reworking SP logic into composed handler

.github/workflows/base.yml

@@ -52,12 +52,6 @@ jobs:
             python_version: "3.11"
             pytorch: nightly
             torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-          - cuda: "128"
-            cuda_version: 12.8.1
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: next
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -79,7 +73,7 @@ jobs:
         uses: docker/build-push-action@v4
         with:
           context: .
-          file: ${{ matrix.pytorch == 'nightly' && './docker/Dockerfile-base-nightly' || matrix.pytorch == 'next' && './docker/Dockerfile-base-next' || './docker/Dockerfile-base' }}
+          file: ${{ matrix.pytorch == 'nightly' && './docker/Dockerfile-base-nightly' || './docker/Dockerfile-base' }}
           push: ${{ github.event_name != 'pull_request' }}
           tags: ${{ steps.metadata.outputs.tags }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
           labels: ${{ steps.metadata.outputs.labels }}

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

@@ -24,13 +24,6 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 124
-            cuda_version: 12.4.1
-            python_version: "3.11"
-            pytorch: 2.6.0
-            axolotl_extras: vllm
-            num_gpus: 2
-            nightly_build: "true"
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"
@@ -45,6 +38,13 @@ jobs:
             axolotl_extras: vllm
             num_gpus: 2
             nightly_build: "true"
+          - cuda: 124
+            cuda_version: 12.4.1
+            python_version: "3.11"
+            pytorch: 2.6.0
+            axolotl_extras: vllm
+            num_gpus: 2
+            nightly_build: "true"
     runs-on: [self-hosted, modal]
     timeout-minutes: 120
     steps:

.github/workflows/tests.yml

@@ -211,7 +211,7 @@ jobs:
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"
-            pytorch: 2.6.0
+            pytorch: 2.5.1
             num_gpus: 1
             axolotl_extras: vllm
     steps:
@@ -258,7 +258,7 @@ jobs:
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"
-            pytorch: 2.5.1
+            pytorch: 2.6.0
             num_gpus: 1
             axolotl_extras: vllm
     steps:

_quarto.yml

@@ -231,7 +231,6 @@ website:
             - docs/reward_modelling.qmd
             - docs/lr_groups.qmd
             - docs/lora_optims.qmd
-            - docs/dataset_loading.qmd
 
         - section: "Core Concepts"
           contents:

cicd/multigpu.sh

@@ -2,5 +2,5 @@
 set -e
 
 # only run one test at a time so as not to OOM the GPU
-pytest -v  --durations=10 -n2 /workspace/axolotl/tests/e2e/multigpu/ --ignore=/workspace/axolotl/tests/e2e/multigpu/solo/
-pytest -v  --durations=10 -n1 /workspace/axolotl/tests/e2e/multigpu/solo/
+pytest -v -n2 /workspace/axolotl/tests/e2e/multigpu/ --ignore=/workspace/axolotl/tests/e2e/multigpu/solo/
+pytest -v -n1 /workspace/axolotl/tests/e2e/multigpu/solo/

docker/Dockerfile

@@ -20,9 +20,9 @@ WORKDIR /workspace/axolotl
 
 # If AXOLOTL_EXTRAS is set, append it in brackets
 RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
-        pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
+        pip install --no-build-isolation -e .[deepspeed,flash-attn,optimizers,ray,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
     else \
-        pip install --no-build-isolation -e .[deepspeed,flash-attn,ring-flash-attn,optimizers,ray] $AXOLOTL_ARGS; \
+        pip install --no-build-isolation -e .[deepspeed,flash-attn,optimizers,ray] $AXOLOTL_ARGS; \
     fi
 
 RUN python scripts/unsloth_install.py | sh

docker/Dockerfile-base

@@ -29,7 +29,7 @@ ENV PATH="/root/miniconda3/envs/py${PYTHON_VERSION}/bin:${PATH}"
 WORKDIR /workspace
 
 RUN python3 -m pip install --upgrade pip && pip3 install -U packaging==23.2 setuptools==75.8.0 wheel && \
-    python3 -m pip install --no-cache-dir -U torch==${PYTORCH_VERSION}+cu${CUDA} torchvision --extra-index-url https://download.pytorch.org/whl/cu$CUDA && \
+    python3 -m pip install --no-cache-dir -U torch==${PYTORCH_VERSION}+cu${CUDA} --extra-index-url https://download.pytorch.org/whl/cu$CUDA && \
     python3 -m pip install --no-cache-dir "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@main" && \
     python3 -m pip install --no-cache-dir "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main"
 

docker/Dockerfile-base-next

@@ -1,38 +0,0 @@
-ARG CUDA_VERSION="12.8.1"
-ARG CUDNN_VERSION="8"
-ARG UBUNTU_VERSION="22.04"
-ARG MAX_JOBS=4
-
-FROM nvidia/cuda:$CUDA_VERSION-cudnn$CUDNN_VERSION-devel-ubuntu$UBUNTU_VERSION AS base-builder
-
-ENV PATH="/root/miniconda3/bin:${PATH}"
-
-ARG PYTHON_VERSION="3.11"
-ARG PYTORCH_VERSION="next"
-ARG CUDA="128"
-ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
-
-ENV PYTHON_VERSION=$PYTHON_VERSION
-ENV TORCH_CUDA_ARCH_LIST=$TORCH_CUDA_ARCH_LIST
-
-RUN apt-get update \
-    && apt-get install -y wget git build-essential ninja-build git-lfs libaio-dev pkg-config && rm -rf /var/lib/apt/lists/* \
-    && wget \
-    https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh \
-    && mkdir /root/.conda \
-    && bash Miniconda3-latest-Linux-x86_64.sh -b \
-    && rm -f Miniconda3-latest-Linux-x86_64.sh \
-    && conda create -n "py${PYTHON_VERSION}" python="${PYTHON_VERSION}"
-
-ENV PATH="/root/miniconda3/envs/py${PYTHON_VERSION}/bin:${PATH}"
-
-WORKDIR /workspace
-
-RUN python3 -m pip install --upgrade pip && pip3 install packaging && \
-    python3 -m pip install --no-cache-dir -U torch==2.7.0 --extra-index-url https://download.pytorch.org/whl/test/cu$CUDA && \
-    python3 -m pip install --no-cache-dir "causal_conv1d @ git+https://github.com/Dao-AILab/causal-conv1d.git@main" && \
-    python3 -m pip install --no-cache-dir "mamba_ssm @ git+https://github.com/state-spaces/mamba.git@main"
-
-RUN git lfs install --skip-repo && \
-    pip3 install awscli && \
-    pip3 install -U --no-cache-dir pydantic==2.10.6

docs/config.qmd

@@ -109,7 +109,7 @@ datasets:
     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
-    split: train # Optional[str] name of dataset split to load from
+    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.
     trust_remote_code: # Optional[bool] Trust remote code for untrusted source
 
@@ -165,9 +165,7 @@ datasets:
       content: value
       # ...
 
-    # Optional[Dict[str, List]]. Roles mapping in the messages.
-    # The format is {target_role: [source_roles]}. All source roles will be mapped to the target role.
-    # The default is:
+    # Optional[Dict[str, List]]. Roles mapping in the messages. The default is:
     roles:
       user: ["human", "user"]
       assistant: ["gpt", "assistant"]
@@ -512,8 +510,7 @@ train_on_inputs: false
 # Note that training loss may have an oscillating pattern with this enabled.
 group_by_length: false
 
-# Whether to use gradient checkpointing. Available options are: true, false, "offload".
-# https://huggingface.co/docs/transformers/v4.18.0/en/performance#gradient-checkpointing
+# Whether to use gradient checkpointing https://huggingface.co/docs/transformers/v4.18.0/en/performance#gradient-checkpointing
 gradient_checkpointing: false
 # additional kwargs to pass to the trainer for gradient checkpointing
 # gradient_checkpointing_kwargs:
@@ -689,9 +686,10 @@ ddp_broadcast_buffers:
 # E.g., if 4 GPUs are available, set this value to 2 to split each sequence into two equal-sized
 # subsequences, or set to 4 to split into four equal-sized subsequences.
 # See https://axolotl-ai-cloud.github.io/axolotl/docs/sequence_parallelism.html for more details.
-sequence_parallel_degree:
-# Optional; strides across the key dimension. Larger values use more memory but should make training faster.
-# Must evenly divide the number of KV heads in your model.
+sequence_parallel_degree: 4  # Set to the number of GPUs to split sequences across
+flash_attention: true  # SP requires flash attention
+micro_batch_size: 1  # SP requires this is set to 1
+# (optional) strides across the key dimension; larger values use more memory but should make training a bit faster
 heads_k_stride: 1
 
 # Path to torch distx for optim 'adamw_anyprecision'

docs/dataset-formats/index.qmd

@@ -13,13 +13,6 @@ As there are a lot of available options in Axolotl, this guide aims to provide a
 
 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.
 
-::: {.callout-tip}
-
-This guide will mainly use JSONL as an introduction. Please refer to the [dataset loading docs](../dataset_loading.qmd) to understand how to load datasets from other sources.
-
-For `pretraining_dataset:` specifically, please refer to the [Pre-training section](#pre-training).
-:::
-
 ## 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.

docs/dataset_loading.qmd

@@ -1,276 +0,0 @@
----
-title: Dataset Loading
-description: Understanding how to load datasets from different sources
-back-to-top-navigation: true
-toc: true
-toc-depth: 5
----
-
-## Overview
-
-Datasets can be loaded in a number of different ways depending on the how it is saved (the extension of the file) and where it is stored.
-
-## Loading Datasets
-
-We use the `datasets` library to load datasets and a mix of `load_dataset` and `load_from_disk` to load them.
-
-You may recognize the similar named configs between `load_dataset` and the `datasets` section of the config file.
-
-```yaml
-datasets:
-  - path:
-    name:
-    data_files:
-    split:
-    revision:
-    trust_remote_code:
-```
-
-::: {.callout-tip}
-
-Do not feel overwhelmed by the number of options here. A lot of them are optional. In fact, the most common config to use would be `path` and sometimes `data_files`.
-
-:::
-
-This matches the API of [`datasets.load_dataset`](https://github.com/huggingface/datasets/blob/0b5998ac62f08e358f8dcc17ec6e2f2a5e9450b6/src/datasets/load.py#L1838-L1858), so if you're familiar with that, you will feel right at home.
-
-For HuggingFace's guide to load different dataset types, see [here](https://huggingface.co/docs/datasets/loading).
-
-For full details on the config, see [config.qmd](config.qmd).
-
-::: {.callout-note}
-
-You can set multiple datasets in the config file by more than one entry under `datasets`.
-
-```yaml
-datasets:
-  - path: /path/to/your/dataset
-  - path: /path/to/your/other/dataset
-```
-
-:::
-
-### Local dataset
-
-#### Files
-
-Usually, to load a JSON file, you would do something like this:
-
-```python
-from datasets import load_dataset
-
-dataset = load_dataset("json", data_files="data.json")
-```
-
-Which translates to the following config:
-
-```yaml
-datasets:
-  - path: json
-    data_files: /path/to/your/file.jsonl
-```
-
-However, to make things easier, we have added a few shortcuts for loading local dataset files.
-
-You can just point the `path` to the file or directory along with the `ds_type` to load the dataset. The below example shows for a JSON file:
-
-```yaml
-datasets:
-  - path: /path/to/your/file.jsonl
-    ds_type: json
-```
-
-This works for CSV, JSON, Parquet, and Arrow files.
-
-::: {.callout-tip}
-
-If `path` points to a file and `ds_type` is not specified, we will automatically infer the dataset type from the file extension, so you could omit `ds_type` if you'd like.
-
-:::
-
-#### Directory
-
-If you're loading a directory, you can point the `path` to the directory.
-
-Then, you have two options:
-
-##### Loading entire directory
-
-You do not need any additional configs.
-
-We will attempt to load in the following order:
-- datasets saved with `datasets.save_to_disk`
-- loading entire directory of files (such as with parquet/arrow files)
-
-```yaml
-datasets:
-  - path: /path/to/your/directory
-```
-
-##### Loading specific files in directory
-
-Provide `data_files` with a list of files to load.
-
-```yaml
-datasets:
-    # single file
-  - path: /path/to/your/directory
-    ds_type: csv
-    data_files: file1.csv
-
-    # multiple files
-  - path: /path/to/your/directory
-    ds_type: json
-    data_files:
-      - file1.jsonl
-      - file2.jsonl
-
-    # multiple files for parquet
-  - path: /path/to/your/directory
-    ds_type: parquet
-    data_files:
-      - file1.parquet
-      - file2.parquet
-
-```
-
-### HuggingFace Hub
-
-The method you use to load the dataset depends on how the dataset was created, whether a folder was uploaded directly or a HuggingFace Dataset was pushed.
-
-::: {.callout-note}
-
-If you're using a private dataset, you will need to enable the `hf_use_auth_token` flag in the root-level of the config file.
-
-:::
-
-#### Folder uploaded
-
-This would mean that the dataset is a single file or file(s) uploaded to the Hub.
-
-```yaml
-datasets:
-  - path: org/dataset-name
-    data_files:
-      - file1.jsonl
-      - file2.jsonl
-```
-
-#### HuggingFace Dataset
-
-This means that the dataset is created as a HuggingFace Dataset and pushed to the Hub via `datasets.push_to_hub`.
-
-```yaml
-datasets:
-  - path: org/dataset-name
-```
-
-::: {.callout-note}
-
-There are some other configs which may be required like `name`, `split`, `revision`, `trust_remote_code`, etc depending on the dataset.
-
-:::
-
-### Remote Filesystems
-
-Via the `storage_options` config under `load_dataset`, you can load datasets from remote filesystems like S3, GCS, Azure, and OCI.
-
-::: {.callout-warning}
-
-This is currently experimental. Please let us know if you run into any issues!
-
-:::
-
-The only difference between the providers is that you need to prepend the path with the respective protocols.
-
-```yaml
-datasets:
-    # Single file
-  - path: s3://bucket-name/path/to/your/file.jsonl
-
-    # Directory
-  - path: s3://bucket-name/path/to/your/directory
-```
-
-For directory, we load via `load_from_disk`.
-
-#### S3
-
-Prepend the path with `s3://`.
-
-The credentials are pulled in the following order:
-
-- `AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY`, and `AWS_SESSION_TOKEN` environment variables
-- from the `~/.aws/credentials` file
-- for nodes on EC2, the IAM metadata provider
-
-::: {.callout-note}
-
-We assume you have credentials setup and not using anonymous access. If you want to use anonymous access, let us know! We may have to open a config option for this.
-
-:::
-
-Other environment variables that can be set can be found in [boto3 docs](https://boto3.amazonaws.com/v1/documentation/api/latest/guide/configuration.html#using-environment-variables)
-
-#### GCS
-
-Prepend the path with `gs://` or `gcs://`.
-
-The credentials are loaded in the following order:
-
-- gcloud credentials
-- for nodes on GCP, the google metadata service
-- anonymous access
-
-#### Azure
-
-##### Gen 1
-
-Prepend the path with `adl://`.
-
-Ensure you have the following environment variables set:
-
-- `AZURE_STORAGE_TENANT_ID`
-- `AZURE_STORAGE_CLIENT_ID`
-- `AZURE_STORAGE_CLIENT_SECRET`
-
-##### Gen 2
-
-Prepend the path with `abfs://` or `az://`.
-
-Ensure you have the following environment variables set:
-
-- `AZURE_STORAGE_ACCOUNT_NAME`
-- `AZURE_STORAGE_ACCOUNT_KEY`
-
-Other environment variables that can be set can be found in [adlfs docs](https://github.com/fsspec/adlfs?tab=readme-ov-file#setting-credentials)
-
-#### OCI
-
-Prepend the path with `oci://`.
-
-It would attempt to read in the following order:
-
-- `OCIFS_IAM_TYPE`, `OCIFS_CONFIG_LOCATION`, and `OCIFS_CONFIG_PROFILE` environment variables
-- when on OCI resource, resource principal
-
-Other environment variables:
-
-- `OCI_REGION_METADATA`
-
-Please see the [ocifs docs](https://ocifs.readthedocs.io/en/latest/getting-connected.html#Using-Environment-Variables).
-
-### HTTPS
-
-The path should start with `https://`.
-
-```yaml
-datasets:
-  - path: https://path/to/your/dataset/file.jsonl
-```
-
-This must be publically accessible.
-
-## Next steps
-
-Now that you know how to load datasets, you can learn more on how to load your specific dataset format into your target output format [dataset formats docs](dataset-formats).

docs/lora_optims.qmd

@@ -17,7 +17,6 @@ We currently support several common model architectures, including (but not limi
 - `qwen2`
 - `gemma`
 - `gemma2`
-- `gemma3`
 
 <details>
 

docs/multimodal.qmd

@@ -9,7 +9,6 @@ format:
 ## Supported Models
 
 - [Mllama](#sec-mllama)
-- [Llama4](#sec-llama4)
 - [Pixtral](#sec-pixtral)
 - [Llava-1.5](#sec-llava-15)
 - [Mistral-Small-3.1](#sec-mistral-small-31)
@@ -64,14 +63,6 @@ base_model: meta-llama/Llama-3.2-11B-Vision-Instruct
 chat_template: llama3_2_vision
 ```
 
-### Llama4 {#sec-llama4}
-
-```yaml
-base_model: meta-llama/Llama-4-Scout-17B-16E-Instruct
-
-chat_template: llama4
-```
-
 ### Pixtral {#sec-pixtral}
 
 ```yaml

docs/sequence_parallelism.qmd

@@ -23,9 +23,10 @@ Use sequence parallelism when:
 To enable sequence parallelism, add the following to your configuration file:
 
 ```yaml
-# Set to a divisor (> 1) of the number of GPUs available
-sequence_parallel_degree: 4  # Split sequences across 4 GPUs
-# Optional; strides across the key dimension. Larger values use more memory but should make training faster.
+sequence_parallel_degree: 4  # Split each sequence into 4 parts, one per GPU
+flash_attention: true  # SP requires flash attention
+micro_batch_size: 1  # SP requires this is set to 1
+# (optional) strides across the key dimension; larger values use more memory but should make training a bit faster
 heads_k_stride: 1
 ```
 
@@ -66,15 +67,16 @@ sequence_len: 8192
 ...
 
 sequence_parallel_degree: 4  # Split each sequence into 4 parts, one per GPU
-flash_attention: true  # Required with sequence parallelism
-# Optional; strides across the key dimension. Larger values use more memory but should make training faster.
+flash_attention: true  # SP requires flash attention
+micro_batch_size: 1  # SP requires this is set to 1
+# (optional) strides across the key dimension; larger values use more memory but should make training a bit faster
 heads_k_stride: 1
 
 ...
 ```
 
-This will train the Llama 3 8B model with 8K context length, with each sequence split
-into 2 subsequences of length 4096 across 2 GPUs.
+This will train the Llama 3 8B model with 8192 context length, with each sequence split
+into 4 subsequences of length 2048 across 4 GPUs.
 
 ## Sample Packing with Sequence Parallelism
 
@@ -86,12 +88,14 @@ Sequence parallelism is compatible with Axolotl's sample packing functionality.
 
 ## Effect on Batch Size
 
+First, note that sequence parallelism supports only the case where `micro_batch_size: 1`.
+
 When using sequence parallelism, your effective global batch size is **divided** by the `sequence_parallel_degree`. This happens because:
 
 - Each group of `sequence_parallel_degree` GPUs works on the same batch (just different parts of each sequence)
 - The number of batches processed per step decreases
 
 For example:
-- With 8 GPUs and no sequence parallelism: 8 different batches processed per step
+- With 8 GPUs and no sequence parallelism: 8 different batches are processed per step
 - With 8 GPUs and `sequence_parallel_degree=4`: Only 2 different batches processed per step (each split across 4 GPUs)
-- If your per-GPU `micro_batch_size` is 2, the global batch size decreases from 16 to 4
+- If your per-GPU `micro_batch_size` is 1, the global batch size decreases from 8 to 2

examples/cerebras/btlm-ft.yml

@@ -8,6 +8,9 @@ tokenizer_type: GPT2Tokenizer
 trust_remote_code: true
 tokenizer_use_fast: true
 tokenizer_legacy: true
+
+load_in_8bit: false
+load_in_4bit: false
 strict: false
 push_dataset_to_hub:
 hf_use_auth_token: true
@@ -31,6 +34,7 @@ lora_alpha:
 lora_dropout:
 lora_target_modules:
 lora_target_linear:
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -54,12 +58,16 @@ learning_rate: 0.000085
 train_on_inputs: true
 group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 
 gradient_checkpointing: false
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
 
+xformers_attention:
 flash_attention: true
 sdp_attention:
 flash_optimum:
@@ -72,6 +80,8 @@ evals_per_epoch: 4
 saves_per_epoch: 1
 save_total_limit:
 
+debug:
+deepspeed:
 weight_decay: 0.1
 special_tokens:
   pad_token: "<|endoftext|>"

examples/cerebras/qlora.yml

@@ -22,6 +22,7 @@ lora_target_modules:
   - c_attn
   - c_proj
 lora_target_linear:
+lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
@@ -35,10 +36,15 @@ optimizer: paged_adamw_8bit
 torchdistx_path:
 lr_scheduler: cosine
 learning_rate: 0.0002
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 gradient_checkpointing: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
 xformers_attention: true
 flash_attention:
@@ -47,6 +53,10 @@ gptq_model_v1:
 warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.1
+fsdp:
+fsdp_config:
 special_tokens:
   pad_token: "<|endoftext|>"

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

@@ -26,6 +26,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -40,18 +41,29 @@ optimizer: adamw_bnb_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
+s2_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   bos_token: "<s>"
   eos_token: "</s>"

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

@@ -26,7 +26,9 @@ pad_to_sequence_len: true
 lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
+lora_target_modules:
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -41,18 +43,28 @@ optimizer: paged_adamw_32bit
 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
 
 warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   bos_token: "<s>"
   eos_token: "</s>"

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

@@ -26,6 +26,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -40,18 +41,29 @@ optimizer: adamw_bnb_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
+s2_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   bos_token: "<s>"
   eos_token: "</s>"

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

@@ -26,7 +26,9 @@ pad_to_sequence_len: true
 lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
+lora_target_modules:
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -41,18 +43,28 @@ optimizer: paged_adamw_32bit
 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
 
 warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   bos_token: "<s>"
   eos_token: "</s>"

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

@@ -26,6 +26,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -40,18 +41,29 @@ optimizer: adamw_bnb_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
+s2_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   bos_token: "<s>"
   eos_token: "</s>"

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

@@ -26,7 +26,9 @@ pad_to_sequence_len: true
 lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
+lora_target_modules:
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -41,18 +43,28 @@ optimizer: paged_adamw_32bit
 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
 
 warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   bos_token: "<s>"
   eos_token: "</s>"

examples/cohere/command-r-7b-qlora.yml

@@ -44,16 +44,28 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 
 gradient_checkpointing: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_ratio: 0.1
 evals_per_epoch:
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/dbrx/16bit-lora.yaml

@@ -3,6 +3,9 @@ base_model: LnL-AI/dbrx-base-converted-v2
 # hub_model_id: username/custom_model_name
 
 trust_remote_code: true
+
+load_in_8bit: false
+load_in_4bit: false
 strict: false
 
 datasets:
@@ -45,20 +48,26 @@ optimizer: paged_adamw_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: false
 
 gradient_checkpointing: false  # don't use with fsdp_activation_checkpointing
 gradient_checkpointing_kwargs:
   use_reentrant: false
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
 evals_per_epoch:
 saves_per_epoch: 1
-
+debug:
 weight_decay: 0.0
 fsdp:
   - full_shard

examples/dbrx/8bit-lora.yaml

@@ -48,20 +48,26 @@ optimizer: paged_adamw_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: false
 
 gradient_checkpointing: false  # don't use with fsdp_activation_checkpointing
 gradient_checkpointing_kwargs:
   use_reentrant: false
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
 evals_per_epoch:
 saves_per_epoch: 1
-
+debug:
 weight_decay: 0.0
 fsdp:
   - full_shard

examples/dbrx/fft-ds-zero3.yaml

@@ -3,6 +3,9 @@ base_model: LnL-AI/dbrx-base-converted-v2
 # hub_model_id: username/custom_model_name
 
 trust_remote_code: true
+
+load_in_8bit: false
+load_in_4bit: false
 strict: false
 
 datasets:
@@ -32,19 +35,25 @@ optimizer: paged_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
 gradient_checkpointing_kwargs:
   use_reentrant: false
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
 evals_per_epoch:
 saves_per_epoch: 1
-
+debug:
 weight_decay: 0.0
 deepspeed: deepspeed_configs/zero3_bf16.json

examples/deepseek-v2/fft-fsdp-16b.yaml

@@ -2,6 +2,9 @@ base_model: deepseek-ai/DeepSeek-V2-Lite
 # Automatically upload checkpoint and final model to HF
 # hub_model_id: username/custom_model_name
 trust_remote_code: true
+
+load_in_8bit: false
+load_in_4bit: false
 strict: false
 
 datasets:
@@ -28,19 +31,27 @@ optimizer: adamw_torch_fused
 lr_scheduler: cosine
 learning_rate: 2e-5
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: false
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: false
+early_stopping_patience:
 resume_from_checkpoint:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 100
 evals_per_epoch: 2
+eval_table_size:
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
 special_tokens:
 fsdp:

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

@@ -52,19 +52,27 @@ optimizer: adamw_torch_fused
 lr_scheduler: cosine
 learning_rate: 2e-5
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: false
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: false
+early_stopping_patience:
 resume_from_checkpoint:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 100
 evals_per_epoch: 2
+eval_table_size:
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
 special_tokens:
 fsdp:

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

@@ -25,7 +25,9 @@ max_packed_sequence_len:
 lora_r: 16
 lora_alpha: 32
 lora_dropout: 0.0
+lora_target_modules:
 lora_target_linear: true
+lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
@@ -39,10 +41,15 @@ optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
 learning_rate: 0.00003
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 gradient_checkpointing: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
 xformers_attention: true
 flash_attention:
@@ -51,7 +58,11 @@ gptq_model_v1:
 warmup_steps: 40
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   pad_token: "<|endoftext|>"
   bos_token: "<|endoftext|>"

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

@@ -38,7 +38,9 @@ lora_alpha: 16
 # 0.05 for 33B and 65B models
 lora_dropout: 0.05
 # add LoRA modules on all linear layers of the base model
+lora_target_modules:
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -65,7 +67,10 @@ lr_scheduler: cosine
 # - 2e-4 for 7b & 13b
 # - 1e-4 for 33b & 64b
 learning_rate: 0.0002
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 gradient_checkpointing: true
 # stop training after this many evaluation losses have increased in a row
@@ -73,6 +78,7 @@ gradient_checkpointing: true
 early_stopping_patience: 3
 resume_from_checkpoint:
 auto_resume_from_checkpoints: true
+local_rank:
 logging_steps: 1
 xformers_attention: true
 flash_attention:
@@ -81,7 +87,11 @@ gptq_model_v1:
 warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.000001
+fsdp:
+fsdp_config:
 special_tokens:
   pad_token: "<|endoftext|>"
   bos_token: "<|endoftext|>"

examples/falcon/config-7b.yml

@@ -7,6 +7,9 @@ tokenizer_type: AutoTokenizer
 
 # required by falcon custom model code: https://huggingface.co/tiiuae/falcon-7b/tree/main
 trust_remote_code: true
+
+load_in_8bit: false
+load_in_4bit: false
 gptq: false
 strict: false
 push_dataset_to_hub:
@@ -22,7 +25,9 @@ max_packed_sequence_len:
 lora_r: 64
 lora_alpha: 32
 lora_dropout: 0.0
+lora_target_modules:
 lora_target_linear: true
+lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
@@ -36,10 +41,15 @@ optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
 learning_rate: 0.00003
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 gradient_checkpointing: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
 xformers_attention: true
 flash_attention:
@@ -48,7 +58,11 @@ gptq_model_v1:
 warmup_steps: 40
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   pad_token: "<|endoftext|>"
   bos_token: "<|endoftext|>"

examples/gemma/qlora.yml

@@ -42,16 +42,28 @@ optimizer: adamw_bnb_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
 
 warmup_ratio: 0.1
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/gemma2/qlora.yml

@@ -48,16 +48,28 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 
 gradient_checkpointing: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_ratio: 0.1
 evals_per_epoch:
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/gemma2/reward-model.yaml

@@ -5,6 +5,9 @@ num_labels: 1
 tokenizer_type: AutoTokenizer
 # 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
 
 reward_model: true
@@ -35,6 +38,8 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: true
 fp16:
 tf32: true
@@ -42,12 +47,21 @@ tf32: true
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: false
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_ratio: 0.1
 evals_per_epoch:
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/gemma3/gemma-3-1b-qlora.yml

@@ -50,18 +50,30 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: false
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_ratio: 0.1
 evals_per_epoch:
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/gemma3/gemma-3-4b-lora.yml

@@ -2,8 +2,6 @@ base_model: google/gemma-3-4b-it
 processor_type: AutoProcessor
 strict: false
 
-load_in_4bit: true
-
 # these 3 lines are needed for now to handle vision chat templates w images
 skip_prepare_dataset: true
 remove_unused_columns: false
@@ -22,7 +20,7 @@ dataset_prepared_path: last_run_prepared
 val_set_size: 0.01
 output_dir: ./outputs/out
 
-adapter: qlora
+adapter: lora
 lora_model_dir:
 
 sequence_len: 2048
@@ -46,6 +44,8 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: true
 fp16:
 tf32: true
@@ -53,6 +53,7 @@ tf32: true
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: false
+local_rank:
 logging_steps: 1
 flash_attention: true
 eager_attention:
@@ -60,4 +61,8 @@ eager_attention:
 warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:

examples/gemma3/gemma-3-4b-qlora.yml

@@ -1,61 +0,0 @@
-base_model: google/gemma-3-4b-it
-strict: false
-
-load_in_4bit: true
-
-# gemma3 doesn't seem to play nice with ddp
-ddp_find_unused_parameters: true
-
-chat_template: gemma3
-datasets:
-  - path: cgato/SlimOrcaDedupCleaned
-    type: chat_template
-    field_messages: conversations
-    message_property_mappings:
-      role: from
-      content: value
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
-output_dir: ./outputs/out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 2048
-sample_packing: true
-pad_to_sequence_len: true
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules: 'language_model.model.layers.[\d]+.(mlp|cross_attn|self_attn).(up|down|gate|q|k|v|o)_proj'
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 1
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: true
-fp16:
-tf32: true
-
-gradient_checkpointing: true
-gradient_checkpointing_kwargs:
-  use_reentrant: false
-logging_steps: 1
-flash_attention: true
-eager_attention:
-
-warmup_ratio: 0.1
-evals_per_epoch: 1
-saves_per_epoch: 1
-weight_decay: 0.0

examples/gptj/qlora.yml

@@ -18,7 +18,9 @@ max_packed_sequence_len:
 lora_r: 8
 lora_alpha: 32
 lora_dropout: 0.05
+lora_target_modules:
 lora_target_linear: true
+lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
@@ -32,10 +34,15 @@ optimizer: paged_adamw_8bit
 torchdistx_path:
 lr_scheduler: cosine
 learning_rate: 0.0001
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 gradient_checkpointing: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
 xformers_attention: true
 flash_attention:
@@ -44,6 +51,10 @@ gptq_model_v1:
 warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.1
+fsdp:
+fsdp_config:
 special_tokens:
   pad_token: "<|endoftext|>"

examples/jamba/qlora.yaml

@@ -40,18 +40,26 @@ optimizer: paged_adamw_8bit
 lr_scheduler: cosine
 learning_rate: 0.00001
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: false
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: false
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
 evals_per_epoch:
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
 special_tokens:

examples/jamba/qlora_deepspeed.yaml

@@ -39,20 +39,26 @@ optimizer: paged_adamw_8bit
 lr_scheduler: cosine
 learning_rate: 0.00001
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: false
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: false
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
 evals_per_epoch:
 saves_per_epoch: 1
-
+debug:
 deepspeed: deepspeed_configs/zero2.json
 weight_decay: 0.0
 special_tokens:

examples/jamba/qlora_fsdp_large.yaml

@@ -39,6 +39,8 @@ optimizer: adamw_torch_fused
 lr_scheduler: cosine
 learning_rate: 0.00001
 
+train_on_inputs: false
+group_by_length: false
 bf16: true
 tf32: true
 

examples/jeopardy-bot/config.yml

@@ -33,9 +33,13 @@ optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
 learning_rate: 0.00003
+train_on_inputs: false
+group_by_length: false
 bf16: auto
 tf32: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 5
 xformers_attention: true
 flash_attention:
@@ -44,7 +48,11 @@ gptq_model_v1:
 warmup_steps: 20
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.1
+fsdp:
+fsdp_config:
 tokens:
   bos_token: "<s>"
   eos_token: "</s>"

examples/llama-2/fft_optimized.yml

@@ -4,6 +4,9 @@ model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 # 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:
@@ -23,6 +26,7 @@ lora_r:
 lora_alpha:
 lora_dropout:
 lora_target_linear:
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -37,12 +41,18 @@ optimizer: adamw_bnb_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
 flash_attn_cross_entropy: false
 flash_attn_rms_norm: true
@@ -51,8 +61,11 @@ flash_attn_fuse_mlp: true
 
 warmup_steps: 100
 evals_per_epoch: 4
+eval_table_size:
 saves_per_epoch: 1
-
+debug:
 deepspeed: #deepspeed_configs/zero2.json # multi-gpu only
 weight_decay: 0.1
+fsdp:
+fsdp_config:
 special_tokens:

examples/llama-2/gptq-lora.yml

@@ -10,6 +10,8 @@ gptq_disable_exllama: true
 
 tokenizer_use_fast: true
 tokenizer_legacy: true
+load_in_8bit: false
+load_in_4bit: false
 strict: false
 push_dataset_to_hub:
 hf_use_auth_token: true
@@ -31,6 +33,7 @@ lora_target_modules:
   - q_proj
   - v_proj
 lora_target_linear:
+lora_fan_in_fan_out:
 wandb_project:
 wandb_watch:
 wandb_name:
@@ -47,19 +50,26 @@ torchdistx_path:
 lr_scheduler: cosine
 lr_quadratic_warmup: true
 learning_rate: 0.000017
+train_on_inputs: false
+group_by_length: false
 bf16: false
 fp16: false
 float16: true
 tf32: true
 gradient_checkpointing: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention:
 sdp_attention:
 flash_optimum:
 warmup_steps: 100
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.1
 special_tokens:
   bos_token: "<s>"

examples/llama-2/lisa.yml

@@ -4,6 +4,9 @@ model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 # 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:
@@ -23,6 +26,7 @@ lora_r:
 lora_alpha:
 lora_dropout:
 lora_target_linear:
+lora_fan_in_fan_out:
 
 lisa_n_layers: 4
 lisa_step_interval: 20
@@ -41,12 +45,18 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 5e-5 # recommendation from lisa paper for 7b
 
+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
 flash_attn_cross_entropy: false
 flash_attn_rms_norm: true
@@ -55,8 +65,13 @@ flash_attn_fuse_mlp: true
 
 warmup_steps: 100
 evals_per_epoch: 4
+eval_table_size:
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.1
+fsdp:
+fsdp_config:
 special_tokens:
   bos_token: "<s>"
   eos_token: "</s>"

examples/llama-2/loftq.yml

@@ -4,6 +4,9 @@ model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 # 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:
@@ -23,6 +26,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 peft:
   loftq_config:
     loftq_bits: 4
@@ -40,16 +44,29 @@ optimizer: adamw_bnb_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
+s2_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/llama-2/lora.yml

@@ -26,6 +26,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -40,16 +41,29 @@ optimizer: adamw_bnb_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
+s2_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/llama-2/qlora-fsdp.yml

@@ -26,7 +26,9 @@ pad_to_sequence_len: true
 lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
+lora_target_modules:
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -41,19 +43,28 @@ optimizer: adamw_torch_fused
 lr_scheduler: cosine
 learning_rate: 0.00001
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: false
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
 fsdp:
   - full_shard

examples/llama-2/qlora.yml

@@ -26,7 +26,9 @@ pad_to_sequence_len: true
 lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
+lora_target_modules:
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -41,16 +43,27 @@ optimizer: paged_adamw_32bit
 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
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/llama-2/relora.yml

@@ -24,7 +24,9 @@ pad_to_sequence_len: true
 lora_r: 8
 lora_alpha: 16
 lora_dropout: 0.05
+lora_target_modules:
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 relora_steps: 150
 relora_warmup_steps: 10
@@ -43,18 +45,28 @@ optimizer: adamw_bnb_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
 
 warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   bos_token: "<s>"
   eos_token: "</s>"

examples/llama-3-vision/lora-11b.yaml

@@ -45,11 +45,14 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: true
 fp16:
 tf32: true
 
 gradient_checkpointing: true
+local_rank:
 logging_steps: 1
 flash_attention: true
 eager_attention:
@@ -57,4 +60,8 @@ eager_attention:
 warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:

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

@@ -42,19 +42,27 @@ optimizer: adamw_torch_fused
 lr_scheduler: cosine
 learning_rate: 2e-5
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: false
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: false
+early_stopping_patience:
 resume_from_checkpoint:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 100
 evals_per_epoch: 2
+eval_table_size:
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
 fsdp:
   - full_shard

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

@@ -1,6 +1,9 @@
 base_model: NousResearch/Meta-Llama-3.1-8B
 # 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:
@@ -27,19 +30,29 @@ optimizer: paged_adamw_8bit
 lr_scheduler: cosine
 learning_rate: 2e-5
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: false
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: false
+early_stopping_patience:
 resume_from_checkpoint:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 100
 evals_per_epoch: 2
+eval_table_size:
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   pad_token: <|end_of_text|>

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

@@ -42,6 +42,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -56,15 +57,31 @@ optimizer: adamw_bnb_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
+s2_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
+
+special_tokens:
+  pad_token: "<|end_of_text|>"

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

@@ -37,6 +37,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -51,17 +52,30 @@ optimizer: adamw_bnb_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
+s2_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
    pad_token: <|end_of_text|>

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

@@ -58,6 +58,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -72,15 +73,28 @@ optimizer: adamw_bnb_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
+s2_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:

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

@@ -31,6 +31,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 lora_modules_to_save:
   - embed_tokens
   - lm_head
@@ -48,17 +49,30 @@ optimizer: adamw_bnb_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
+s2_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
    pad_token: <|end_of_text|>

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

@@ -1,6 +1,9 @@
 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:
@@ -21,6 +24,7 @@ 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
@@ -49,12 +53,18 @@ 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
@@ -63,6 +73,10 @@ 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/llama-3/lora-1b-ray.yml

@@ -1,6 +1,9 @@
 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:
@@ -21,6 +24,7 @@ pad_to_sequence_len: true
 lora_r: 16
 lora_alpha: 32
 lora_dropout: 0.05
+lora_fan_in_fan_out:
 lora_target_modules:
   - gate_proj
   - down_proj
@@ -43,12 +47,18 @@ 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
@@ -57,9 +67,11 @@ loss_watchdog_patience: 3
 warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
-
+debug:
 deepspeed: deepspeed_configs/zero3.json
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   pad_token: "<|end_of_text|>"
 

examples/llama-3/lora-1b-sample-packing-sequentially.yml

@@ -33,6 +33,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 lora_modules_to_save:
   - embed_tokens
   - lm_head
@@ -50,17 +51,30 @@ optimizer: adamw_bnb_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
+s2_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   pad_token: <|end_of_text|>

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

@@ -1,6 +1,9 @@
 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:
@@ -21,6 +24,7 @@ pad_to_sequence_len: true
 lora_r: 16
 lora_alpha: 32
 lora_dropout: 0.05
+lora_fan_in_fan_out:
 lora_target_modules:
   - gate_proj
   - down_proj
@@ -43,12 +47,18 @@ 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
@@ -57,6 +67,10 @@ 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/llama-3/lora-8b.yml

@@ -27,6 +27,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 lora_modules_to_save:
   - embed_tokens
   - lm_head
@@ -44,17 +45,30 @@ optimizer: adamw_bnb_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
+s2_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
    pad_token: <|end_of_text|>

examples/llama-3/qlora-1b-kto.yaml

@@ -32,6 +32,7 @@ lora_r: 32
 lora_alpha: 64
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -46,19 +47,31 @@ optimizer: adamw_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: false
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 20
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   pad_token: "<|end_of_text|>"

examples/llama-3/qlora-1b.yml

@@ -24,6 +24,7 @@ pad_to_sequence_len: true
 lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
+lora_fan_in_fan_out:
 lora_target_modules:
   - gate_proj
   - down_proj
@@ -46,12 +47,18 @@ optimizer: adamw_bnb_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
@@ -59,7 +66,13 @@ loss_watchdog_patience: 3
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   pad_token: "<|end_of_text|>"

examples/llama-3/qlora-fsdp-405b.yaml

@@ -24,6 +24,7 @@ pad_to_sequence_len: true
 lora_r: 16
 lora_alpha: 16
 lora_dropout: 0.05
+lora_target_modules:
 lora_target_linear: true
 
 gradient_accumulation_steps: 4
@@ -33,6 +34,8 @@ optimizer: adamw_torch_fused
 lr_scheduler: cosine
 learning_rate: 0.00001
 
+train_on_inputs: false
+group_by_length: false
 bf16: true
 tf32: true
 

examples/llama-3/qlora-fsdp-70b.yaml

@@ -26,7 +26,9 @@ pad_to_sequence_len: true
 lora_r: 8
 lora_alpha: 16
 lora_dropout: 0.05
+lora_target_modules:
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -41,19 +43,28 @@ optimizer: adamw_torch_fused
 lr_scheduler: cosine
 learning_rate: 0.00001
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: false
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
 fsdp:
   - full_shard

examples/llama-3/qlora.yml

@@ -26,7 +26,9 @@ pad_to_sequence_len: true
 lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
+lora_target_modules:
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -41,17 +43,28 @@ optimizer: paged_adamw_32bit
 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
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   pad_token: "<|end_of_text|>"

examples/llama-4/scout-lora.yaml

@@ -1,75 +0,0 @@
-base_model: meta-llama/Llama-4-Scout-17B-16E
-model_type: Llama4ForConditionalGeneration
-  # Automatically upload checkpoint and final model to HF
-  # hub_model_id: username/custom_model_name
-
-strict: false
-
-  # torch_compile: true
-
-adapter: lora
-lora_r: 32
-lora_alpha: 64
-lora_target_modules:
-  - self_attn.q_proj
-  - self_attn.k_proj
-  - self_attn.v_proj
-  - self_attn.o_proj
-lora_modules_to_save:
-  - lm_head
-  - embed_tokens
-
-chat_template: llama4
-datasets:
-  - path: mlabonne/FineTome-100k
-    type: chat_template
-    split: train[:20%]
-    field_messages: conversations
-    message_property_mappings:
-      role: from
-      content: value
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.0
-output_dir: ./outputs/out
-
-sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
-
-gradient_accumulation_steps: 1
-micro_batch_size: 1
-num_epochs: 1
-optimizer: adamw_torch_8bit
-lr_scheduler: cosine
-learning_rate: 2e-5
-
-bf16: true
-tf32: true
-
-# gradient_checkpointing: true
-# gradient_checkpointing_kwargs:
-#   use_reentrant: false
-logging_steps: 1
-flash_attention: true
-
-warmup_steps: 100
-evals_per_epoch: 2
-saves_per_epoch: 1
-weight_decay: 0.0
-fsdp:
-  - auto_wrap
-  - full_shard
-fsdp_config:
-  fsdp_version: 2
-  fsdp_offload_params: false
-  fsdp_cpu_ram_efficient_loading: true
-  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
-  fsdp_transformer_layer_cls_to_wrap: Llama4TextDecoderLayer
-  fsdp_state_dict_type: SHARDED_STATE_DICT
-  fsdp_sharding_strategy: FULL_SHARD
-  fsdp_reshard_after_forward: true
-  fsdp_activation_checkpointing: true
-special_tokens:
-  pad_token: <|finetune_right_pad_id|>
-  eos_token: <|eot|>

examples/llama-4/scout-qlora-fsdp1.yaml

@@ -1,93 +0,0 @@
-base_model: axolotl-quants/Llama-4-Scout-17B-16E-Linearized-bnb-nf4-bf16
-model_type: Llama4ForConditionalGeneration
-  # Automatically upload checkpoint and final model to HF
-  # hub_model_id: username/custom_model_name
-
-strict: false
-
-# torch_compile: true
-plugins:
-  - axolotl.integrations.liger.LigerPlugin
-
-liger_glu_activation: true
-liger_rms_norm: true
-liger_layer_norm: true
-
-llama4_linearized_experts: true
-load_in_4bit: true
-adapter: qlora
-lora_r: 32
-lora_alpha: 64
-lora_target_modules:
-  - self_attn.q_proj
-  - self_attn.k_proj
-  - self_attn.v_proj
-  - self_attn.o_proj
-  - shared_expert.gate_proj
-  - shared_expert.up_proj
-  - shared_expert.down_proj
-    # - experts.gate_projs.[0-9]+$
-    # - experts.up_projs.[0-9]+$
-    # - experts.down_projs.[0-9]+$
-lora_modules_to_save:
-  - lm_head
-  - embed_tokens
-
-chat_template: llama4
-datasets:
-  - path: mlabonne/FineTome-100k
-    type: chat_template
-    split: train[:20%]
-    field_messages: conversations
-    message_property_mappings:
-      role: from
-      content: value
-
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.0
-output_dir: ./outputs/out
-
-sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 1
-micro_batch_size: 1
-num_epochs: 1
-optimizer: adamw_torch_fused
-lr_scheduler: cosine
-learning_rate: 2e-5
-
-bf16: true
-tf32: true
-
-logging_steps: 1
-flash_attention: true
-
-warmup_steps: 100
-evals_per_epoch: 1
-saves_per_epoch: 1
-weight_decay: 0.0
-fsdp:
-  - auto_wrap
-  - full_shard
-fsdp_config:
-  fsdp_transformer_layer_cls_to_wrap: Llama4TextDecoderLayer
-  fsdp_limit_all_gathers: true
-  fsdp_sync_module_states: true
-  fsdp_offload_params: true
-  fsdp_use_orig_params: false
-  fsdp_cpu_ram_efficient_loading: true
-  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
-  fsdp_state_dict_type: FULL_STATE_DICT
-  fsdp_sharding_strategy: FULL_SHARD
-  fsdp_activation_checkpointing: true
-special_tokens:
-  pad_token: <|finetune_right_pad_id|>
-  eos_token: <|eot|>

examples/llava/lora-7b.yaml

@@ -41,11 +41,14 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: true
 fp16:
 tf32: true
 
 gradient_checkpointing: true
+local_rank:
 logging_steps: 1
 flash_attention: true
 eager_attention:
@@ -53,4 +56,8 @@ eager_attention:
 warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:

examples/mamba/config.yml

@@ -5,6 +5,9 @@ tokenizer_type: AutoTokenizer
 tokenizer_config: EleutherAI/gpt-neox-20b
 # 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:
@@ -35,17 +38,27 @@ train_on_inputs: false
 group_by_length: true
 
 bf16: auto
+fp16:
 tf32: true
 
 gradient_checkpointing: false
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
 tokens:
 save_safetensors: False

examples/mistral/bigstral-ds-zero3.yaml

@@ -6,6 +6,9 @@ tokenizer_type: LlamaTokenizer
 # hub_model_id: username/custom_model_name
 
 trust_remote_code: true
+
+load_in_8bit: false
+load_in_4bit: false
 strict: false
 
 unfrozen_parameters:
@@ -37,19 +40,27 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0001
 
+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
 
 save_total_limit: 1
 save_steps:
-
+debug:
 deepspeed: deepspeed_configs/zero3_bf16_cpuoffload_params.json
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   eos_token: "<|im_end|>"
 tokens:

examples/mistral/config.yml

@@ -4,6 +4,9 @@ model_type: MistralForCausalLM
 tokenizer_type: LlamaTokenizer
 # 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:
@@ -31,16 +34,28 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.000005
 
+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
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/mistral/lora-mps.yml

@@ -4,6 +4,9 @@ model_type: MistralForCausalLM
 tokenizer_type: LlamaTokenizer
 # 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:
@@ -25,6 +28,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 lora_target_modules:
   - gate_proj
   - down_proj
@@ -47,13 +51,18 @@ optimizer: adamw_torch_fused
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
 fp16: false
 tf32: true
 
 gradient_checkpointing: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: false
 sdp_attention: true
 
@@ -62,6 +71,12 @@ loss_watchdog_patience: 3
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_table_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/mistral/lora.yml

@@ -27,6 +27,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 lora_target_modules:
   - gate_proj
   - down_proj
@@ -49,12 +50,18 @@ optimizer: adamw_bnb_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
@@ -62,6 +69,12 @@ loss_watchdog_patience: 3
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/mistral/mistral-dpo-qlora.yml

@@ -40,6 +40,7 @@ lora_r: 8
 lora_alpha: 16
 lora_dropout: 0.2
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 lora_target_modules:
   - gate_proj
@@ -66,18 +67,31 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0001
 
+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: false
+s2_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   bos_token: "<|im_start|>"
   eos_token: "<|im_end|>"

examples/mistral/mistral-qlora-fsdp.yml

@@ -32,6 +32,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -46,12 +47,18 @@ optimizer: paged_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
@@ -59,8 +66,10 @@ loss_watchdog_patience: 3
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
-
+debug:
 weight_decay: 0.0
 fsdp:
   - full_shard

examples/mistral/mistral-qlora-orpo.yml

@@ -32,6 +32,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 lora_target_modules:
   - gate_proj
   - down_proj
@@ -54,12 +55,18 @@ optimizer: adamw_bnb_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
@@ -67,6 +74,12 @@ loss_watchdog_patience: 3
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/mistral/mistral-small-3.1-24B-lora.yml

@@ -43,11 +43,14 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: true
 fp16:
 tf32: true
 
 gradient_checkpointing: true
+local_rank:
 logging_steps: 1
 flash_attention: false # PixtralVisionModel does not support Flash Attention 2.0 yet.
 eager_attention:
@@ -55,5 +58,9 @@ eager_attention:
 warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/mistral/mixtral-8x22b-qlora-fsdp.yml

@@ -30,6 +30,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -44,12 +45,18 @@ optimizer: adamw_torch_fused
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 
 gradient_checkpointing: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 loss_watchdog_threshold: 5.0
@@ -57,8 +64,10 @@ loss_watchdog_patience: 3
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
-
+debug:
 weight_decay: 0.0
 fsdp:
   - full_shard

examples/mistral/mixtral-qlora-fsdp.yml

@@ -32,6 +32,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -46,12 +47,18 @@ optimizer: adamw_torch_fused
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 
 gradient_checkpointing: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 loss_watchdog_threshold: 5.0
@@ -59,8 +66,10 @@ loss_watchdog_patience: 3
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
-
+debug:
 weight_decay: 0.0
 fsdp:
   - full_shard

examples/mistral/mixtral.yml

@@ -41,6 +41,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 #lora_target_modules:
 #  - gate
 #  - q_proj
@@ -64,12 +65,18 @@ optimizer: adamw_bnb_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
@@ -77,8 +84,12 @@ loss_watchdog_patience: 3
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
-
+debug:
 deepspeed: deepspeed_configs/zero2.json
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/mistral/mixtral_22.yml

@@ -6,6 +6,9 @@ tokenizer_type: LlamaTokenizer
 # hub_model_id: username/custom_model_name
 
 trust_remote_code: true
+
+load_in_8bit: false
+load_in_4bit: false
 strict: false
 
 unfrozen_parameters:
@@ -35,19 +38,27 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0001
 
+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
 
 save_total_limit: 1
 save_steps:
-
+debug:
 deepspeed: deepspeed_configs/zero3_bf16_cpuoffload_all.json
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   eos_token: "<|im_end|>"
 tokens:

examples/mistral/qlora.yml

@@ -27,6 +27,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 lora_target_modules:
   - gate_proj
   - down_proj
@@ -49,12 +50,18 @@ optimizer: adamw_bnb_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
@@ -62,6 +69,12 @@ loss_watchdog_patience: 3
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/mpt-7b/config.yml

@@ -35,17 +35,26 @@ optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
 learning_rate: 0.0000002
+train_on_inputs: false
+group_by_length: false
 bf16: auto
 tf32: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 5
+xformers_attention:
 flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0001
+fsdp:
+fsdp_config:
 tokens:
   pad_token: "<|padding|>"
   bos_token: "<|endoftext|>"

examples/openllama-3b/config.yml

@@ -4,6 +4,9 @@ model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 # 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
 push_dataset_to_hub:
 datasets:
@@ -20,6 +23,7 @@ lora_alpha:
 lora_dropout:
 lora_target_modules:
 lora_target_linear:
+lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
@@ -33,20 +37,29 @@ optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
 learning_rate: 0.000003
+train_on_inputs: false
+group_by_length: false
 float16: true
 bf16: false
 fp16: false
 tf32: false
 gradient_checkpointing: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.1
+fsdp:
+fsdp_config:
 special_tokens:
   bos_token: "<s>"
   eos_token: "</s>"

examples/openllama-3b/lora.yml

@@ -29,6 +29,7 @@ lora_target_modules:
   - v_proj
   - k_proj
   - o_proj
+lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
@@ -42,19 +43,29 @@ optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
 learning_rate: 0.0002
+train_on_inputs: false
+group_by_length: false
 bf16: false
 fp16: true
 tf32: false
 gradient_checkpointing: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 gptq_groupsize:
+s2_attention:
 gptq_model_v1:
 warmup_steps: 20
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.1
+fsdp:
+fsdp_config:
 special_tokens:
   bos_token: "<s>"
   eos_token: "</s>"

examples/openllama-3b/qlora.yml

@@ -21,7 +21,9 @@ sample_packing: true
 lora_r: 8
 lora_alpha: 32
 lora_dropout: 0.05
+lora_target_modules:
 lora_target_linear: true
+lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
@@ -35,19 +37,28 @@ optimizer: paged_adamw_32bit
 torchdistx_path:
 lr_scheduler: cosine
 learning_rate: 0.0002
+train_on_inputs: false
+group_by_length: false
 bf16: false
 fp16: true
 tf32: false
 gradient_checkpointing: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.1
+fsdp:
+fsdp_config:
 special_tokens:
   bos_token: "<s>"
   eos_token: "</s>"

examples/phi/lora-3.5.yaml

@@ -37,6 +37,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -51,16 +52,28 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bfloat16: true
 bf16: true
 fp16:
 tf32: false
 
 gradient_checkpointing: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
+s2_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 4
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:

examples/phi/phi-ft.yml

@@ -4,6 +4,9 @@ model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
 # 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:
@@ -24,6 +27,7 @@ lora_r:
 lora_alpha:
 lora_dropout:
 lora_target_linear:
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -41,20 +45,30 @@ max_grad_norm: 1.0
 lr_scheduler: cosine
 learning_rate: 0.000003
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: True
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 100
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.1
+fsdp:
+fsdp_config:
 resize_token_embeddings_to_32x: true
 special_tokens:
   pad_token: "<|endoftext|>"

examples/phi/phi-qlora.yml

@@ -27,6 +27,7 @@ lora_r: 64
 lora_alpha: 32
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -44,20 +45,30 @@ max_grad_norm: 1.0
 lr_scheduler: cosine
 learning_rate: 0.000003
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: True
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 100
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.1
+fsdp:
+fsdp_config:
 resize_token_embeddings_to_32x: true
 special_tokens:
   pad_token: "<|endoftext|>"

examples/phi/phi2-ft.yml

@@ -4,6 +4,9 @@ model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
 # 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:
@@ -24,6 +27,7 @@ lora_r:
 lora_alpha:
 lora_dropout:
 lora_target_linear:
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -41,20 +45,30 @@ max_grad_norm: 1.0
 lr_scheduler: cosine
 learning_rate: 0.000003
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: True
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 100
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.1
+fsdp:
+fsdp_config:
 resize_token_embeddings_to_32x: true
 special_tokens:
   pad_token: "<|endoftext|>"

examples/phi/phi3-ft-fsdp.yml

@@ -4,6 +4,9 @@ model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
 # 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:
@@ -25,6 +28,7 @@ lora_r:
 lora_alpha:
 lora_dropout:
 lora_target_linear:
+lora_fan_in_fan_out:
 
 wandb_project: phi3
 wandb_entity:
@@ -42,19 +46,27 @@ max_grad_norm: 1.0
 lr_scheduler: cosine
 learning_rate: 0.000003
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 100
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.1
 fsdp:
   - full_shard

examples/phi/phi3-ft.yml

@@ -7,6 +7,9 @@ tokenizer_type: AutoTokenizer
 # hub_model_id: username/custom_model_name
 
 chat_template: phi_3
+
+load_in_8bit: false
+load_in_4bit: false
 strict: false
 
 datasets:
@@ -27,6 +30,7 @@ lora_r: 64
 lora_alpha: 32
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 gradient_accumulation_steps: 1
 micro_batch_size: 2
@@ -38,6 +42,8 @@ max_grad_norm: 1.0
 lr_scheduler: cosine
 learning_rate: 5.0e-6
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
 
 gradient_checkpointing: true
@@ -49,9 +55,9 @@ flash_attention: true
 
 eval_steps: 1000
 save_steps: 5000
+eval_table_size: 2
 eval_batch_size: 2
 eval_sample_packing: false
-eval_table_size: 2
 eval_max_new_tokens: 32
 eval_causal_lm_metrics: ["perplexity"]
 do_causal_lm_eval: true

examples/pixtral/lora-12b.yml

@@ -41,11 +41,14 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: true
 fp16:
 tf32: true
 
 gradient_checkpointing: true
+local_rank:
 logging_steps: 1
 flash_attention: false # PixtralVisionModel does not support Flash Attention 2.0 yet
 eager_attention:
@@ -53,6 +56,10 @@ eager_attention:
 warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:
   pad_token: <pad>

examples/pythia-12b/config.yml

@@ -5,6 +5,9 @@ model_type: GPTNeoXForCausalLM
 tokenizer_type: AutoTokenizer
 # Automatically upload checkpoint and final model to HF
 # hub_model_id: username/custom_model_name
+
+load_in_8bit: false
+load_in_4bit: false
 gptq: false
 device_map: auto
 datasets:
@@ -19,6 +22,7 @@ max_packed_sequence_len: 2048
 lora_r: 64
 lora_alpha: 32
 lora_dropout: 0.0
+lora_target_modules:
 lora_target_linear: true
 lora_fan_in_fan_out: true  # pythia/GPTNeoX lora specific
 wandb_project:
@@ -33,10 +37,16 @@ num_epochs: 5
 learning_rate: 0.00003
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
+train_on_inputs: false
+group_by_length: false
 bf16: false
 fp16: false
 float16: true
 tf32: true
 flash_optimum: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 gradient_checkpointing: true
+fsdp:
+fsdp_config:

examples/pythia/lora.yml

@@ -28,9 +28,13 @@ gradient_accumulation_steps: 1
 micro_batch_size: 4
 num_epochs: 4
 learning_rate: 0.00001
+train_on_inputs: false
+group_by_length: false
 bf16: auto
 tf32: true
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 weight_decay: 0.1
 evals_per_epoch: 4
 logging_steps: 1

examples/qwen/lora.yml

@@ -28,6 +28,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -42,16 +43,28 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: false
 
 gradient_checkpointing: false
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/qwen/qlora.yml

@@ -28,6 +28,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -42,16 +43,28 @@ optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: false
 
 gradient_checkpointing: false
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention:
 
 warmup_steps: 10
 evals_per_epoch: 4
+eval_table_size:
+eval_max_new_tokens: 128
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens:

examples/qwen/qwen2-moe-lora.yaml

@@ -3,6 +3,9 @@ base_model: Qwen/Qwen1.5-MoE-A2.7B
 # hub_model_id: username/custom_model_name
 
 trust_remote_code: true
+
+load_in_8bit: false
+load_in_4bit: false
 strict: false
 
 datasets:
@@ -22,6 +25,7 @@ lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
 lora_target_linear: true
+lora_fan_in_fan_out:
 
 wandb_project:
 wandb_entity:
@@ -36,18 +40,28 @@ optimizer: paged_adamw_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
 
+train_on_inputs: false
+group_by_length: false
 bf16: auto
+fp16:
 tf32: true
 
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: false
+early_stopping_patience:
 resume_from_checkpoint:
+local_rank:
 logging_steps: 1
+xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
 evals_per_epoch: 4
 saves_per_epoch: 1
+debug:
+deepspeed:
 weight_decay: 0.0
+fsdp:
+fsdp_config:
 special_tokens: