
+        X: torch.Tensor,
+        gate_weight: torch.Tensor,
+        gate_quant: object | None,
+        gate_A: torch.Tensor | None,
+        gate_B: torch.Tensor | None,
+        gate_scale: float,
+        up_weight: torch.Tensor,
+        up_quant: object | None,
+        up_A: torch.Tensor | None,
+        up_B: torch.Tensor | None,
+        up_scale: float,
+        down_weight: torch.Tensor,
+        down_quant: object | None,
+        down_A: torch.Tensor | None,
+        down_B: torch.Tensor | None,
+        down_scale: float,
+        activation_fn: Callable,
+        activation_fn_backward: Callable,
+        inplace: bool | None = True,
+    ) -> torch.Tensor:
+        """
+        Forward pass for LoRA MLP.
+
+        Args:
+            ctx: Autograd context
+            X: Input features
+            gate_weight: Gate projection weight
+            gate_quant: Gate quantization state
+            gate_A: Gate LoRA A matrix
+            gate_B: Gate LoRA B matrix
+            gate_scale: Gate LoRA scale
+            up_weight: Up-projection weight
+            up_quant: Up-projection quantization state
+            up_A: Up-projection LoRA A matrix
+            up_B: Up-projection LoRA B matrix
+            up_scale: Up-projection LoRA scale
+            down_weight: Down-projection weight
+            down_quant: Down-projection quantization state
+            down_A: Down-projection LoRA A matrix
+            down_B: Down-projection LoRA B matrix
+            down_scale: Down-projection LoRA scale
+            activation_fn: Forward activation function
+            activation_fn_backward: Backward activation function
+            inplace: Whether to perform operations in-place
+
+        Returns:
+            Output transformed by multi-layer perceptron and activation function
+        """
+        # Compute projections
+        gate = matmul_lora(X, gate_weight, gate_quant, gate_A, gate_B, gate_scale)
+        up = matmul_lora(X, up_weight, up_quant, up_A, up_B, up_scale)
+
+        # Activation
+        hidden = activation_fn(gate, up)
+
+        # Down projection
+        output = matmul_lora(
+            hidden, down_weight, down_quant, down_A, down_B, down_scale
+        )
+
+        # Save for backward
+        ctx.save_for_backward(X, gate, up, gate_A, gate_B, up_A, up_B, down_A, down_B)
+        ctx.scales = (gate_scale, up_scale, down_scale)
+        ctx.quants = (gate_quant, up_quant, down_quant)
+        ctx.weights = (gate_weight, up_weight, down_weight)
+        ctx.activation_fn = activation_fn
+        ctx.activation_fn_backward = activation_fn_backward
+        ctx.inplace = inplace
+
+        return output
+
+    @staticmethod
+    @torch_amp_custom_bwd
+    def backward(
+        ctx: torch.autograd.function.FunctionCtx,
+        grad_output: torch.Tensor,
+    ) -> tuple[
+        torch.Tensor | None,
+        None,
+        None,
+        torch.Tensor | None,
+        torch.Tensor | None,
+        None,
+        None,
+        None,
+        torch.Tensor | None,
+        torch.Tensor | None,
+        None,
+        None,
+        None,
+        torch.Tensor | None,
+        torch.Tensor | None,
+        None,
+        None,
+        None,
+        None,
+    ]:
+        """
+        Performs backward pass computation for LoRA MLP.
+
+        Args:
+            ctx: Context object storing tensors saved during forward pass
+            grad_output: Gradient of loss with respect to layer output
+
+        Returns:
+            Tuple containing gradients for all inputs from forward pass:
+            - Input gradient tensor (or `None`)
+            - `None` for weights/quantization states
+            - LoRA A/B matrix gradients (or `None`)
+            - `None` for scaling factors
+            - `None` for activation functions and flags
+        """
+        (
+            X,
+            gate,
+            up,
+            gate_A,
+            gate_B,
+            up_A,
+            up_B,
+            down_A,
+            down_B,
+        ) = ctx.saved_tensors
+        gate_scale, up_scale, down_scale = ctx.scales
+        gate_quant, up_quant, down_quant = ctx.quants
+        gate_weight, up_weight, down_weight = ctx.weights
+
+        # Transpose all LoRA matrices
+        gate_A, gate_B = (
+            gate_A.t() if gate_A is not None else None,
+            gate_B.t() if gate_B is not None else None,
+        )
+        up_A, up_B = (
+            up_A.t() if up_A is not None else None,
+            up_B.t() if up_B is not None else None,
+        )
+        down_A, down_B = (
+            down_A.t() if down_A is not None else None,
+            down_B.t() if down_B is not None else None,
+        )
+
+        # Reshape inputs
+        batch, seq_len, hd = X.shape
+        grad_output = grad_output.view(-1, grad_output.shape[-1])
+        X = X.view(-1, X.shape[-1])
+        gate = gate.view(-1, gate.shape[-1])
+        up = up.view(-1, up.shape[-1])
+        dtype = X.dtype
+
+        # Down projection
+        DW = matmul_lora(
+            grad_output,
+            down_weight.t(),
+            down_quant,
+            down_B,
+            down_A,
+            down_scale,
+        )
+
+        # Activation backward
+        h, grad_gate, grad_up = ctx.activation_fn_backward(DW, gate, up)
+
+        # Initialize and compute LoRA gradients
+        d_down_A = d_down_B = d_up_A = d_up_B = d_gate_A = d_gate_B = None
+
+        if down_A is not None:
+            d_down_A = h.t() @ (grad_output @ down_B.t())
+            d_down_B = (down_A.t() @ h.t()) @ grad_output
+            d_down_A *= down_scale
+            d_down_B *= down_scale
+
+        if up_A is not None:
+            d_up_A = X.t() @ (grad_up @ up_B.t())
+            d_up_B = (up_A.t() @ X.t()) @ grad_up
+            d_up_A *= up_scale
+            d_up_B *= up_scale
+
+        if gate_A is not None:
+            d_gate_A = X.t() @ (grad_gate @ gate_B.t())
+            d_gate_B = (gate_A.t() @ X.t()) @ grad_gate
+            d_gate_A *= gate_scale
+            d_gate_B *= gate_scale
+
+        # Compute input gradients
+        dX = torch.zeros_like(X) if ctx.needs_input_grad[0] else None
+
+        if dX is not None:
+            # Up projection gradients
+            up_weight = dequantize(up_weight.t(), up_quant)
+            if ctx.inplace:
+                dX = torch.matmul(grad_up, up_weight.t(), out=X)
+            else:
+                dX = torch.matmul(grad_up, up_weight.t())
+            del up_weight
+
+            # Note the .to(dtype) only where mixing LoRA with base weights
+            if up_A is not None:
+                dX += grad_up @ up_B.to(dtype).t() @ (up_scale * up_A.to(dtype).t())
+
+            # Gate projection gradients
+            gate_weight = dequantize(gate_weight.t(), gate_quant)
+            dX += grad_gate @ gate_weight.t()
+            del gate_weight
+
+            if gate_A is not None:
+                dX += (
+                    grad_gate
+                    @ gate_B.to(dtype).t()
+                    @ (gate_scale * gate_A.to(dtype).t())
+                )
+
+            # Reshape back
+            dX = dX.view(batch, seq_len, hd)
+
+        # Return gradients in correct order matching forward inputs
+        return (
+            dX,
+            None,
+            None,
+            d_gate_A.t() if d_gate_A is not None else None,
+            d_gate_B.t() if d_gate_B is not None else None,
+            None,
+            None,
+            None,
+            d_up_A.t() if d_up_A is not None else None,
+            d_up_B.t() if d_up_B is not None else None,
+            None,
+            None,
+            None,
+            d_down_A.t() if d_down_A is not None else None,
+            d_down_B.t() if d_down_B is not None else None,
+            None,
+            None,
+            None,
+            None,
+        )
+
+
+def apply_lora_mlp_swiglu(self, X: torch.Tensor, inplace: bool = True) -> torch.Tensor:
+    """
+    Applies LoRA to MLP layer with SwiGLU activation.
+
+    Args:
+        X: Input tensor for the MLP layer
+        inplace: Whether to perform operations in-place to save memory
+
+    Returns:
+        Output tensor after applying LoRA-adapted MLP with SwiGLU activation
+    """
+    gateW, gateW_quant, gateA, gateB, gateS = get_lora_parameters(self.gate_proj)
+    upW, upW_quant, upA, upB, upS = get_lora_parameters(self.up_proj)
+    downW, downW_quant, downA, downB, downS = get_lora_parameters(self.down_proj)
+
+    out = LoRA_MLP.apply(
+        X,
+        gateW,
+        gateW_quant,
+        gateA,
+        gateB,
+        gateS,
+        upW,
+        upW_quant,
+        upA,
+        upB,
+        upS,
+        downW,
+        downW_quant,
+        downA,
+        downB,
+        downS,
+        swiglu_forward,
+        swiglu_backward,
+        inplace,
+    )
+
+    return out
+
+
+def apply_lora_mlp_geglu(self, X: torch.Tensor, inplace: bool = True) -> torch.Tensor:
+    """
+    Applies LoRA to MLP layer with GEGLU activation.
+
+    Args:
+        X: Input tensor for the MLP layer
+        inplace: Whether to perform operations in-place to save memory
+
+    Returns:
+        Output tensor after applying LoRA-adapted MLP with GEGLU activation
+    """
+    gateW, gateW_quant, gateA, gateB, gateS = get_lora_parameters(self.gate_proj)
+    upW, upW_quant, upA, upB, upS = get_lora_parameters(self.up_proj)
+    downW, downW_quant, downA, downB, downS = get_lora_parameters(self.down_proj)
+    out = LoRA_MLP.apply(
+        X,
+        gateW,
+        gateW_quant,
+        gateA,
+        gateB,
+        gateS,
+        upW,
+        upW_quant,
+        upA,
+        upB,
+        upS,
+        downW,
+        downW_quant,
+        downA,
+        downB,
+        downS,
+        geglu_forward,
+        geglu_backward,
+        inplace,
+    )
+
+    return out
+
+
+class LoRA_QKV(torch.autograd.Function):
+    """
+    Optimized LoRA QKV implementation with quantization support.
+
+    Implements efficient computation of query, key, value projections with LoRA,
+    supporting quantization and memory optimization.
+    """
+
+    @staticmethod
+    @torch_amp_custom_fwd
+    def forward(
+        ctx: torch.autograd.function.FunctionCtx,
+        X: torch.Tensor,
+        q_weight: torch.Tensor,
+        q_quant: QuantState | None,
+        q_A: torch.Tensor | None,
+        q_B: torch.Tensor | None,
+        q_scale: float,
+        k_weight: torch.Tensor,
+        k_quant: QuantState | None,
+        k_A: torch.Tensor | None,
+        k_B: torch.Tensor | None,
+        k_scale: float,
+        v_weight: torch.Tensor,
+        v_quant: QuantState | None,
+        v_A: torch.Tensor | None,
+        v_B: torch.Tensor | None,
+        v_scale: float,
+        inplace: bool = True,
+    ) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        """
+        Forward pass computing Q, K, V projections with LoRA.
+
+        Args:
+            ctx: Autograd context
+            X: Input tensor
+            q_weight: Query projection weight
+            q_quant: Query quantization state
+            q_A: Query LoRA A matrix
+            q_B: Query LoRA B matrix
+            q_scale: Query LoRA scale
+            k_weight: Key projection weight
+            k_quant: Key quantization state
+            k_A: Key LoRA A matrix
+            k_B: Key LoRA B matrix
+            k_scale: Key LoRA scale
+            v_weight: Value projection weight
+            v_quant: Value quantization state
+            v_A: Value LoRA A matrix
+            v_B: Value LoRA B matrix
+            v_scale: Value LoRA scale
+            inplace: Whether to perform operations in-place
+
+        Returns:
+            Tuple of (Query, Key, Value) projection tensors
+        """
+        Q = matmul_lora(X, q_weight, q_quant, q_A, q_B, q_scale)
+        K = matmul_lora(X, k_weight, k_quant, k_A, k_B, k_scale)
+        V = matmul_lora(X, v_weight, v_quant, v_A, v_B, v_scale)
+
+        ctx.save_for_backward(X, q_A, q_B, k_A, k_B, v_A, v_B)
+        ctx.scales = (q_scale, k_scale, v_scale)
+        ctx.quants = (q_quant, k_quant, v_quant)
+        ctx.weights = (q_weight, k_weight, v_weight)
+        ctx.inplace = inplace
+
+        return Q, K, V
+
+    @staticmethod
+    @torch_amp_custom_fwd
+    def backward(
+        ctx: torch.autograd.function.FunctionCtx,
+        q_grad: torch.Tensor,
+        k_grad: torch.Tensor,
+        v_grad: torch.Tensor,
+    ) -> tuple[
+        torch.Tensor,
+        None,
+        None,
+        torch.Tensor | None,
+        torch.Tensor | None,
+        None,
+        None,
+        None,
+        torch.Tensor | None,
+        torch.Tensor | None,
+        None,
+        None,
+        None,
+        torch.Tensor | None,
+        torch.Tensor | None,
+        None,
+        None,
+    ]:
+        """
+        Backward pass computing gradients for LoRA QKV.
+
+        Args:
+            ctx: Autograd context
+            q_grad: Gradient for query projection
+            k_grad: Gradient for key projection
+            v_grad: Gradient for value projection
+
+        Returns:
+            Tuple containing gradients for all forward inputs
+        """
+        X, A_q, B_q, A_k, B_k, A_v, B_v = ctx.saved_tensors
+        q_weight, k_weight, v_weight = ctx.weights
+        q_quant, k_quant, v_quant = ctx.quants
+        q_scale, k_scale, v_scale = ctx.scales
+        dtype = X.dtype
+
+        # Reshape gradients
+        batch, seq_len = X.shape[:2]
+        q_grad = q_grad.view(-1, q_grad.shape[-1])
+        k_grad = k_grad.reshape(-1, k_grad.shape[-1])
+        v_grad = v_grad.view(-1, v_grad.shape[-1])
+        X = X.view(-1, X.shape[-1])
+
+        # Pre-transpose X once
+        X_t = X.t()
+
+        # Initialize LoRA gradients as None
+        d_A_q = d_B_q = d_A_k = d_B_k = d_A_v = d_B_v = None
+
+        # Compute q path LoRA gradients if adapters exist
+        if A_q is not None and B_q is not None:
+            A_q_scaled = (q_scale * A_q).to(dtype)
+            B_q_scaled = B_q.to(dtype)
+            d_A_q = torch.mm(X_t, torch.mm(q_grad, B_q_scaled))
+            d_B_q = torch.mm(torch.mm(A_q_scaled, X_t), q_grad)
+
+        # Compute k path LoRA gradients if adapters exist
+        if A_k is not None and B_k is not None:
+            A_k_scaled = (k_scale * A_k).to(dtype)
+            B_k_scaled = B_k.to(dtype)
+            d_A_k = torch.mm(X_t, torch.mm(k_grad, B_k_scaled))
+            d_B_k = torch.mm(torch.mm(A_k_scaled, X_t), k_grad)
+
+        # Compute v path LoRA gradients if adapters exist
+        if A_v is not None and B_v is not None:
+            A_v_scaled = (v_scale * A_v).to(dtype)
+            B_v_scaled = B_v.to(dtype)
+            d_A_v = torch.mm(X_t, torch.mm(v_grad, B_v_scaled))
+            d_B_v = torch.mm(torch.mm(A_v_scaled, X_t), v_grad)
+
+        # Compute input gradient, reusing X memory if possible
+        out_buffer = X if ctx.inplace else None
+
+        # Q path
+        q_weight_t = dequantize(q_weight, q_quant)
+        grad_X = torch.mm(q_grad, q_weight_t, out=out_buffer)
+        del q_weight
+        del q_weight_t
+        if A_q is not None and B_q is not None:
+            grad_X.addmm_(q_grad, torch.mm(B_q_scaled, A_q_scaled))
+
+        # K path
+        k_weight_t = dequantize(k_weight, k_quant)
+        grad_X.addmm_(k_grad, k_weight_t)
+        del k_weight
+        del k_weight_t
+        if A_k is not None and B_k is not None:
+            grad_X.addmm_(k_grad, torch.mm(B_k_scaled, A_k_scaled))
+
+        # V path
+        v_weight_t = dequantize(v_weight, v_quant)
+        grad_X.addmm_(v_grad, v_weight_t)
+        del v_weight
+        del v_weight_t
+        if A_v is not None and B_v is not None:
+            grad_X.addmm_(v_grad, torch.mm(B_v_scaled, A_v_scaled))
+
+        # Transpose gradients if needed
+        if d_A_q is not None:
+            d_A_q = d_A_q.t()
+        if d_B_q is not None:
+            d_B_q = d_B_q.t()
+        if d_A_k is not None:
+            d_A_k = d_A_k.t()
+        if d_B_k is not None:
+            d_B_k = d_B_k.t()
+        if d_A_v is not None:
+            d_A_v = d_A_v.t()
+        if d_B_v is not None:
+            d_B_v = d_B_v.t()
+
+        return (
+            grad_X.view(batch, seq_len, -1),
+            None,
+            None,
+            d_A_q,
+            d_B_q,
+            None,
+            None,
+            None,
+            d_A_k,
+            d_B_k,
+            None,
+            None,
+            None,
+            d_A_v,
+            d_B_v,
+            None,
+            None,
+        )
+
+
+def apply_lora_qkv(
+    self, X: torch.Tensor, inplace: bool = True
+) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+    """
+    Applies LoRA to compute Query, Key, Value projections.
+
+    Args:
+        X: Input tensor
+        inplace: Whether to perform operations in-place
+
+    Returns:
+        Tuple of (Query, Key, Value) projection tensors
+    """
+    QW, QW_quant, QA, QB, QS = get_lora_parameters(self.q_proj)
+    KW, KW_quant, KA, KB, KS = get_lora_parameters(self.k_proj)
+    VW, VW_quant, VA, VB, VS = get_lora_parameters(self.v_proj)
+    Q, K, V = LoRA_QKV.apply(
+        X,
+        QW,
+        QW_quant,
+        QA,
+        QB,
+        QS,
+        KW,
+        KW_quant,
+        KA,
+        KB,
+        KS,
+        VW,
+        VW_quant,
+        VA,
+        VB,
+        VS,
+        inplace,
+    )
+
+    return Q, K, V
+
+
+class LoRA_O(torch.autograd.Function):
+    """Optimized LoRA implementation for output projection."""
+
+    @staticmethod
+    @torch_amp_custom_fwd
+    def forward(
+        ctx: torch.autograd.function.FunctionCtx,
+        X: torch.Tensor,
+        W: torch.Tensor,
+        W_quant: QuantState | None,
+        A: torch.Tensor | None,
+        B: torch.Tensor | None,
+        S: float,
+    ) -> torch.Tensor:
+        """
+        Forward pass for output projection with LoRA.
+
+        Args:
+            ctx: Autograd context
+            X: Input tensor
+            W: Output projection weight
+            W_quant: Weight quantization state
+            A: LoRA A matrix
+            B: LoRA B matrix
+            S: LoRA scaling factor
+
+        Returns:
+            Output projection tensor
+        """
+        XW = matmul_lora(X, W, W_quant, A, B, S)
+        ctx.custom_saved_tensors = (
+            W,
+            W_quant,
+            S,
+        )
+        ctx.save_for_backward(A, B, X)
+
+        return XW
+
+    @staticmethod
+    @torch_amp_custom_bwd
+    def backward(
+        ctx: torch.autograd.function.FunctionCtx,
+        dY: torch.Tensor,
+    ) -> tuple[
+        torch.Tensor,
+        None,
+        None,
+        torch.Tensor | None,
+        torch.Tensor | None,
+        None,
+    ]:
+        """
+        Backward pass computing gradients for LoRA output projection.
+
+        Args:
+            ctx: Autograd context
+            dY: Gradient of loss with respect to output
+
+        Returns:
+            Tuple containing gradients for all forward inputs
+        """
+        W, W_quant, S = ctx.custom_saved_tensors
+        A, B, X = ctx.saved_tensors
+
+        batch, seq_len, hd = X.shape
+        dY = dY.reshape(-1, dY.shape[-1])
+        X = X.reshape(-1, X.shape[-1])
+        dtype = X.dtype
+
+        # Weight projection
+        dY_X = X.t() @ dY
+        d_A = S * dY_X @ B
+        d_B = S * A @ dY_X
+
+        # Get derivative for dX
+        W = dequantize(W.t(), W_quant)
+        dX = dY @ W.t()
+        del W
+        dX += dY @ B.to(dtype) @ (S * A.to(dtype))
+
+        # W, W_quant, A, B, S
+        return dX.view(batch, seq_len, hd), None, None, d_A.t(), d_B.t(), None
+
+
+def apply_lora_o(self, X: torch.Tensor) -> torch.Tensor:
+    """
+    Applies LoRA to output projection layer.
+
+    Args:
+        X: Input tensor
+
+    Returns:
+        Transformed output tensor
+    """
+    OW, OW_quant, OA, OB, OS = get_lora_parameters(self.o_proj)
+    output = LoRA_O.apply(X, OW, OW_quant, OA, OB, OS)
+
+    return output

src/axolotl/kernels/quantize.py

@@ -0,0 +1,149 @@
+"""Dequantization utilities for `bitsandbytes` integration."""
+# pylint: disable=invalid-name,global-statement
+
+import ctypes
+
+import bitsandbytes as bnb
+import torch
+from bitsandbytes.functional import QuantState, get_ptr
+from packaging.version import Version
+
+cdequantize_blockwise_fp32 = bnb.functional.lib.cdequantize_blockwise_fp32
+cdequantize_blockwise_fp16_nf4 = bnb.functional.lib.cdequantize_blockwise_fp16_nf4
+cdequantize_blockwise_bf16_nf4 = bnb.functional.lib.cdequantize_blockwise_bf16_nf4
+
+CUDA_STREAM: torch.cuda.Stream | None = None
+HAS_CUDA_STREAM: bool = Version(bnb.__version__) > Version("0.43.3")
+
+
+def dequantize(
+    W: torch.Tensor,
+    quant_state: QuantState | list | None = None,
+    out: torch.Tensor | None = None,
+) -> torch.Tensor:
+    """
+    Fast NF4 dequantization using `bitsandbytes` CUDA kernels.
+
+    Performs efficient dequantization of weights from NF4 format using `bitsandbytes`'
+    optimized CUDA implementations. Supports both legacy list and new `QuantState`
+    formats.
+
+    Args:
+        W: Quantized weight tensor to dequantize
+        quant_state: Quantization state containing metadata needed for
+            dequantization. Can be either a `QuantState` object or legacy list format.
+            If None, returns `W` unchanged.
+        out: Optional output tensor for storing dequantized results. Must match
+            expected shape and dtype if provided.
+
+    Returns:
+        Dequantized tensor in the specified dtype (fp16 or bf16). Will be transposed if
+        input `W` was transposed.
+
+    Raises:
+        AssertionError: If provided output tensor doesn't match expected shape / dtype.
+
+    Note:
+        Uses CUDA streams for better performance when available in newer `bitsandbytes`
+        versions (>0.43.3).
+    """
+    if quant_state is None:
+        return W
+
+    # Get the target device from input tensor W
+    target_device = W.device
+
+    # Extract quantization state
+    if not isinstance(quant_state, list):
+        # New style quant_state class
+        absmax = quant_state.absmax.to(target_device)
+        shape = quant_state.shape
+        dtype = quant_state.dtype
+        blocksize = quant_state.blocksize
+        offset = quant_state.offset.to(target_device)
+        state2 = quant_state.state2
+        absmax2 = state2.absmax.to(target_device)
+        code2 = state2.code.to(target_device)
+        blocksize2 = state2.blocksize
+    else:
+        # Legacy list format
+        absmax, shape, dtype, blocksize, compressed_stats, _, _ = quant_state
+        absmax = absmax.to(target_device)
+        offset, state2 = compressed_stats
+        offset = offset.to(target_device)
+        absmax2, code2, blocksize2, _, _, _, _ = state2
+        absmax2 = absmax2.to(target_device)
+        code2 = code2.to(target_device)
+
+    # Setup output tensor on the same device as input
+    if out is None:
+        out = torch.empty(shape, dtype=dtype, device=target_device)
+    else:
+        assert out.shape == shape and out.dtype == dtype
+        out = out.to(target_device)
+
+    # Dequantize statistics on the target device
+    n_elements_absmax: int = absmax.numel()
+    out_absmax: torch.Tensor = torch.empty(
+        n_elements_absmax, dtype=torch.float32, device=target_device
+    )
+    ptr_out_absmax: int = get_ptr(out_absmax)
+
+    # Use CUDA stream if available
+    if HAS_CUDA_STREAM:
+        global CUDA_STREAM
+        if CUDA_STREAM is None:
+            CUDA_STREAM = torch.cuda.current_stream(target_device)
+
+        cdequantize_blockwise_fp32(
+            get_ptr(code2),
+            get_ptr(absmax),
+            get_ptr(absmax2),
+            ptr_out_absmax,
+            ctypes.c_int(blocksize2),
+            ctypes.c_int(n_elements_absmax),
+            CUDA_STREAM,
+        )
+    else:
+        cdequantize_blockwise_fp32(
+            get_ptr(code2),
+            get_ptr(absmax),
+            get_ptr(absmax2),
+            ptr_out_absmax,
+            ctypes.c_int(blocksize2),
+            ctypes.c_int(n_elements_absmax),
+        )
+
+    out_absmax += offset
+
+    # Choose appropriate dequantization function
+    fx = (
+        cdequantize_blockwise_fp16_nf4
+        if dtype == torch.float16
+        else cdequantize_blockwise_bf16_nf4
+    )
+
+    # Dequantize weights
+    if HAS_CUDA_STREAM:
+        fx(
+            get_ptr(None),
+            get_ptr(W),
+            ptr_out_absmax,
+            get_ptr(out),
+            ctypes.c_int(blocksize),
+            ctypes.c_int(out.numel()),
+            CUDA_STREAM,
+        )
+    else:
+        fx(
+            get_ptr(None),
+            get_ptr(W),
+            ptr_out_absmax,
+            get_ptr(out),
+            ctypes.c_int(blocksize),
+            ctypes.c_int(out.numel()),
+        )
+
+    # Handle transposed data
+    is_transposed: bool = W.shape[0] == 1
+    return out.t() if is_transposed else out

src/axolotl/kernels/swiglu.py

@@ -0,0 +1,163 @@
+"""
+Module for definition of SwiGLU Triton kernels.
+
+See "GLU Variants Improve Transformer" (https://arxiv.org/abs/2002.05202).
+
+Credit to `unsloth` (https://unsloth.ai/) for inspiration for this implementation.
+"""
+import torch
+import triton
+import triton.language as tl
+
+
+@triton.jit
+def _swiglu_fwd_kernel(
+    gate_ptr,
+    up_ptr,
+    out_ptr,
+    n_elements,
+    block_size: tl.constexpr,
+):
+    """
+    SwiGLU forward kernel. The kernel computes activation in fp32 precision for better
+    numerical stability, then converts back to original dtype for the final result.
+
+    Args:
+        gate_ptr: Pointer to gate tensor `[*, hidden_dim]`.
+        up_ptr: Pointer to up-projection tensor `[*, hidden_dim]`.
+        out_ptr: Pointer to output tensor `[*, hidden_dim]`.
+        n_elements: Total number of elements in the input tensors.
+        block_size: Size of thread blocks for parallel computation.
+    """
+    block_idx = tl.program_id(0)
+    offsets = block_idx * block_size + tl.arange(0, block_size)
+    mask = offsets < n_elements
+
+    # Load gate in fp32, keep up in original dtype
+    gate = tl.load(gate_ptr + offsets, mask=mask, other=0).to(tl.float32)
+    up = tl.load(up_ptr + offsets, mask=mask, other=0)
+
+    # Compute activation in fp32 then convert back
+    f = gate * tl.sigmoid(gate)
+    f = f.to(up.dtype)
+    result = f * up
+
+    tl.store(out_ptr + offsets, result, mask=mask)
+
+
+@triton.jit
+def _swiglu_bwd_kernel(
+    grad_out_ptr,
+    gate_ptr,
+    up_ptr,
+    n_elements,
+    block_size: tl.constexpr,
+):
+    """
+    SwiGLU backward kernel. Stores gradient results in-place.
+
+    Args:
+        grad_out_ptr: Pointer to gradient output tensor `[*, hidden_dim]`.
+        gate_ptr: Pointer to gate tensor `[*, hidden_dim]`.
+        up_ptr: Pointer to up-projection tensor `[*, hidden_dim]`.
+        n_elements: Total number of elements in the input tensors.
+        block_size: Size of thread blocks for parallel computation.
+
+    Note:
+        After kernel execution, tensors are modified in-place:
+        - `grad_out_ptr` contains forward output (`h`)
+        - `gate_ptr` contains gradient w.r.t gate (`grad_gate`)
+        - `up_ptr` contains gradient w.r.t up (`grad_up`)
+    """
+    block_idx = tl.program_id(0)
+    offsets = block_idx * block_size + tl.arange(0, block_size)
+    mask = offsets < n_elements
+
+    # Load values - only convert gate to fp32
+    grad_out = tl.load(grad_out_ptr + offsets, mask=mask, other=0)
+    gate = tl.load(gate_ptr + offsets, mask=mask, other=0).to(tl.float32)
+    up = tl.load(up_ptr + offsets, mask=mask, other=0)
+
+    # Compute SiLU and forward output
+    sigmoid_gate = tl.sigmoid(gate)
+    silu_gate = sigmoid_gate * gate
+    silu_gate = silu_gate.to(grad_out.dtype)
+    h = silu_gate * up
+
+    # Compute gradients
+    grad_up = grad_out * silu_gate  # gradient for up is grad_out * SiLU(gate)
+
+    # Compute gate gradient
+    temp = grad_out * up
+    grad_gate = temp.to(tl.float32) * sigmoid_gate * (1.0 + gate * (1.0 - sigmoid_gate))
+    grad_gate = grad_gate.to(grad_out.dtype)
+
+    # Store results with correct gradient ordering
+    tl.store(grad_out_ptr + offsets, h, mask=mask)
+    tl.store(gate_ptr + offsets, grad_gate, mask=mask)  # grad wrt gate
+    tl.store(up_ptr + offsets, grad_up, mask=mask)  # grad wrt up
+
+
+# pylint: disable=unnecessary-lambda-assignment
+def swiglu_forward(gate: torch.Tensor, up: torch.Tensor) -> torch.Tensor:
+    """
+    SwiGLU forward pass. Computes SwiGLU activation: `x * sigmoid(x) * up`, where
+    `x` is the gate tensor.
+
+    Args:
+        gate: Input gate tensor of shape `[batch, seq_len, hidden_dim]`.
+        up: Up-projection tensor of shape `[batch, seq_len, hidden_dim]`.
+
+    Returns:
+        Output tensor of shape `[batch, seq_len, hidden_dim]`.
+    """
+    batch, seq_len, hidden_dim = gate.shape
+    n_elements = gate.numel()
+    out = torch.empty((batch, seq_len, hidden_dim), dtype=gate.dtype, device="cuda")
+
+    grid = lambda meta: (triton.cdiv(n_elements, meta["block_size"]),)  # noqa: E731
+    _swiglu_fwd_kernel[grid](
+        gate_ptr=gate,
+        up_ptr=up,
+        out_ptr=out,
+        n_elements=n_elements,
+        block_size=1024,
+    )
+
+    return out
+
+
+# pylint: disable=unnecessary-lambda-assignment
+def swiglu_backward(
+    grad_output: torch.Tensor, gate: torch.Tensor, up: torch.Tensor
+) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+    """
+    SwiGLU backward pass using in-place operations.
+
+    Args:
+        grad_output: Gradient of loss with respect to output, shape `[batch, seq_len, hidden_dim]`.
+        gate: Gate tensor from forward pass, shape `[batch, seq_len, hidden_dim]`.
+        up: Up-projection tensor from forward pass, shape `[batch, seq_len, hidden_dim]`.
+
+    Returns:
+        Tuple containing:
+            - Forward pass output (`h`)
+            - Gradient with respect to gate (`df`)
+            - Gradient with respect to up-projection (`de`)
+    """
+    n_elements = grad_output.numel()
+
+    grid = lambda meta: (triton.cdiv(n_elements, meta["block_size"]),)  # noqa: E731
+    _swiglu_bwd_kernel[grid](
+        grad_out_ptr=grad_output,
+        gate_ptr=gate,
+        up_ptr=up,
+        n_elements=n_elements,
+        block_size=1024,
+    )
+
+    # After kernel execution, tensors contain:
+    # grad_output: h (forward output)
+    # gate: grad_gate (grad wrt gate)
+    # up: grad_up (grad wrt up)
+    return grad_output, gate, up

src/axolotl/kernels/utils.py

@@ -0,0 +1,11 @@
+"""Utilities for `axolotl.kernels` submodules."""
+
+import torch
+from packaging.version import Version
+
+if Version(torch.__version__) < Version("2.4.0"):
+    torch_amp_custom_fwd = torch.cuda.amp.custom_fwd
+    torch_amp_custom_bwd = torch.cuda.amp.custom_bwd
+else:
+    torch_amp_custom_fwd = torch.amp.custom_fwd(device_type="cuda")
+    torch_amp_custom_bwd = torch.amp.custom_bwd(device_type="cuda")

src/axolotl/monkeypatch/lora_kernels.py

@@ -0,0 +1,333 @@
+"""Module for patching custom LoRA Triton kernels and `torch.autograd` functions."""
+
+import importlib
+import inspect
+import logging
+import types
+from typing import Type
+
+import torch
+from accelerate.logging import get_logger
+from peft import PeftModelForCausalLM
+from torch import nn
+from transformers import AutoConfig
+
+from axolotl.kernels.lora import (
+    apply_lora_mlp_geglu,
+    apply_lora_mlp_swiglu,
+    apply_lora_o,
+    apply_lora_qkv,
+)
+from axolotl.monkeypatch.utils import detab_code
+from axolotl.utils.dict import DictDefault
+
+LOG = get_logger(__name__)
+
+ORIGINAL_QKV_CODE = """
+    query_states = self.q_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+    key_states = self.k_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+    value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+""".lstrip(
+    "\n"
+)
+
+PATCHED_QKV_CODE = """
+    query_states, key_states, value_states = self.apply_qkv(hidden_states)
+    query_states = query_states.view(hidden_shape).transpose(1, 2)
+    key_states = key_states.view(hidden_shape).transpose(1, 2)
+    value_states = value_states.view(hidden_shape).transpose(1, 2)
+""".lstrip(
+    "\n"
+)
+
+ORIGINAL_O_CODE = """
+    attn_output = self.o_proj(attn_output)
+""".lstrip(
+    "\n"
+)
+
+PATCHED_O_CODE = """
+    attn_output = self.apply_o(attn_output)
+""".lstrip(
+    "\n"
+)
+
+SUPPORTED_ACTIVATIONS = ["silu", "gelu"]
+APPLY_FN_MAPPING = {
+    "silu": apply_lora_mlp_swiglu,
+    "gelu": apply_lora_mlp_geglu,
+}
+
+
+def original_apply_qkv(
+    self: nn.Module, hidden_states: torch.Tensor
+) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+    """
+    Original implementation of QKV projection without optimizations.
+
+    Args:
+        self: The attention module instance.
+        hidden_states: Input tensor of shape [batch_size, seq_len, hidden_dim].
+
+    Returns:
+        A tuple `(query_states, key_states, value_states)` containing the projected
+            states for query, key, and value.
+    """
+    query_states = self.q_proj(hidden_states)
+    key_states = self.k_proj(hidden_states)
+    value_states = self.v_proj(hidden_states)
+
+    return query_states, key_states, value_states
+
+
+def original_apply_o(self: nn.Module, hidden_states: torch.Tensor) -> torch.Tensor:
+    """
+    Original implementation of output projection without optimizations.
+
+    Args:
+        self: The attention module instance.
+        hidden_states: Input tensor of shape `[`batch_size, seq_len, hidden_dim]`.
+
+    Returns:
+        The output projection result.
+    """
+    attn_output = self.o_proj(hidden_states)
+
+    return attn_output
+
+
+def get_attention_cls_from_config(cfg: DictDefault) -> Type[nn.Module]:
+    """
+    Get the appropriate attention class by inspecting the model config.
+    Uses dynamic import to support any model architecture that follows
+    the standard transformers naming convention.
+
+    Args:
+        cfg: Dictionary mapping `axolotl` config keys to values.
+
+    Returns:
+        The appropriate attention class for the model.
+
+    Raises:
+        ValueError: If `base_model` not specified or attention class cannot be imported
+        ImportError: If the model module or attention class doesn't exist
+    """
+    if "base_model" not in cfg:
+        raise ValueError("base_model must be specified in config")
+
+    # Get model config without loading the model
+    model_config = AutoConfig.from_pretrained(cfg["base_model"])
+    model_type = model_config.model_type
+
+    # Special case for model_type = "qwen2"
+    if model_type == "qwen2":
+        from transformers.models.qwen2.modeling_qwen2 import Qwen2Attention
+
+        return Qwen2Attention
+
+    try:
+        # Dynamically import the module and attention class
+        module_path = f"transformers.models.{model_type}.modeling_{model_type}"
+        module = __import__(
+            module_path, fromlist=[f"{model_type.capitalize()}Attention"]
+        )
+        attention_cls = getattr(module, f"{model_type.capitalize()}Attention")
+
+        return attention_cls
+    except (ImportError, AttributeError) as e:
+        raise ValueError(
+            f"Could not import attention class for model_type: {model_type}. "
+            f"Error: {str(e)}"
+        ) from e
+
+
+# pylint: disable=protected-access
+def patch_self_attn_lora(cfg: DictDefault):
+    """
+    Given an `axolotl` config, this method patches the inferred attention class forward
+    pass with optimized LoRA implementations.
+
+    It modifies the attention class to use optimized QKV and output projections. The
+    original implementation is preserved and can be restored if needed.
+
+    Args:
+        cfg: Dictionary mapping `axolotl` config keys to values.
+
+    Raises:
+        AssertionError: If the required code blocks are not found in the attention
+            implementation.
+    """
+    attention_cls = get_attention_cls_from_config(cfg)
+
+    # Check if already patched
+    if hasattr(attention_cls, "_original_forward"):
+        LOG.info(f"{attention_cls.__name__} already patched")
+        return
+
+    self_attn_forward = inspect.getsource(attention_cls.forward)
+    attention_cls._original_forward = self_attn_forward
+    self_attn_forward, _ = detab_code(self_attn_forward)
+
+    assert ORIGINAL_QKV_CODE in self_attn_forward, "Original QKV code not found"
+    assert ORIGINAL_O_CODE in self_attn_forward, "Original O code not found"
+
+    self_attn_forward = self_attn_forward.replace(ORIGINAL_QKV_CODE, PATCHED_QKV_CODE)
+    self_attn_forward = self_attn_forward.replace(ORIGINAL_O_CODE, PATCHED_O_CODE)
+    self_attn_forward = self_attn_forward.replace(
+        "def forward(",
+        "def axolotl_attn_forward(",
+        1,
+    )
+
+    # Load necessary imports
+    module_name = attention_cls.__module__
+    module = importlib.import_module(module_name)
+
+    items_to_import = []
+    for item in dir(module):
+        if item in self_attn_forward:
+            items_to_import.append(item)
+
+    exec(  # pylint: disable=exec-used  # nosec B102
+        f"from {module_name} import ({', '.join(items_to_import)})",
+        globals(),
+    )
+    exec(self_attn_forward, globals())  # pylint: disable=exec-used  # nosec B102
+
+    LOG.info(f"Patched attention class with LoRA optims: {attention_cls.__name__}")
+    attention_cls.forward = (
+        axolotl_attn_forward  # pylint: disable=undefined-variable  # noqa: F821
+    )
+
+
+def apply_lora_kernel_patches(
+    model: PeftModelForCausalLM, cfg: DictDefault
+) -> PeftModelForCausalLM:
+    """
+    Applies optimized Triton kernel patches to a PEFT model.
+
+    Patches a PEFT model with optimized implementations for MLP and attention
+    computations. The optimizations include custom Triton kernels for activation
+    functions and specialized autograd functions for LoRA computations.
+
+    Args:
+        model: A PEFT model to be patched with optimized kernels.
+        cfg: Dictionary mapping `axolotl` config keys to values.
+
+    Returns:
+        PeftModelForCausalLM: The patched model with optimized kernels.
+
+    Raises:
+        TypeError: If the provided model is not a `PeftModelForCausalLM`.
+        NotImplementedError: If the model type is not supported.
+        AssertionError: If multiple adapters are active (currently unsupported).
+
+    Note:
+        The optimizations require LoRA adapters with no dropout and no bias terms. The
+            function will skip patching if these conditions aren't met.
+    """
+    if not isinstance(model, PeftModelForCausalLM):
+        raise TypeError("Model must be a PeftModelForCausalLM")
+
+    # Get active LoRA adapter config
+    if hasattr(model, "active_adapters"):
+        assert (
+            len(model.active_adapters) == 1
+        ), "Axolotl currently does not support LoRA Triton kernels for multiple adapters"
+        active_adapter = model.active_adapters[0]
+    else:
+        active_adapter = model.active_adapter
+    lora_config = model.model.peft_config[active_adapter]
+
+    # Only patch if conditions are met
+    can_patch = lora_config.lora_dropout == 0 and lora_config.bias == "none"
+
+    if not can_patch:
+        LOG.warning("Cannot patch layers - requires no dropout and no bias")
+        LOG.warning("Please specify `lora_dropout: 0` in your axolotl config file")
+        return model
+
+    # This needs to be reset after patching
+    original_level = LOG.getEffectiveLevel()
+    LOG.setLevel(logging.INFO)
+
+    # Choose activation based on model type
+    activation = model.config.hidden_act
+    if activation not in SUPPORTED_ACTIVATIONS:
+        raise NotImplementedError(f"Activation {activation} is not supported")
+
+    # Patch each layer
+    for layer in model.model.model.layers:
+        # Add QKV, O fallback implementations to start
+        # These will be overwritten later (if some conditions apply)
+        layer.self_attn.apply_qkv = types.MethodType(
+            original_apply_qkv, layer.self_attn
+        )
+        layer.self_attn.apply_o = types.MethodType(original_apply_o, layer.self_attn)
+
+        if cfg.lora_mlp_kernel:
+            # MLP patching
+            gate_proj = layer.mlp.gate_proj
+            up_proj = layer.mlp.up_proj
+            down_proj = layer.mlp.down_proj
+
+            can_patch_mlp = all(
+                hasattr(proj, "lora_A")
+                and getattr(proj, "base_layer", proj).bias is None
+                and len(getattr(proj, "lora_magnitude_vector", []) or []) == 0
+                for proj in (gate_proj, up_proj, down_proj)
+            )
+
+            if can_patch_mlp:
+                apply_fn = APPLY_FN_MAPPING[activation]
+                layer.mlp.forward = types.MethodType(apply_fn, layer.mlp)
+            else:
+                LOG.warning_once(
+                    "Cannot patch some MLP layers - requires LoRA adapters with no bias"
+                )
+        if cfg.lora_qkv_kernel:
+            # Query, key, value patching
+            layer_modules = [
+                getattr(layer.self_attn, linear_proj)
+                for linear_proj in ["q_proj", "k_proj", "v_proj"]
+            ]
+            can_patch_qkv = all(
+                hasattr(module, "lora_A")
+                and getattr(module, "base_layer", module).bias is None
+                and len(getattr(module, "lora_magnitude_vector", []) or []) == 0
+                for module in layer_modules
+            )
+
+            if can_patch_qkv:
+                # Add optimized implementation
+                layer.self_attn.apply_qkv = types.MethodType(
+                    apply_lora_qkv, layer.self_attn
+                )
+            else:
+                LOG.warning_once(
+                    "Cannot patch some attention QKV projections - requires LoRA adapters with no bias"
+                )
+        if cfg.lora_o_kernel:
+            # Output patching
+            layer_modules = [
+                getattr(layer.self_attn, linear_proj) for linear_proj in ["o_proj"]
+            ]
+            can_patch_o = all(
+                hasattr(module, "lora_A")
+                and getattr(module, "base_layer", module).bias is None
+                and len(getattr(module, "lora_magnitude_vector", []) or []) == 0
+                for module in layer_modules
+            )
+
+            if can_patch_o:
+                layer.self_attn.apply_o = types.MethodType(
+                    apply_lora_o, layer.self_attn
+                )
+            else:
+                LOG.warning_once(
+                    "Cannot patch some attention output projection - requires LoRA adapters with no bias"
+                )
+
+    LOG.setLevel(original_level)
+
+    return model

src/axolotl/monkeypatch/multipack.py

@@ -25,6 +25,7 @@ SUPPORTED_MULTIPACK_MODEL_TYPES = [
     "gemmoe",
     "starcoder2",
     "deepseek_v2",
+    "deepseek_v3",
 ]
 
 

src/axolotl/monkeypatch/relora.py

@@ -127,6 +127,8 @@ class ReLoRACallback(TrainerCallback):
         optimizer: torch.optim.Optimizer,
         **_kwargs,
     ):
+        if not optimizer:
+            optimizer = state.optimizer
         if state.global_step > 0 and state.global_step % self.relora_steps == 0:
             checkpoint_folder = os.path.join(
                 args.output_dir,

src/axolotl/prompt_strategies/__init__.py

@@ -41,10 +41,10 @@ def load(strategy, tokenizer, cfg, ds_cfg, processor=None):
                 load_kwargs["ds_cfg"] = ds_cfg
             if "processor" in sig.parameters:
                 load_kwargs["processor"] = processor
+
         return func(tokenizer, cfg, **load_kwargs)
     except ModuleNotFoundError:
         return None
     except Exception as exc:  # pylint: disable=broad-exception-caught
         LOG.error(f"Failed to load prompt strategy `{strategy}`: {str(exc)}")
         raise exc
-    return None

src/axolotl/prompt_strategies/base.py

@@ -13,8 +13,19 @@ def load(strategy, cfg, module_base=None, **kwargs):
         if len(strategy.split(".")) == 1:
             strategy = strategy + ".default"
         load_fn = strategy.split(".")[-1]
-        strategy = ".".join(strategy.split(".")[:-1])
-        mod = importlib.import_module(f".{strategy}", module_base)
+        if len(strategy.split(".")) > 1:
+            try:
+                importlib.import_module(
+                    strategy.split(".")[-2],
+                    ".".join(strategy.split(".")[:-2]),
+                )
+                module_base = ".".join(strategy.split(".")[:-2])
+                strategy = strategy.split(".")[-2]
+            except ModuleNotFoundError:
+                strategy = "." + ".".join(strategy.split(".")[:-1])
+        else:
+            strategy = "." + ".".join(strategy.split(".")[:-1])
+        mod = importlib.import_module(strategy, module_base)
         func = getattr(mod, load_fn)
         return func(cfg, **kwargs)
     except Exception:  # pylint: disable=broad-exception-caught

src/axolotl/prompt_strategies/bradley_terry/chat_template.py

@@ -34,15 +34,12 @@ class BTChatTemplateStrategy(ChatTemplateStrategy):
 
         max_length = self.prompter.max_length
 
-        self.messages = "chosen_messages"
         # pylint: disable=duplicate-code
-        prompt[self.messages] = []
+        prompt["messages"] = []
         if prompt["system"]:
-            prompt[self.messages].append(
-                {"role": "system", "content": prompt["system"]}
-            )
-        prompt[self.messages].append({"role": "user", "content": prompt["input"]})
-        prompt[self.messages].append({"role": "assistant", "content": prompt["chosen"]})
+            prompt["messages"].append({"role": "system", "content": prompt["system"]})
+        prompt["messages"].append({"role": "user", "content": prompt["input"]})
+        prompt["messages"].append({"role": "assistant", "content": prompt["chosen"]})
         chosen_tokenized = super()._tokenize_single_prompt(prompt)
 
         if len(chosen_tokenized["input_ids"]) > max_length:
@@ -55,17 +52,12 @@ class BTChatTemplateStrategy(ChatTemplateStrategy):
                 :max_length
             ]
 
-        self.messages = "rejected_messages"
         # pylint: disable=duplicate-code
-        prompt[self.messages] = []
+        prompt["messages"] = []
         if prompt["system"]:
-            prompt[self.messages].append(
-                {"role": "system", "content": prompt["system"]}
-            )
-        prompt[self.messages].append({"role": "user", "content": prompt["input"]})
-        prompt[self.messages].append(
-            {"role": "assistant", "content": prompt["rejected"]}
-        )
+            prompt["messages"].append({"role": "system", "content": prompt["system"]})
+        prompt["messages"].append({"role": "user", "content": prompt["input"]})
+        prompt["messages"].append({"role": "assistant", "content": prompt["rejected"]})
         rejected_tokenized = super()._tokenize_single_prompt(prompt)
 
         if len(rejected_tokenized["input_ids"]) > max_length:
@@ -99,8 +91,13 @@ def load(tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None):
     prompter_params = {
         "tokenizer": tokenizer,
         "chat_template": chat_template_string,
-        "message_field_role": ds_cfg.get("message_field_role", "role"),
-        "message_field_content": ds_cfg.get("message_field_content", "content"),
+        "message_property_mappings": ds_cfg.get(
+            "message_property_mappings",
+            {
+                "role": "role",
+                "content": "content",
+            },
+        ),
         "message_field_training": ds_cfg.get("message_field_training", None),
         "message_field_training_detail": ds_cfg.get(
             "message_field_training_detail", None
@@ -124,7 +121,4 @@ def load(tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None):
         ChatTemplatePrompter(**prompter_params), tokenizer=tokenizer, **strategy_params
     )
 
-    if "field_messages" in ds_cfg and hasattr(strategy, "messages"):
-        strategy.messages = ds_cfg["field_messages"]
-
     return strategy

src/axolotl/prompt_strategies/chat_template.py

@@ -4,13 +4,16 @@ HF Chat Templates prompt strategy
 
 import logging
 from collections import defaultdict
-from typing import Any, Dict, List, Optional
+from typing import Any, Dict, List, Optional, Set, Union
 
+from pydantic import BaseModel
 from transformers import ProcessorMixin
 
+from axolotl.prompt_strategies.jinja_template_analyzer import JinjaTemplateAnalyzer
 from axolotl.prompt_tokenizers import PromptTokenizingStrategy
 from axolotl.prompters import IGNORE_TOKEN_ID, Prompter
 from axolotl.utils.chat_templates import get_chat_template_from_config
+from axolotl.utils.config.models.input.v0_4_1 import DatasetConfig
 
 # Configure the logger
 LOG = logging.getLogger("axolotl")
@@ -23,16 +26,23 @@ class ChatTemplatePrompter(Prompter):
     def __init__(
         self,
         tokenizer,
+        chat_template: str,
         processor=None,
-        chat_template=None,
         max_length=2048,
-        message_field_role: str = "role",
-        message_field_content: str = "content",
+        message_property_mappings: Optional[Dict[str, str]] = None,
         message_field_training: Optional[str] = None,
         message_field_training_detail: Optional[str] = None,
+        field_messages: str = "messages",
         roles: Optional[Dict[str, List[str]]] = None,
         drop_system_message: bool = False,
     ):
+        # check if message_property_mappings is None or empty dict
+        if message_property_mappings is None or (not message_property_mappings):
+            message_property_mappings = {
+                "role": "role",
+                "content": "content",
+            }
+
         if roles:
             self.roles = {s: t for t, sources in roles.items() for s in sources}
         else:
@@ -45,18 +55,28 @@ class ChatTemplatePrompter(Prompter):
                 "tool": "tool",
             }
 
-        self.message_field_role = message_field_role
-        self.message_field_content = message_field_content
+        self._chat_template_msg_variables = self.get_chat_template_msg_variables(
+            chat_template, field_messages
+        )
+        self.message_property_mappings = message_property_mappings
         self.message_field_training = message_field_training
         self.message_field_training_detail = message_field_training_detail
+        self.field_messages = field_messages
         self.tokenizer = tokenizer
-        self.processor: ProcessorMixin = processor
+        self.processor: Optional[ProcessorMixin] = processor
         self.chat_template = chat_template
         self.max_length = max_length
         self.drop_system_message = drop_system_message
 
+    @property
+    def chat_template_msg_variables(self) -> Set[str]:
+        return self._chat_template_msg_variables
+
     def build_prompt(self, conversation, add_generation_prompt=False, images=None):
         if self.processor:
+            if not callable(self.processor):
+                raise TypeError("Processor must be callable")
+
             text = self.processor.apply_chat_template(
                 conversation,
                 chat_template=self.chat_template,
@@ -184,17 +204,21 @@ class ChatTemplatePrompter(Prompter):
 
         return adjusted_details
 
+    def get_chat_template_msg_variables(
+        self, chat_template: str, field_messages: str
+    ) -> Set[str]:
+        template_analyzer = JinjaTemplateAnalyzer(chat_template)
+        return template_analyzer.get_message_vars(field_messages)
+
 
 class ChatTemplateStrategy(PromptTokenizingStrategy):
     """
     Tokenizing strategy for instruction-based prompts.
     """
 
-    _messages = "messages"
-
     def __init__(
         self,
-        prompter: ChatTemplatePrompter,
+        prompter: "ChatTemplatePrompter",
         tokenizer,
         train_on_inputs,
         sequence_len,
@@ -202,6 +226,7 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
         train_on_eos=None,
     ):
         super().__init__(prompter, tokenizer, train_on_inputs, sequence_len)
+        self.prompter: ChatTemplatePrompter = prompter
 
         self.roles_to_train = []
         if roles_to_train:
@@ -213,13 +238,9 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
         self.train_on_eos = train_on_eos
         self.images = "images"
 
-    @property
-    def messages(self):
-        return self._messages
-
-    @messages.setter
-    def messages(self, messages):
-        self._messages = messages
+        LOG.debug(
+            f"The chat template uses the following properites on the message: {self.prompter.chat_template_msg_variables}"
+        )
 
     @property
     def supports_batched(self) -> bool:
@@ -229,7 +250,7 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
     def is_prompt_batched(self, prompt: dict[str, Any]) -> bool:
         try:
             return all(isinstance(v, list) for v in prompt.values()) and all(
-                isinstance(v, list) for v in prompt[self.messages]
+                isinstance(v, list) for v in prompt[self.prompter.field_messages]
             )
         except KeyError:
             return False
@@ -251,8 +272,7 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
                 dict(zip(feature_names, row))
             )
             for key, val in tokenized_prompt.items():
-                for i in range(0, len(val), self.sequence_len):
-                    res[key].append(val[i : i + self.sequence_len])
+                res[key].append(val)
 
         # If there are no examples left, return an empty dictionary
         if not res:
@@ -464,30 +484,17 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
 
     def get_conversation_thread(self, prompt):
         turns = []
-        optional_keys = [
-            "tool_calls",  # tool that 'assistant' calls
-            "name",  # name of tool given by 'tool'
-            "tool_call_id",  # mistral/mixtral requires this
-        ]
-        for message in prompt[self.messages]:
+        for message in prompt[self.prompter.field_messages]:
+            transformed_message = self.transform_message(message)
+
             turn = {
-                "role": self.prompter.roles[message[self.prompter.message_field_role]],
+                **transformed_message,
                 "training": message.get(self.prompter.message_field_training),
                 "training_detail": message.get(
                     self.prompter.message_field_training_detail
                 ),
             }
 
-            # do not add content if None as it may conflict with some templates due to tools
-            content = message.get(self.prompter.message_field_content, None)
-            if content is not None:
-                turn["content"] = content
-
-            for key in optional_keys:
-                value = message.get(key, None)
-                if value is not None:
-                    turn[key] = value
-
             turns.append(turn)
 
         if self.prompter.drop_system_message and turns[0]["role"] == "system":
@@ -495,6 +502,37 @@ class ChatTemplateStrategy(PromptTokenizingStrategy):
 
         return turns
 
+    def transform_message(self, message):
+        # Build the initial transformed message from the mappings
+        transformed_message = {}
+        for key, value in self.prompter.message_property_mappings.items():
+            if message.get(value) is not None:
+                transformed_message[key] = message[value]
+            else:
+                LOG.debug(
+                    f"Could not find value for property {value} in message: {message}"
+                )
+
+        # Map the role if necessary
+        if "role" in transformed_message:
+            transformed_message["role"] = self.prompter.roles.get(
+                transformed_message["role"], transformed_message["role"]
+            )
+
+        # Determine which keys in the original message were not mapped
+        mapped_values = set(self.prompter.message_property_mappings.values())
+        remaining_keys = set(message) - mapped_values
+
+        # Keep only the properties defined in the chat template
+        # and not already mapped
+        for key in self.prompter.chat_template_msg_variables:
+            if key in remaining_keys:
+                val = message.get(key)
+                if val is not None:
+                    transformed_message[key] = val
+
+        return transformed_message
+
     def get_images(self, prompt):
         return prompt.get(self.images, None)
 
@@ -516,33 +554,46 @@ class StrategyLoader:
         }
 
     def __call__(
-        self, tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None, processor=None
+        self,
+        tokenizer,
+        cfg,
+        ds_cfg: Optional[Union[Dict[str, Any], DatasetConfig]] = None,
+        processor=None,
     ):
-        # pylint: disable=duplicate-code
-        ds_cfg = ds_cfg or {}
+        if ds_cfg is None:
+            dataset_config = {}
+        elif isinstance(ds_cfg, BaseModel):
+            dataset_config = ds_cfg.model_dump()
+        else:
+            dataset_config = ds_cfg
+
         chat_template_string = get_chat_template_from_config(
-            cfg=cfg, ds_cfg=ds_cfg, tokenizer=tokenizer
+            cfg=cfg, ds_cfg=dataset_config, tokenizer=tokenizer
         )
         LOG.info(f"Using chat template:\n---\n{chat_template_string!s}\n---")
 
         prompter_params = {
             "tokenizer": tokenizer,
             "chat_template": chat_template_string,
-            "message_field_role": ds_cfg.get("message_field_role", "role"),
-            "message_field_content": ds_cfg.get("message_field_content", "content"),
-            "message_field_training": ds_cfg.get("message_field_training", None),
-            "message_field_training_detail": ds_cfg.get(
+            "message_property_mappings": dataset_config.get(
+                "message_property_mappings", {}
+            ),
+            "message_field_training": dataset_config.get(
+                "message_field_training", None
+            ),
+            "message_field_training_detail": dataset_config.get(
                 "message_field_training_detail",
                 None,
             ),
-            "roles": ds_cfg.get("roles"),
-            "drop_system_message": ds_cfg.get("drop_system_message", False),
+            "field_messages": dataset_config.get("field_messages", "messages"),
+            "roles": dataset_config.get("roles"),
+            "drop_system_message": dataset_config.get("drop_system_message", False),
             # we need to add one for detecting sequences with exceeding the `sequence_len` limit.
             "max_length": cfg.sequence_len + 1,
             "processor": processor,
         }
 
-        strategy_params = self._get_strategy_params(cfg, ds_cfg)
+        strategy_params = self._get_strategy_params(cfg, dataset_config)
         strategy_cls = self._get_strategy_cls()
 
         strategy = strategy_cls(
@@ -551,9 +602,6 @@ class StrategyLoader:
             **strategy_params,
         )
 
-        if "field_messages" in ds_cfg and hasattr(strategy, "messages"):
-            strategy.messages = ds_cfg["field_messages"]
-
         return strategy
 
 

src/axolotl/prompt_strategies/dpo/chat_template.py

@@ -3,20 +3,28 @@ DPO prompt strategies for using tokenizer chat templates.
 """
 
 from axolotl.utils.chat_templates import extract_chat_template_args, get_chat_template
+from axolotl.utils.config.models.input.v0_4_1 import handle_legacy_message_fields_logic
 
 
 def default(
     cfg, dataset_idx=0, **kwargs
 ):  # pylint: disable=possibly-unused-variable,unused-argument
     ds_cfg = cfg["datasets"][dataset_idx]
+    ds_cfg = handle_legacy_message_fields_logic(ds_cfg)
+
     chat_template_choice, chat_template_jinja = extract_chat_template_args(
         cfg=cfg, ds_cfg=ds_cfg
     )
     field_messages = ds_cfg.get("field_messages", "messages")
     field_chosen = ds_cfg.get("field_chosen", "chosen")
     field_rejected = ds_cfg.get("field_rejected", "rejected")
-    field_message_role = ds_cfg.get("message_field_role", "role")
-    field_message_content = ds_cfg.get("message_field_content", "content")
+    message_property_mappings = ds_cfg.get(
+        "message_property_mappings",
+        {
+            "role": "role",
+            "content": "content",
+        },
+    )
     role_map_inv = ds_cfg.get(
         "roles",
         {
@@ -40,18 +48,18 @@ def default(
         messages = sample[field_messages]
         messages = [
             {
-                "role": role_map[m[field_message_role]],
-                "content": m[field_message_content],
+                "role": role_map[m[message_property_mappings["role"]]],
+                "content": m[message_property_mappings["content"]],
             }
             for m in messages
         ]
         chosen = {
-            "role": role_map[sample[field_chosen][field_message_role]],
-            "content": sample[field_chosen][field_message_content],
+            "role": role_map[sample[field_chosen][message_property_mappings["role"]]],
+            "content": sample[field_chosen][message_property_mappings["content"]],
         }
         rejected = {
-            "role": role_map[sample[field_rejected][field_message_role]],
-            "content": sample[field_rejected][field_message_content],
+            "role": role_map[sample[field_rejected][message_property_mappings["role"]]],
+            "content": sample[field_rejected][message_property_mappings["content"]],
         }
         dummy_user_message = {"role": "user", "content": "[[dummy_message]]"}
 

src/axolotl/prompt_strategies/dpo/passthrough.py

@@ -0,0 +1,14 @@
+"""
+DPO prompt strategies passthrough/zero-processing strategy
+"""
+
+
+def default(
+    cfg, dataset_idx=0, **kwargs
+):  # pylint: disable=possibly-unused-variable,unused-argument
+    def transform_fn(
+        sample, tokenizer=None
+    ):  # pylint: disable=possibly-unused-variable,unused-argument
+        return sample
+
+    return transform_fn

src/axolotl/prompt_strategies/jinja_template_analyzer.py

@@ -0,0 +1,318 @@
+"""Module for inspect jinja templates for the variables they use"""
+from typing import Dict, Optional, Set, TypedDict, Union
+
+from jinja2 import Environment, meta, nodes
+
+
+class JinjaTemplateAnalysis(TypedDict):
+    """
+    Represents the detailed analysis of a Jinja template variable.
+
+    Attributes:
+        accessed_properties (Set[str]): A set of properties accessed from the variable
+            (e.g., `foo.bar` results in 'bar' being accessed for 'foo').
+        accessed_indices (Set[Union[int, float]]): A set of indices accessed from the variable.
+        is_iterated (bool): Indicates if the variable is used as an iteration source in a `for` loop.
+        is_conditional (bool): Indicates if the variable is referenced within a conditional statement (e.g., an `if` block).
+        iteration_source (Optional[str]): The name of the variable being iterated over, if applicable.
+        iteration_target (Optional[Union[str, list[str]]]): The loop target(s) assigned in the iteration.
+    """
+
+    accessed_properties: Set[str]
+    accessed_indices: Set[Union[int, float]]
+    is_iterated: bool
+    is_conditional: bool
+    iteration_source: Optional[str]
+    iteration_target: Optional[Union[str, list[str]]]
+
+
+class JinjaTemplateAnalyzer:
+    """
+    Analyzes Jinja templates to extract information about variable usage,
+    including accessed properties, iteration, and conditional references.
+
+    Attributes:
+        env (jinja2.Environment): The Jinja2 environment used for parsing templates.
+        property_access (Dict[str, Set[str]]): Tracks accessed properties for variables.
+        iteration_targets (Dict[str, str]): Maps iteration target variables to their sources.
+
+    Methods:
+        get_template_variables(template: str) -> Dict[str, Set[str]]:
+            Parse a Jinja template and return a mapping of variables to their accessed properties.
+
+        analyze_template(template: str) -> Dict[str, JinjaTemplateAnalysis]:
+            Perform a detailed analysis of the template, including variable usage,
+            iteration, and conditional references.
+
+    Private Methods:
+        _visit_node(node) -> None:
+            Recursively visit AST nodes to detect attribute access and iteration targets.
+
+        _get_base_name(node) -> Optional[str]:
+            Extract the base variable name from a node.
+
+        _get_target_name(node) -> Optional[Union[str, list[str]]]:
+            Extract the target name(s) from a `For` node.
+    """
+
+    def __init__(self, template: str):
+        self.env: Environment = Environment(autoescape=True)
+        self.property_access: Dict[str, Set[str]] = {}
+        self.iteration_targets: Dict[str, Union[str, list[str]]] = {}
+        self.index_access: Dict[str, Set[Union[int, float]]] = {}
+        self.ast: nodes.Node = self.env.parse(template)
+        self.template: str = template
+        self.variable_assignments: Dict[str, str] = {}
+
+    def _visit_node(self, node) -> None:
+        """Recursively visit AST nodes to find attribute access."""
+        # Handle attribute access (dot notation)
+        if isinstance(node, nodes.Getattr):
+            base_name = self._get_base_name(node.node)
+            if base_name:
+                self.property_access.setdefault(base_name, set()).add(node.attr)
+
+        # Handle dictionary access (subscript notation)
+        elif isinstance(node, nodes.Getitem):
+            base_name = self._get_base_name(node.node)
+            if base_name and isinstance(node.arg, nodes.Const):
+                value = node.arg.value
+                if isinstance(value, (int, float)):
+                    self.index_access.setdefault(base_name, set()).add(value)
+                else:
+                    self.property_access.setdefault(base_name, set()).add(value)
+
+        elif isinstance(node, nodes.Test) and node.name == "defined":
+            base_name = self._get_base_name(node.node)
+            if base_name:
+                if isinstance(node.node, nodes.Getattr):
+                    self.property_access.setdefault(base_name, set()).add(
+                        node.node.attr
+                    )
+
+        # Handle loop variables
+        elif isinstance(node, nodes.For):
+            iter_name = self._get_base_name(node.iter)
+            target_name = self._get_target_name(node.target)
+            if iter_name and target_name:
+                self.iteration_targets[target_name] = iter_name
+                self.property_access.setdefault(iter_name, set())
+
+        elif isinstance(node, nodes.Assign):
+            target_name = self._get_target_name(node.target)
+            source_name = self._get_base_name(node.node)
+            if target_name and source_name:
+                self.variable_assignments[target_name] = source_name
+
+        elif isinstance(node, nodes.Filter):
+            if node.name == "selectattr":
+                target = self._get_base_name(node.node)
+                if target:
+                    self.variable_assignments[f"filtered_{target}"] = target
+
+        for child in node.iter_child_nodes():
+            self._visit_node(child)
+
+    def _get_target_name(self, node) -> Optional[str]:
+        """Get the target variable name from a For node.
+
+        Args:
+            node: A Jinja AST node representing either a Name or Tuple node
+
+        Returns:
+            - str: For simple variable targets (e.g., "item" in "for item in items")
+            - None: If the node type is not recognized or is a tuple
+        """
+        if isinstance(node, nodes.Name):
+            return node.name
+        return None
+
+    def _get_target_names(self, node) -> list[str]:
+        """Get all target variable names from a For node, including tuple unpacking.
+
+        Args:
+            node: A Jinja AST node representing either a Name or Tuple node
+
+        Returns:
+            List of target variable names
+        """
+        if isinstance(node, nodes.Name):
+            return [node.name]
+
+        if isinstance(node, nodes.Tuple):
+            names = []
+            for n in node.items:
+                if isinstance(n, nodes.Name):
+                    names.append(n.name)
+            return names
+
+        return []
+
+    def _get_base_name(self, node) -> Optional[str]:
+        """Get the base variable name from a node."""
+        if isinstance(node, nodes.Name):
+            return node.name
+
+        if isinstance(node, nodes.Getattr):
+            return self._get_base_name(node.node)
+
+        if isinstance(node, nodes.Getitem):
+            return self._get_base_name(node.node)
+
+        return None
+
+    def get_template_variables(self) -> Dict[str, Set[str]]:
+        """
+        Parse a Jinja template and return both variables and their accessed properties.
+
+        Args:
+            template (str): The Jinja template string
+
+        Returns:
+            Dict[str, Set[str]]: Dictionary mapping variable names to sets of accessed properties
+        """
+        # Parse the template
+        ast = self.env.parse(self.template)
+
+        # Get all undeclared variables
+        variables = meta.find_undeclared_variables(ast)
+
+        # Reset property access tracking
+        self.property_access = {}
+
+        # Visit all nodes to find property access
+        self._visit_node(ast)
+
+        # Create result dictionary
+        result: Dict[str, Set[str]] = {var: set() for var in variables}
+        # Merge in any discovered sub-properties
+        for var, props in self.property_access.items():
+            if var not in result:
+                result[var] = set()
+            result[var].update(props)
+
+        return result
+
+    def analyze_template(self) -> Dict[str, JinjaTemplateAnalysis]:
+        """
+        Provide a detailed analysis of template variables and their usage.
+        """
+        variables = self.get_template_variables()
+        self.iteration_targets = {}
+
+        analysis: Dict[str, JinjaTemplateAnalysis] = {
+            var: JinjaTemplateAnalysis(
+                accessed_properties=props,
+                accessed_indices=set(),
+                is_iterated=False,
+                is_conditional=False,
+                iteration_source=None,
+                iteration_target=None,
+            )
+            for var, props in variables.items()
+        }
+
+        for var, indices in self.index_access.items():
+            if var in analysis:
+                analysis[var]["accessed_indices"] = indices
+
+        def visit_node(node):
+            if isinstance(node, nodes.If):
+
+                def find_test_vars(test_node):
+                    if isinstance(test_node, nodes.Name):
+                        if test_node.name in analysis:
+                            analysis[test_node.name]["is_conditional"] = True
+                    for child in test_node.iter_child_nodes():
+                        find_test_vars(child)
+
+                find_test_vars(node.test)
+
+            if isinstance(node, nodes.For):
+                iter_target = self._get_base_name(node.iter)
+                target_name = self._get_target_name(node.target)
+                if iter_target in analysis:
+                    analysis[iter_target]["is_iterated"] = True
+                    if target_name:
+                        analysis[iter_target]["iteration_target"] = target_name
+                        if isinstance(target_name, str) and target_name not in analysis:
+                            analysis[target_name] = {
+                                "accessed_properties": set(),
+                                "is_iterated": False,
+                                "is_conditional": False,
+                                "iteration_source": iter_target,
+                                "iteration_target": None,
+                            }
+
+            for child in node.iter_child_nodes():
+                visit_node(child)
+
+        visit_node(self.ast)
+        return analysis
+
+    def get_downstream_properties(self, start_var: str) -> Dict[str, Set[str]]:
+        """
+        Get all properties accessed on a variable and its downstream assignments.
+
+        Args:
+            start_var: The starting variable to trace
+
+        Returns:
+            Dict mapping variable names to their accessed properties
+        """
+        visited = set()
+        properties = {}
+
+        def trace_variable(var_name: str):
+            if var_name in visited:
+                return
+            visited.add(var_name)
+
+            # Get direct properties
+            if var_name in self.property_access:
+                properties[var_name] = self.property_access[var_name]
+
+            # Get properties from iteration targets
+            if var_name in self.iteration_targets:
+                target = self.iteration_targets[var_name]
+                if isinstance(target, str):
+                    trace_variable(target)
+                elif isinstance(target, list):
+                    for t in target:
+                        trace_variable(t)
+
+            # Follow assignments
+            for target, source in self.variable_assignments.items():
+                if source == var_name:
+                    trace_variable(target)
+
+            # Check for array slicing
+            analysis = self.analyze_template()
+            if var_name in analysis:
+                var_info = analysis[var_name]
+                if var_info["accessed_indices"]:
+                    # If this variable is sliced, follow the resulting assignment
+                    slice_result = f"{var_name}_slice"
+                    if slice_result in self.property_access:
+                        trace_variable(slice_result)
+
+        trace_variable(start_var)
+        return properties
+
+    def get_message_vars(self, field_messages: str = "messages") -> Set[str]:
+        """
+        Get all properties accessed on messages and derived variables.
+        """
+        all_properties = self.get_downstream_properties(field_messages)
+
+        # Combine all properties from all related variables
+        combined_properties = set()
+        for properties in all_properties.values():
+            combined_properties.update(properties)
+
+        # Also include properties from the message iteration variable
+        analysis = self.analyze_template()
+        if "message" in analysis:
+            combined_properties.update(analysis["message"]["accessed_properties"])
+
+        return combined_properties

src/axolotl/prompt_strategies/messages/chat.py

@@ -51,8 +51,13 @@ def load(tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None):
     ds_cfg = ds_cfg or {}
 
     field_messages = ds_cfg.get("field_messages")
-    message_field_role = ds_cfg.get("message_field_role")
-    message_field_content = ds_cfg.get("message_field_content")
+    message_property_mappings = ds_cfg.get("message_property_mappings")
+    message_field_role = (
+        message_property_mappings.get("role") if message_property_mappings else None
+    )
+    message_field_content = (
+        message_property_mappings.get("content") if message_property_mappings else None
+    )
     message_field_training = ds_cfg.get("message_field_training")
 
     builder_kwargs = {}

src/axolotl/train.py

@@ -175,6 +175,7 @@ def train(
         LOG.info("hang tight... sorting dataset for group_by_length")
 
     pretrain_hooks(cfg, trainer)
+
     if cfg.flash_optimum:
         with torch.backends.cuda.sdp_kernel(
             # TODO configure these from the YAML w/ sdp_kernel_kwargs: ...
@@ -185,6 +186,7 @@ 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}")

src/axolotl/utils/config/__init__.py

@@ -18,6 +18,7 @@ from axolotl.utils.config.models.input.v0_4_1 import (
 from axolotl.utils.config.models.input.v0_4_1 import (
     AxolotlInputConfig as AxolotlInputConfigBase,
 )
+from axolotl.utils.config.models.input.v0_4_1 import DPODataset, KTODataset, SFTDataset
 from axolotl.utils.dict import DictDefault
 from axolotl.utils.models import load_model_config
 
@@ -258,7 +259,7 @@ def validate_config(
     cfg: DictDefault,
     capabilities: Optional[dict] = None,
     env_capabilities: Optional[dict] = None,
-):
+) -> DictDefault:
     AxolotlConfigWCapabilities = AxolotlConfigWCapabilitiesBase
     AxolotlInputConfig = AxolotlInputConfigBase
 
@@ -268,6 +269,16 @@ def validate_config(
             AxolotlInputConfig,  # pylint: disable=invalid-name
         ) = merge_input_args()
 
+    # Convert datasets to proper format if needed
+    if cfg.get("datasets"):
+        for idx, ds_cfg in enumerate(cfg["datasets"]):
+            if cfg.get("rl") == "dpo" and not isinstance(ds_cfg, DPODataset):
+                cfg["datasets"][idx] = DPODataset(**ds_cfg)
+            elif cfg.get("rl") == "kto" and not isinstance(ds_cfg, KTODataset):
+                cfg["datasets"][idx] = KTODataset(**dict(ds_cfg))
+            elif not isinstance(ds_cfg, SFTDataset):
+                cfg["datasets"][idx] = SFTDataset(**dict(ds_cfg))
+
     if capabilities or env_capabilities:
         if (capabilities and env_capabilities is None) or (
             env_capabilities and capabilities is None

src/axolotl/utils/config/models/input/v0_4_1/__init__.py

@@ -1,7 +1,4 @@
-"""
-Module for pydantic models for configuration
-"""
-
+"""Module with Pydantic models for configuration."""
 # pylint: disable=too-many-lines
 
 import logging
@@ -9,12 +6,13 @@ import os
 from enum import Enum
 from typing import Annotated, Any, Dict, List, Literal, Optional, Tuple, Union
 
+from annotated_types import MinLen
 from packaging import version
 from pydantic import (
     BaseModel,
     Field,
     StringConstraints,
-    conlist,
+    field_serializer,
     field_validator,
     model_validator,
 )
@@ -24,6 +22,8 @@ from transformers.utils.import_utils import is_torch_npu_available
 
 from axolotl.utils.config.models.internals import EnvCapabilities, GPUCapabilities
 
+from .trl import TRLConfig
+
 LOG = logging.getLogger("axolotl.utils.config.models.input")
 
 SUPPORTED_METRICS = {"sacrebleu", "comet", "ter", "chrf", "perplexity"}
@@ -33,6 +33,7 @@ class RLType(str, Enum):
     """RL trainer type configuration subset"""
 
     dpo = "dpo"  # pylint: disable=invalid-name
+    grpo = "grpo"  # pylint: disable=invalid-name
     ipo = "ipo"  # pylint: disable=invalid-name
     orpo = "orpo"  # pylint: disable=invalid-name
     kto = "kto"  # pylint: disable=invalid-name
@@ -54,6 +55,7 @@ class ChatTemplate(str, Enum):
     phi_3 = "phi_3"  # pylint: disable=invalid-name
     phi_35 = "phi_35"  # pylint: disable=invalid-name
     deepseek_v2 = "deepseek_v2"  # pylint: disable=invalid-name
+    deepseek_v3 = "deepseek_v3"  # pylint: disable=invalid-name
     jamba = "jamba"  # pylint: disable=invalid-name
     jinja = "jinja"  # pylint: disable=invalid-name
     qwen_25 = "qwen_25"  # pylint: disable=invalid-name
@@ -166,6 +168,7 @@ class SFTDataset(BaseModel):
     type: Optional[Union[str, UserDefinedPrompterType]] = None
     input_transform: Optional[str] = None
     shards: Optional[int] = None
+    shards_idx: Optional[int] = None
     preprocess_shards: Optional[int] = None
     conversation: Optional[str] = None
     # Do not make this too strict or it will break the validator to choose different dataset class
@@ -185,8 +188,13 @@ class SFTDataset(BaseModel):
     field_human: Optional[str] = None
     field_model: Optional[str] = None
     field_messages: Optional[str] = None
-    message_field_role: Optional[str] = None
-    message_field_content: Optional[str] = None
+    message_field_role: Optional[
+        str
+    ] = None  # deprecated, use message_property_mappings
+    message_field_content: Optional[
+        str
+    ] = None  # deprecated, use message_property_mappings
+    message_property_mappings: Optional[Dict[str, str]] = None
     message_field_training: Optional[str] = None
     message_field_training_detail: Optional[str] = None
     logprobs_field: Optional[str] = None
@@ -198,9 +206,18 @@ class SFTDataset(BaseModel):
     trust_remote_code: Optional[bool] = False
     revision: Optional[str] = None
 
+    @model_validator(mode="before")
+    @classmethod
+    def handle_legacy_message_fields(cls, data):
+        """Handle backwards compatibility between legacy message field mapping and new property mapping system."""
+        return handle_legacy_message_fields_logic(data)
+
     @model_validator(mode="before")
     @classmethod
     def check_chat_template_config(cls, data):
+        if isinstance(data, BaseModel):
+            data = data.model_dump()
+
         # Set chat_template to tokenizer_default if not set
         if data.get("type") == "chat_template" and not data.get("chat_template"):
             data["chat_template"] = ChatTemplate.tokenizer_default
@@ -240,6 +257,7 @@ class DPODataset(BaseModel):
     type: Optional[Union[UserDefinedDPOType, str]] = None
     data_files: Optional[List[str]] = None
     revision: Optional[str] = None
+    field_messages: Optional[str] = None
 
 
 class StepwiseSupervisedDataset(BaseModel):
@@ -276,6 +294,9 @@ class KTODataset(BaseModel):
     revision: Optional[str] = None
 
 
+DatasetConfig = Union[SFTDataset, DPODataset, KTODataset, StepwiseSupervisedDataset]
+
+
 class LoftQConfig(BaseModel):
     """LoftQ configuration subset"""
 
@@ -322,6 +343,7 @@ class LoraConfig(BaseModel):
     peft_use_dora: Optional[bool] = None
     peft_use_rslora: Optional[bool] = None
     peft_layer_replication: Optional[List[Tuple[int, int]]] = None
+    peft_init_lora_weights: Optional[Union[bool, str]] = None
 
     qlora_sharded_model_loading: Optional[bool] = Field(
         default=False,
@@ -415,6 +437,8 @@ class ReLoRAConfig(BaseModel):
 class ModelInputConfig(BaseModel):
     """model to train on configuration subset"""
 
+    model_config = {"protected_namespaces": ()}
+
     base_model: str
     base_model_config: Optional[str] = None
     cls_model_config: Optional[str] = None
@@ -481,7 +505,7 @@ class HyperparametersConfig(BaseModel):
                 "adopt_adamw",
             ],
         ]
-    ] = OptimizerNames.ADAMW_HF.value
+    ] = OptimizerNames.ADAMW_HF
     optim_args: Optional[Union[str, Dict[str, Any]]] = Field(
         default=None,
         json_schema_extra={"description": "Optional arguments to supply to optimizer."},
@@ -493,7 +517,9 @@ class HyperparametersConfig(BaseModel):
         },
     )
     torchdistx_path: Optional[str] = None
-    lr_scheduler: Optional[Union[SchedulerType, Literal["one_cycle"]]] = "cosine"
+    lr_scheduler: Optional[
+        Union[SchedulerType, Literal["one_cycle"]]
+    ] = SchedulerType.COSINE
     lr_scheduler_kwargs: Optional[Dict[str, Any]] = None
     lr_quadratic_warmup: Optional[bool] = None
     cosine_min_lr_ratio: Optional[float] = None
@@ -617,19 +643,19 @@ class RayConfig(BaseModel):
     use_ray: bool = Field(default=False)
     ray_run_name: Optional[str] = Field(
         default=None,
-        metadata={
+        json_schema_extra={
             "help": "The training results will be saved at `saves/ray_run_name`."
         },
     )
     ray_num_workers: int = Field(
         default=1,
-        metadata={
+        json_schema_extra={
             "help": "The number of workers for Ray training. Default is 1 worker."
         },
     )
     resources_per_worker: dict = Field(
         default_factory=lambda: {"GPU": 1},
-        metadata={
+        json_schema_extra={
             "help": "The resources per worker for Ray training. Default is to use 1 GPU per worker."
         },
     )
@@ -654,35 +680,49 @@ class AxolotlInputConfig(
 ):
     """wrapper of all config options"""
 
-    class Config:
-        """Config for alias"""
-
-        populate_by_name = True
+    model_config = {"populate_by_name": True}
 
     strict: Optional[bool] = Field(default=False)
     resume_from_checkpoint: Optional[str] = None
     auto_resume_from_checkpoints: Optional[bool] = None
     resize_token_embeddings_to_32x: Optional[bool] = None
     mean_resizing_embeddings: Optional[bool] = False
+    # optionally shrink the embeddings when the tokenizer vocab size is smaller
+    shrink_embeddings: Optional[bool] = None
 
     rl: Optional[RLType] = None
+    trl: Optional[TRLConfig] = Field(
+        default_factory=lambda: TRLConfig(),  # pylint: disable=unnecessary-lambda
+    )
     reward_model: Optional[bool] = None
     process_reward_model: Optional[bool] = None
     num_labels: Optional[int] = None
     dpo_use_weighting: Optional[
         bool
     ] = None  # whether to use weighting in DPO trainer. If none, default is false in the trainer.
+    dpo_use_logits_to_keep: Optional[bool] = None
 
-    datasets: Optional[conlist(Union[SFTDataset, DPODataset, KTODataset, StepwiseSupervisedDataset], min_length=1)] = None  # type: ignore
-    test_datasets: Optional[conlist(Union[SFTDataset, DPODataset, KTODataset, StepwiseSupervisedDataset], min_length=1)] = None  # type: ignore
+    datasets: Optional[
+        Annotated[
+            list[Union[SFTDataset, DPODataset, KTODataset, StepwiseSupervisedDataset]],
+            MinLen(1),
+        ]
+    ] = None
+
+    test_datasets: Optional[
+        Annotated[
+            list[Union[SFTDataset, DPODataset, KTODataset, StepwiseSupervisedDataset]],
+            MinLen(1),
+        ]
+    ] = None
     shuffle_merged_datasets: Optional[bool] = True
     dataset_prepared_path: Optional[str] = None
     dataset_shard_num: Optional[int] = None
     dataset_shard_idx: Optional[int] = None
     skip_prepare_dataset: Optional[bool] = False
 
-    pretraining_dataset: Optional[  # type: ignore
-        conlist(Union[PretrainingDataset, SFTDataset], min_length=1)
+    pretraining_dataset: Optional[
+        Annotated[list[Union[PretrainingDataset, SFTDataset]], MinLen(1)]
     ] = Field(
         default=None,
         json_schema_extra={"description": "streaming dataset to use for pretraining"},
@@ -708,6 +748,8 @@ class AxolotlInputConfig(
     local_rank: Optional[int] = None
     ddp: Optional[bool] = None
 
+    tensor_parallel: Optional[bool] = None
+
     seed: Optional[int] = None
     ddp_timeout: Optional[int] = None
     ddp_bucket_cap_mb: Optional[int] = None
@@ -800,6 +842,10 @@ class AxolotlInputConfig(
     unsloth_rms_norm: Optional[bool] = None
     unsloth_rope: Optional[bool] = None
 
+    lora_mlp_kernel: Optional[bool] = None
+    lora_qkv_kernel: Optional[bool] = None
+    lora_o_kernel: Optional[bool] = None
+
     deepspeed: Optional[Union[str, Dict[str, Any]]] = None
     fsdp: Optional[List[str]] = None
     fsdp_config: Optional[Dict[str, Any]] = None
@@ -822,7 +868,7 @@ class AxolotlInputConfig(
     warmup_steps: Optional[int] = None
     warmup_ratio: Optional[float] = None
     eval_steps: Optional[Union[int, float]] = None
-    evals_per_epoch: Optional[Union[int]] = None
+    evals_per_epoch: Optional[int] = None
     eval_strategy: Optional[str] = None
     save_steps: Optional[Union[int, float]] = None
     saves_per_epoch: Optional[int] = None
@@ -834,6 +880,7 @@ class AxolotlInputConfig(
     save_only_model: Optional[bool] = False
     use_tensorboard: Optional[bool] = None
     profiler_steps: Optional[int] = None
+    include_tokens_per_second: Optional[bool] = None
 
     neftune_noise_alpha: Optional[float] = None
 
@@ -883,10 +930,15 @@ class AxolotlInputConfig(
     @classmethod
     def deprecate_sharegpt_datasets(cls, datasets):
         for _, ds_cfg in enumerate(datasets):
-            if not ds_cfg.get("type"):
+            # Handle both dict and pydantic model cases
+            ds_type = (
+                ds_cfg.get("type")
+                if isinstance(ds_cfg, dict)
+                else getattr(ds_cfg, "type", None)
+            )
+            if not ds_type:
                 continue
 
-            ds_type = ds_cfg["type"]
             # skip if it's a dict (for custom user instruction prompt)
             if isinstance(ds_type, dict):
                 continue
@@ -898,6 +950,14 @@ class AxolotlInputConfig(
 
         return datasets
 
+    @field_serializer("datasets")
+    def datasets_serializer(
+        self, ds_configs: Optional[List[DatasetConfig]]
+    ) -> Optional[List[Dict[str, Any]]]:
+        if ds_configs:
+            return [ds_config.model_dump(exclude_none=True) for ds_config in ds_configs]
+        return None
+
     @model_validator(mode="before")
     @classmethod
     def check_batch_size_fields(cls, data):
@@ -1313,6 +1373,13 @@ class AxolotlInputConfig(
             )
         return data
 
+    @model_validator(mode="before")
+    @classmethod
+    def check_fsdp_tp(cls, data):
+        if data.get("fsdp") and data.get("tensor_parallel"):
+            raise ValueError("FSDP with tensor parallelism is not supported yet.")
+        return data
+
     @model_validator(mode="after")
     def check_fft_possible_bad_config(self):
         if (
@@ -1523,12 +1590,42 @@ class AxolotlInputConfig(
             or data.get("unsloth_lora_qkv")
             or data.get("unsloth_lora_o")
         ):
-            if data.get("adapter") == "lora" or data.get("load_in_8bit"):
+            if data.get("adapter") == "lora" and data.get("load_in_8bit"):
                 raise ValueError(
                     "unsloth_lora_mlp, unsloth_lora_qkv, and unsloth_lora_o are not compatible with 8-bit LoRA"
                 )
         return data
 
+    @model_validator(mode="before")
+    @classmethod
+    def check_lora_8bit(cls, data):
+        if (
+            data.get("lora_mlp_kernel")
+            or data.get("lora_qkv_kernel")
+            or data.get("lora_o_kernel")
+        ):
+            if data.get("adapter") == "lora" and data.get("load_in_8bit"):
+                raise ValueError(
+                    "lora_mlp_kernel, lora_mlp_kernel, and lora_mlp_kernel are not compatible with 8-bit LoRA"
+                )
+        return data
+
+    @model_validator(mode="before")
+    @classmethod
+    def check_lora_axolotl_unsloth(cls, data):
+        is_lora_kernel = any(
+            data.get(k) for k in ["lora_mlp_kernel", "lora_qkv_kernel", "lora_o_kernel"]
+        )
+        is_unsloth_lora = any(
+            data.get(k)
+            for k in ["unsloth_lora_mlp", "unsloth_lora_qkv", "unsloth_lora_o"]
+        )
+        if is_lora_kernel and is_unsloth_lora:
+            raise ValueError(
+                "both lora_mlp_kernel and unsloth_lora_mlp cannot be true (similarly for lora_qkv_kernel, lora_o_kernel)"
+            )
+        return data
+
     @model_validator(mode="before")
     @classmethod
     def check_torch_compile_deepspeed(cls, data):
@@ -1661,6 +1758,29 @@ class AxolotlConfigWCapabilities(AxolotlInputConfig):
                 )
         return data
 
+    @model_validator(mode="before")
+    @classmethod
+    def check_multigpu_lora_kernels(cls, data):
+        if (
+            data.get("lora_mlp_kernel")
+            or data.get("lora_qkv_kernel")
+            or data.get("lora_o_kernel")
+        ):
+            capabilities = data.get("capabilities")
+            is_fsdp = data.get("fsdp") is not None
+            is_deepspeed = data.get("deepspeed") is not None
+
+            if capabilities and capabilities.get("n_gpu", 0) > 1:
+                if is_fsdp:
+                    raise ValueError(
+                        "lora_mlp_kernel, lora_qkv_kernel, and lora_o_kernel are not compatible with FSDP."
+                    )
+                if is_deepspeed:
+                    raise ValueError(
+                        "lora_mlp_kernel, lora_qkv_kernel, and lora_o_kernel are not compatible with DeepSpeed."
+                    )
+        return data
+
     @model_validator(mode="before")
     @classmethod
     def check_adopt_torch_version(cls, data):
@@ -1697,3 +1817,77 @@ class AxolotlConfigWCapabilities(AxolotlInputConfig):
             else:
                 data["torch_compile"] = False
         return data
+
+
+def handle_legacy_message_fields_logic(data: dict) -> dict:
+    """
+    Handle backwards compatibility between legacy message field mapping and new property mapping system.
+
+    Previously, the config only supported mapping 'role' and 'content' fields via dedicated config options:
+    - message_field_role: Mapped to the role field
+    - message_field_content: Mapped to the content field
+
+    The new system uses message_property_mappings to support arbitrary field mappings:
+    message_property_mappings:
+        role: source_role_field
+        content: source_content_field
+        additional_field: source_field
+
+    Args:
+        data: Dictionary containing configuration data
+
+    Returns:
+        Updated dictionary with message field mappings consolidated
+
+    Raises:
+        ValueError: If there are conflicts between legacy and new mappings
+    """
+    data = data.copy()  # Create a copy to avoid modifying the original
+
+    if data.get("message_property_mappings") is None:
+        data["message_property_mappings"] = {}
+
+    # Check for conflicts and handle role
+    if "message_field_role" in data:
+        LOG.warning(
+            "message_field_role is deprecated, use message_property_mappings instead. "
+            f"Example: message_property_mappings: {{role: {data['message_field_role']}}}"
+        )
+        if (
+            "role" in data["message_property_mappings"]
+            and data["message_property_mappings"]["role"] != data["message_field_role"]
+        ):
+            raise ValueError(
+                f"Conflicting message role fields: message_field_role='{data['message_field_role']}' "
+                f"conflicts with message_property_mappings.role='{data['message_property_mappings']['role']}'"
+            )
+        data["message_property_mappings"]["role"] = data["message_field_role"] or "role"
+
+        del data["message_field_role"]
+    elif "role" not in data["message_property_mappings"]:
+        data["message_property_mappings"]["role"] = "role"
+
+    # Check for conflicts and handle content
+    if "message_field_content" in data:
+        LOG.warning(
+            "message_field_content is deprecated, use message_property_mappings instead. "
+            f"Example: message_property_mappings: {{content: {data['message_field_content']}}}"
+        )
+        if (
+            "content" in data["message_property_mappings"]
+            and data["message_property_mappings"]["content"]
+            != data["message_field_content"]
+        ):
+            raise ValueError(
+                f"Conflicting message content fields: message_field_content='{data['message_field_content']}' "
+                f"conflicts with message_property_mappings.content='{data['message_property_mappings']['content']}'"
+            )
+        data["message_property_mappings"]["content"] = (
+            data["message_field_content"] or "content"
+        )
+
+        del data["message_field_content"]
+    elif "content" not in data["message_property_mappings"]:
+        data["message_property_mappings"]["content"] = "content"
+
+    return data

src/axolotl/utils/config/models/input/v0_4_1/trl.py

@@ -0,0 +1,35 @@
+"""
+GRPO specific configuration args
+"""
+from typing import List, Optional
+
+from pydantic import BaseModel, Field
+
+
+class TRLConfig(BaseModel):
+    """
+    Input args for TRL.
+    """
+
+    beta: Optional[float] = None
+    max_completion_length: Optional[int] = Field(
+        default=None,
+        json_schema_extra={
+            "description": "Maximum length of the completion for RL training"
+        },
+    )
+
+    # GRPO specific args
+    use_vllm: Optional[bool] = False
+    vllm_device: Optional[str] = "auto"
+    vllm_gpu_memory_utilization: Optional[float] = 0.9
+    vllm_max_model_len: Optional[int] = None
+    vllm_dtype: Optional[str] = "auto"
+
+    reward_funcs: Optional[List[str]] = None
+    num_generations: Optional[int] = None
+    log_completions: Optional[bool] = False
+
+    sync_ref_model: Optional[bool] = False
+    ref_model_mixup_alpha: Optional[float] = 0.9
+    ref_model_sync_steps: Optional[int] = 64

src/axolotl/utils/data/rl.py

@@ -4,15 +4,16 @@ import inspect
 import logging
 from functools import partial
 from pathlib import Path
-from typing import Any, List
+from typing import Any, List, Union
 
 import yaml
-from datasets import DatasetDict, concatenate_datasets, load_dataset, load_from_disk
+from datasets import Dataset, DatasetDict, concatenate_datasets, load_from_disk
 
 from axolotl.common.const import DEFAULT_DATASET_PREPARED_PATH
 from axolotl.prompt_strategies.dpo import load as load_dpo
 from axolotl.prompt_strategies.kto import load as load_kto
 from axolotl.prompt_strategies.orpo import load as load_orpo
+from axolotl.utils.data.shared import datasets_w_name_generator, load_dataset_w_config
 from axolotl.utils.data.utils import deduplicate_and_log_datasets, md5
 from axolotl.utils.dict import DictDefault
 from axolotl.utils.distributed import is_main_process, zero_first
@@ -57,7 +58,7 @@ def _save_preprocessed_ds(cfg, sub_cfg, dataset):
         dataset.save_to_disk(str(prepared_ds_path))
 
 
-def map_dataset(cfg, data_set, ds_transform_fn, tokenizer):
+def map_dataset(cfg, data_set, ds_transform_fn, tokenizer, **map_kwargs):
     sig = inspect.signature(ds_transform_fn)
     if "tokenizer" in sig.parameters:
         if not tokenizer:
@@ -70,6 +71,7 @@ def map_dataset(cfg, data_set, ds_transform_fn, tokenizer):
     data_set = data_set.map(
         ds_transform_fn,
         desc="Mapping RL Dataset",
+        **map_kwargs,
     )
 
     return data_set
@@ -112,29 +114,21 @@ def drop_long_rl_seq(
 
         return (len_prompt + len_completion) <= sequence_len
 
+    if rl == "grpo":
+        return True
+
     raise ValueError("Unknown RL type")
 
 
 def load_prepare_preference_datasets(cfg):
     def load_split(dataset_cfgs, _cfg):
         split_datasets: List[Any] = []
-        for i, ds_cfg in enumerate(dataset_cfgs):
-            if ds_cfg["ds_type"] == "json":
-                for data_file in ds_cfg["data_files"]:
-                    data_files = {ds_cfg["split"]: data_file}
-                    ds = load_dataset(  # pylint: disable=invalid-name
-                        "json",
-                        data_files=data_files,
-                        split=ds_cfg["split"],
-                    )
-                    split_datasets.insert(i, ds)
-            else:
-                ds = load_dataset(  # pylint: disable=invalid-name
-                    ds_cfg["path"],
-                    split=ds_cfg["split"],
-                    revision=ds_cfg.get("revision", None),
-                )
-                split_datasets.insert(i, ds)
+        use_auth_token = _cfg.hf_use_auth_token
+        for config_dataset in datasets_w_name_generator(dataset_cfgs):
+            ds: Union[Dataset, DatasetDict] = load_dataset_w_config(
+                config_dataset, use_auth_token, streaming=False
+            )
+            split_datasets.append(ds)
 
         tokenizer = load_tokenizer(cfg)
 
@@ -150,36 +144,45 @@ def load_prepare_preference_datasets(cfg):
                 else:
                     ds_transform_fn = load_dpo(_type, _cfg, dataset_idx=i)
 
+                map_kwargs = {}
+                if isinstance(ds_transform_fn, tuple):
+                    ds_transform_fn, map_kwargs = ds_transform_fn
                 split_datasets[i] = map_dataset(
-                    cfg, data_set, ds_transform_fn, tokenizer
+                    cfg, data_set, ds_transform_fn, tokenizer, **map_kwargs
                 )
             elif _cfg.rl == "kto":
                 ds_transform_fn = load_kto(_type, _cfg, dataset_idx=i)
+                map_kwargs = {}
+                if isinstance(ds_transform_fn, tuple):
+                    ds_transform_fn, map_kwargs = ds_transform_fn
                 split_datasets[i] = map_dataset(
-                    cfg, data_set, ds_transform_fn, tokenizer
+                    cfg, data_set, ds_transform_fn, tokenizer, **map_kwargs
                 )
             else:
                 # If no `type` is provided, assume the dataset is already in the expected format with
                 # "prompt", "chosen" and "rejected" already preprocessed
                 split_datasets[i] = data_set
 
-            drop_long = partial(
-                drop_long_rl_seq,
-                rl=_cfg.rl,
-                tokenizer=tokenizer,
-                sequence_len=cfg.sequence_len,
-            )
+            if not cfg.skip_prepare_dataset:
+                drop_long = partial(
+                    drop_long_rl_seq,
+                    rl=_cfg.rl,
+                    tokenizer=tokenizer,
+                    sequence_len=cfg.sequence_len,
+                )
 
-            prior_len = len(split_datasets[i])
-            split_datasets[i] = split_datasets[i].filter(
-                drop_long,
-                num_proc=cfg.dataset_processes,
-                load_from_cache_file=not cfg.is_preprocess,
-                desc="Dropping Long Sequences",
-            )
-            dropped = prior_len - len(split_datasets[i])
-            if dropped:
-                LOG.warning(f"Dropped {dropped} long samples from dataset index {i}")
+                prior_len = len(split_datasets[i])
+                split_datasets[i] = split_datasets[i].filter(
+                    drop_long,
+                    num_proc=cfg.dataset_processes,
+                    load_from_cache_file=not cfg.is_preprocess,
+                    desc="Dropping Long Sequences",
+                )
+                dropped = prior_len - len(split_datasets[i])
+                if dropped:
+                    LOG.warning(
+                        f"Dropped {dropped} long samples from dataset index {i}"
+                    )
 
         combined_datasets = concatenate_datasets(split_datasets)
         combined_datasets = combined_datasets.shuffle(seed=cfg.seed)

src/axolotl/utils/data/sft.py

@@ -43,7 +43,7 @@ from axolotl.prompters import (
     UnsupportedPrompter,
 )
 from axolotl.utils.data.pretraining import wrap_pretraining_dataset
-from axolotl.utils.data.shared import load_dataset_w_config
+from axolotl.utils.data.shared import datasets_w_name_generator, load_dataset_w_config
 from axolotl.utils.data.utils import (
     deduplicate_and_log_datasets,
     drop_long_seq_in_dataset,
@@ -180,6 +180,7 @@ def load_tokenized_prepared_datasets(
 ) -> Tuple[DatasetDict, List[Prompter]]:
     cfg_datasets = cfg.test_datasets if split == "test" else cfg.datasets
     tokenizer_name = cfg.tokenizer_config
+
     ds_hash = str(
         md5(
             (
@@ -263,30 +264,11 @@ def load_tokenized_prepared_datasets(
 
         datasets = []
 
-        def for_d_in_datasets(dataset_configs):
-            for dataset in dataset_configs:
-                if dataset.name and isinstance(dataset.name, list):
-                    # load_dataset doesn't properly handle multiple named configurations
-                    # at the same time for a given dataset
-                    for name in dataset.name:
-                        yield DictDefault({**dataset, "name": name})
-                elif dataset.preprocess_shards and not dataset.shards:
-                    for shard in range(dataset.preprocess_shards):
-                        yield DictDefault(
-                            {
-                                **dataset,
-                                "shards": dataset.preprocess_shards,
-                                "shards_idx": shard,
-                            }
-                        )
-                else:
-                    yield dataset
-
         streaming_ds = False
         if preprocess_iterable:
             streaming_ds = True
         # pylint: disable=invalid-name
-        for config_dataset in for_d_in_datasets(cfg_datasets):
+        for config_dataset in datasets_w_name_generator(cfg_datasets):
             ds: Union[Dataset, DatasetDict] = load_dataset_w_config(
                 config_dataset, use_auth_token, streaming=streaming_ds
             )

src/axolotl/utils/data/shared.py

@@ -1,6 +1,7 @@
 """
 dataset loading shared utils
 """
+
 from pathlib import Path
 from typing import Optional, Union
 
@@ -29,9 +30,43 @@ def get_ds_type(config_dataset: DictDefault):
     return ds_type
 
 
+def datasets_w_name_generator(dataset_configs: list[DictDefault]):
+    """
+    Yields dataset configs handling multiple names or preprocess_shards
+
+    Args:
+        dataset_configs: list of dataset configs (equivalent to cfg.datasets)
+    """
+    for dataset in dataset_configs:
+        if dataset.name and isinstance(dataset.name, list):
+            # load_dataset doesn't properly handle multiple named configurations
+            # at the same time for a given dataset
+            for name in dataset.name:
+                yield DictDefault({**dataset, "name": name})
+        elif dataset.preprocess_shards and not dataset.shards:
+            for shard in range(dataset.preprocess_shards):
+                yield DictDefault(
+                    {
+                        **dataset,
+                        "shards": dataset.preprocess_shards,
+                        "shards_idx": shard,
+                    }
+                )
+        else:
+            yield dataset
+
+
 def load_dataset_w_config(
-    config_dataset, auth_token, streaming=False
+    config_dataset: DictDefault, use_auth_token: bool, streaming=False
 ) -> Union[Dataset, DatasetDict]:
+    """
+    Load a dataset from a config
+
+    Args:
+        config_dataset: single dataset config
+        use_auth_token: whether to use HF auth token
+        streaming: whether to stream the dataset
+    """
     # pylint: disable=invalid-name
     ds: Optional[Union[Dataset, DatasetDict]] = None  # pylint: disable=invalid-name
     ds_from_hub = False
@@ -43,7 +78,7 @@ def load_dataset_w_config(
             config_dataset.path,
             name=config_dataset.name,
             streaming=True,
-            token=auth_token,
+            token=use_auth_token,
             revision=config_dataset.revision,
             trust_remote_code=ds_trust_remote_code,
         )
@@ -161,7 +196,7 @@ def load_dataset_w_config(
             name=config_dataset.name,
             streaming=streaming,
             data_files=config_dataset.data_files,
-            token=auth_token,
+            token=use_auth_token,
             revision=config_dataset.revision,
             trust_remote_code=config_dataset.trust_remote_code,
             **load_ds_kwargs,

src/axolotl/utils/data/utils.py

@@ -172,10 +172,11 @@ def drop_long_seq_in_dataset(dataset: Dataset, cfg: DictDefault):
     )
 
     try:
-        min_input_len = np.min(get_dataset_lengths(dataset))
-        LOG.debug(f"min_input_len: {min_input_len}")
-        max_input_len = np.max(get_dataset_lengths(dataset))
-        LOG.debug(f"max_input_len: {max_input_len}")
+        ds_lengths = get_dataset_lengths(dataset, from_arrow=True)
+        min_input_len = np.min(ds_lengths)
+        LOG.info(f"min_input_len: {min_input_len}")
+        max_input_len = np.max(ds_lengths)
+        LOG.info(f"max_input_len: {max_input_len}")
     except AttributeError:
         pass
 

src/axolotl/utils/dict.py

@@ -13,3 +13,26 @@ class DictDefault(Dict):
 
     def __or__(self, other):
         return DictDefault(super().__ror__(other))
+
+    def __setitem__(self, name, value):
+        # workaround for pickle/unpickle issues and __frozen not being available
+        try:
+            isFrozen = hasattr(  # pylint: disable=invalid-name
+                self, "__frozen"
+            ) and object.__getattribute__(self, "__frozen")
+        except AttributeError:
+            isFrozen = False  # pylint: disable=invalid-name
+
+        if isFrozen and name not in super().keys():
+            raise KeyError(name)
+        super(Dict, self).__setitem__(name, value)  # pylint: disable=bad-super-call
+        try:
+            p = object.__getattribute__(self, "__parent")
+            key = object.__getattribute__(self, "__key")
+        except AttributeError:
+            p = None
+            key = None
+        if p is not None:
+            p[key] = self
+            object.__delattr__(self, "__parent")
+            object.__delattr__(self, "__key")

src/axolotl/utils/lora.py

@@ -0,0 +1,75 @@
+# Copyright 2025 Axolotl AI. 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.
+
+"""
+module to get the state dict of a merged lora model
+"""
+import torch
+from peft.tuners.tuners_utils import onload_layer
+from peft.utils import ModulesToSaveWrapper, _get_submodules
+
+
+def get_lora_merged_state_dict(
+    model: torch.nn.Module,
+) -> dict:
+    r"""
+    Create and return a state_dict that has the LoRA deltas
+    merged into the base model’s weights, without modifying `model` in place.
+
+    Arguments:
+        model (torch.nn.Module): A model that has LoRA/PEFT adapters attached.
+
+    Returns:
+        dict: A state_dict of the merged parameters.
+    """
+
+    base_model_prefix = "base_model.model."
+    state_dict = {}
+    key_list = [key for key, _ in model.named_modules() if model.prefix not in key]
+    for key in key_list:
+        try:
+            _, target, _ = _get_submodules(model, key)
+        except AttributeError:
+            continue
+        with onload_layer(target):
+            weight_key = key.replace(base_model_prefix, "") + ".weight"
+            bias_key = key.replace(base_model_prefix, "") + ".bias"
+            if hasattr(target, "base_layer"):
+                target.merge(safe_merge=True, adapter_names=None)
+                # get the state_dict of target.base_layer
+                layer_state_dict = target.base_layer.state_dict()
+                state_dict[weight_key] = layer_state_dict["weight"]
+            elif isinstance(target, ModulesToSaveWrapper):
+                # save any additional trainable modules part of `modules_to_save`
+                new_module = target.modules_to_save[target.active_adapter]
+                if hasattr(new_module, "base_layer"):
+                    # check if the module is itself a tuner layer
+                    new_module.merge(safe_merge=True, adapter_names=None)
+                layer_state_dict = new_module.state_dict()
+                state_dict[weight_key] = layer_state_dict["weight"]
+            elif hasattr(target, "weight"):
+                if any(
+                    skip in key
+                    for skip in [
+                        ".original_module",
+                        ".modules_to_save",
+                        ".base_layer",
+                    ]
+                ):
+                    continue
+                layer_state_dict = target.state_dict()
+                state_dict[weight_key] = layer_state_dict["weight"]
+                if hasattr(target, "bias") and "bias" in layer_state_dict.keys():
+                    state_dict[bias_key] = layer_state_dict["bias"]
+    return state_dict

src/axolotl/utils/models.py

@@ -414,6 +414,7 @@ class ModelLoader:
                 has_remote_code = "AutoModelForCausalLM" in auto_map_config
             else:
                 has_remote_code = False
+
             if has_remote_code and self.cfg.trust_remote_code is False:
                 # if explicitly set in the YAML, we should prefer that, for example if explicitly disabled
                 has_remote_code = self.cfg.trust_remote_code
@@ -425,10 +426,6 @@ class ModelLoader:
 
             if self.cfg.is_llama_derived_model:
                 self.patch_loss_llama()
-                if self.cfg.unsloth_lora_qkv or self.cfg.unsloth_lora_o:
-                    from axolotl.monkeypatch.unsloth_ import patch_self_attn_lora
-
-                    patch_self_attn_lora()
         elif self.cfg.is_llama_derived_model:
             self.patch_llama_derived_model()
 
@@ -442,6 +439,11 @@ class ModelLoader:
 
             patch_mistral_cross_entropy()
 
+        if self.cfg.unsloth_lora_qkv or self.cfg.unsloth_lora_o:
+            from axolotl.monkeypatch.lora_kernels import patch_self_attn_lora
+
+            patch_self_attn_lora(self.cfg)
+
     def patch_attention(self) -> None:
         if hasattr(self.model_config, "model_type"):
             if self.model_config.model_type == "mllama" and self.cfg.flash_attention:
@@ -472,9 +474,7 @@ class ModelLoader:
         return importlib.util.find_spec("flash_attn") is not None
 
     def patch_loss_llama(self) -> None:
-        """
-        Patch loss functions
-        """
+        """Patch loss functions and other optimizations"""
         if self.has_flash_attn:
             from axolotl.monkeypatch.llama_attn_hijack_flash import (
                 patch_fa_llama_cross_entropy,
@@ -494,15 +494,14 @@ class ModelLoader:
             from axolotl.monkeypatch.unsloth_ import patch_unsloth_layernorm
 
             patch_unsloth_layernorm()
+
         if self.cfg.unsloth_lora_qkv or self.cfg.unsloth_lora_o:
             from axolotl.monkeypatch.unsloth_ import patch_self_attn_lora
 
             patch_self_attn_lora()
 
     def patch_llama_derived_model(self) -> None:
-        """
-        Modify all llama derived models in one block
-        """
+        """Modify all llama derived models in one block"""
         self.patch_loss_llama()
 
         if self.cfg.flash_attention:
@@ -763,6 +762,9 @@ class ModelLoader:
             return hf_ds_cfg
 
         skip_move_to_device = False
+        if self.cfg.tensor_parallel:
+            del self.model_kwargs["device_map"]
+
         if (  # pylint: disable=condition-evals-to-constant)
             (self.cfg.fsdp and self.cfg.fsdp_config.fsdp_cpu_ram_efficient_loading)
             and not qlora_fsdp
@@ -1013,7 +1015,8 @@ class ModelLoader:
                 if hasattr(module, "weight"):
                     module.to(dist_dtype)
 
-    def apply_lora_patch(self) -> None:
+    # TODO: Deprecate this.
+    def apply_unsloth_lora_patch(self) -> None:
         if self.cfg.unsloth_lora_mlp:
             from axolotl.monkeypatch.unsloth_ import integrate_lora_mlp_patch
 
@@ -1027,6 +1030,16 @@ class ModelLoader:
 
             integrate_rope_embeddings()
 
+    def apply_lora_patch(self) -> None:
+        if (
+            self.cfg.lora_mlp_kernel
+            or self.cfg.lora_qkv_kernel
+            or self.cfg.lora_o_kernel
+        ):
+            from axolotl.monkeypatch.lora_kernels import apply_lora_kernel_patches
+
+            apply_lora_kernel_patches(self.model, self.cfg)
+
     def load_model(self) -> Tuple[PreTrainedModel, Optional[PeftConfig]]:
         self.apply_patches()
         self.set_auto_model_loader()
@@ -1053,9 +1066,12 @@ class ModelLoader:
             if self.cfg.resize_token_embeddings_to_32x
             else len(self.tokenizer)
         )
-        if (
-            hasattr(self.model, "get_input_embeddings")
-            and self.model.get_input_embeddings().num_embeddings != embeddings_len
+        if hasattr(self.model, "get_input_embeddings") and (
+            self.model.get_input_embeddings().num_embeddings < embeddings_len
+            or (
+                self.model.get_input_embeddings().num_embeddings > embeddings_len
+                and self.cfg.shrink_embeddings
+            )
         ):
             resize_kwargs = {}
             if self.cfg.mean_resizing_embeddings is not None:
@@ -1168,6 +1184,7 @@ class ModelLoader:
         if self.cfg.adapter is not None:
             log_gpu_memory_usage(LOG, "after adapters", self.model.device)
 
+        self.apply_unsloth_lora_patch()
         self.apply_lora_patch()
 
         for _ in range(3):
@@ -1307,8 +1324,11 @@ def load_lora(model, cfg, inference=False, config_only=False):
     if loftq_bits:
         lora_config_kwargs["loftq_config"] = LoftQConfig(loftq_bits=loftq_bits)
         lora_config_kwargs["init_lora_weights"] = "loftq"
+    if cfg.peft_init_lora_weights:
+        lora_config_kwargs["init_lora_weights"] = cfg.peft_init_lora_weights
     if cfg.peft_use_dora:
         lora_config_kwargs["use_dora"] = cfg.peft_use_dora
+        LOG.info("Initializing LoRA weights using dora. This might take longer.")
     if cfg.peft_use_rslora:
         lora_config_kwargs["use_rslora"] = cfg.peft_use_rslora
     if cfg.peft_layer_replication:

src/axolotl/utils/samplers/utils.py

@@ -4,13 +4,17 @@ helper util to calculate dataset lengths
 import numpy as np
 
 
-def get_dataset_lengths(dataset):
-    if "length" in dataset.data.column_names:
-        lengths = np.array(dataset.data.column("length"))
-    elif "position_ids" in dataset.data.column_names:
-        position_ids = dataset.data.column("position_ids")
+def get_dataset_lengths(dataset, from_arrow=False):
+    if "length" in dataset.column_names:
+        lengths = np.array(dataset["length"])
+    elif "position_ids" in dataset.column_names:
+        position_ids = dataset["position_ids"]
         lengths = np.array([x[-1] + 1 for x in position_ids])
     else:
-        input_ids = dataset.data.column("input_ids")
-        lengths = np.vectorize(len)(np.array(input_ids, dtype=object))
+        if from_arrow:
+            input_ids = dataset.data.column("input_ids")
+            lengths = np.vectorize(len)(np.array(input_ids, dtype=object))
+        else:
+            input_ids = dataset["input_ids"]
+            lengths = np.array([len(seq) for seq in input_ids])
     return lengths