Fix(debug): Use space delimiter for debug_text_only also

* Update tests-docker.yml

* Update tests-docker.yml

* run ci tests on ci yaml updates

---------

Co-authored-by: Wing Lian <wing.lian@gmail.com>

.github/ISSUE_TEMPLATE/bug-report.yaml

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

.github/workflows/base.yml

@@ -25,6 +25,16 @@ jobs:
             python_version: "3.10"
             pytorch: 2.0.1
             torch_cuda_arch_list: "7.0 7.5 8.0 8.6 9.0+PTX"
+          - cuda: "118"
+            cuda_version: 11.8.0
+            python_version: "3.10"
+            pytorch: 2.1.1
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 9.0+PTX"
+          - cuda: "121"
+            cuda_version: 12.1.0
+            python_version: "3.10"
+            pytorch: 2.1.1
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 9.0+PTX"
     steps:
       - name: Checkout
         uses: actions/checkout@v3

.github/workflows/main.yml

@@ -13,47 +13,70 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: cu118
+          - cuda: 118
             cuda_version: 11.8.0
             python_version: "3.9"
             pytorch: 2.0.1
             axolotl_extras:
-          - cuda: cu118
+          - cuda: 118
             cuda_version: 11.8.0
             python_version: "3.10"
             pytorch: 2.0.1
             axolotl_extras:
-          - cuda: cu118
+            is_latest: true
+          - cuda: 118
             cuda_version: 11.8.0
-            python_version: "3.9"
-            pytorch: 2.0.1
-            axolotl_extras: gptq
-    runs-on: self-hosted
+            python_version: "3.10"
+            pytorch: 2.1.1
+            axolotl_extras:
+          - cuda: 121
+            cuda_version: 12.1.0
+            python_version: "3.10"
+            pytorch: 2.1.1
+            axolotl_extras:
+    runs-on: [self-hosted, gpu, docker]
     steps:
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
       - name: Docker metadata
         id: metadata
-        uses: docker/metadata-action@v3
+        uses: docker/metadata-action@v5
         with:
           images: winglian/axolotl
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
       - name: Login to Docker Hub
-        uses: docker/login-action@v2
+        uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
-      - name: Build
-        uses: docker/build-push-action@v4
+      # guidance for testing before pushing: https://docs.docker.com/build/ci/github-actions/test-before-push/
+      - name: Build and export to Docker
+        uses: docker/build-push-action@v5
         with:
           context: .
+          load: true
           build-args: |
-            BASE_TAG=${{ github.ref_name }}-base-py${{ matrix.python_version }}-${{ matrix.cuda }}-${{ matrix.pytorch }}
+            BASE_TAG=${{ github.ref_name }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}
+            CUDA=${{ matrix.cuda }}
+            PYTORCH_VERSION=${{ matrix.pytorch }}
           file: ./docker/Dockerfile
-          push: ${{ github.event_name != 'pull_request' }}
-          tags: ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+          tags: |
+            ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+            ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
           labels: ${{ steps.metadata.outputs.labels }}
+      - name: Unit Tests
+        run: |
+          docker run --rm ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }} pytest --ignore=tests/e2e/ /workspace/axolotl/tests/
+      - name: Push to Docker Hub
+        if: github.event_name != 'pull_request'
+        run: |
+          docker push ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+          latest_tag=${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
+          if [ -n "$latest_tag" ]; then
+            docker push "$latest_tag"
+          fi
+
   build-axolotl-runpod:
     needs: build-axolotl
     if: github.repository_owner == 'OpenAccess-AI-Collective'
@@ -74,27 +97,32 @@ jobs:
             is_latest: true
           - cuda: 118
             cuda_version: 11.8.0
-            python_version: "3.9"
-            pytorch: 2.0.1
-            axolotl_extras: gptq
-    runs-on: self-hosted
+            python_version: "3.10"
+            pytorch: 2.1.1
+            axolotl_extras:
+          - cuda: 121
+            cuda_version: 12.1.0
+            python_version: "3.10"
+            pytorch: 2.1.1
+            axolotl_extras:
+    runs-on: [self-hosted, gpu, docker]
     steps:
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
       - name: Docker metadata
         id: metadata
-        uses: docker/metadata-action@v3
+        uses: docker/metadata-action@v5
         with:
           images: winglian/axolotl-runpod
       - name: Login to Docker Hub
-        uses: docker/login-action@v2
+        uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v2
       - name: Build
-        uses: docker/build-push-action@v4
+        uses: docker/build-push-action@v5
         with:
           context: .
           build-args: |

.github/workflows/pre-commit.yml

@@ -1,16 +0,0 @@
-name: pre-commit
-
-on:
-  pull_request:
-  push:
-
-jobs:
-  pre-commit:
-    runs-on: ubuntu-latest
-    steps:
-    - uses: actions/checkout@v3
-    - uses: actions/setup-python@v4
-      with:
-          python-version: "3.9"
-          cache: 'pip' # caching pip dependencies
-    - uses: pre-commit/action@v3.0.0

.github/workflows/pypi.yml

@@ -0,0 +1,45 @@
+name: publish pypi
+
+on:
+  push:
+    tags:
+      - '*'
+
+jobs:
+  pypi-publish:
+    name: Upload release to PyPI
+    runs-on: ubuntu-latest
+    environment:
+      name: pypi
+      url: https://pypi.org/p/axolotl
+    permissions:
+      id-token: write  # IMPORTANT: this permission is mandatory for trusted publishing
+    steps:
+      - name: Check out repository code
+        uses: actions/checkout@v3
+
+      - name: Setup Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: "3.10"
+
+      - name: Install dependencies
+        run: |
+          pip3 install wheel
+          pip3 install -e .
+          pip3 install -r requirements-tests.txt
+
+      - name: Extract tag name
+        id: tag
+        run: echo ::set-output name=TAG_NAME::$(echo $GITHUB_REF | cut -d / -f 3)
+
+      - name: Update version in setup.py
+        run: >-
+          sed -i -E 's/version="([0-9.]+)",/version="${{ steps.tag.outputs.TAG_NAME }}",/g' setup.py
+
+      - name: Build a binary wheel
+        run: >-
+          python setup.py sdist bdist_wheel
+
+      - name: Publish package distributions to PyPI
+        uses: pypa/gh-action-pypi-publish@release/v1

.github/workflows/tests-docker.yml

@@ -0,0 +1,46 @@
+name: e2e-docker-tests
+
+on:
+  pull_request:
+    paths:
+      - '**.py'
+      - 'requirements.txt'
+      - '.github/workflows/*.yml'
+  workflow_dispatch:
+
+jobs:
+  build-axolotl:
+    if: github.repository_owner == 'OpenAccess-AI-Collective'
+    # this job needs to be run on self-hosted GPU runners...
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - cuda: 118
+            cuda_version: 11.8.0
+            python_version: "3.10"
+            pytorch: 2.0.1
+          - cuda: 121
+            cuda_version: 12.1.0
+            python_version: "3.10"
+            pytorch: 2.1.1
+    runs-on: [self-hosted, gpu, docker]
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Build Docker image
+        run: |
+          # Set up build arguments
+          BASE_TAG="main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}"
+          CUDA="${{ matrix.cuda }}"
+          PYTORCH_VERSION="${{ matrix.pytorch }}"
+          # Build the Docker image
+          docker build . \
+            --file ./docker/Dockerfile \
+            --build-arg BASE_TAG=$BASE_TAG \
+            --build-arg CUDA=$CUDA \
+            --build-arg PYTORCH_VERSION=$PYTORCH_VERSION \
+            --tag test-axolotl
+      - name: Unit Tests w docker image
+        run: |
+          docker run --rm test-axolotl pytest --ignore=tests/e2e/ /workspace/axolotl/tests/

.github/workflows/tests.yml

@@ -1,10 +1,32 @@
-name: PyTest
+name: Tests
 on:
+  # check on push/merge to main, PRs, and manual triggers
   push:
+    branches:
+      - "main"
+    paths:
+      - '**.py'
+      - 'requirements.txt'
   pull_request:
+      paths:
+       - '**.py'
+       - 'requirements.txt'
+  workflow_dispatch:
 
 jobs:
-  test:
+  pre-commit:
+    name: pre-commit
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v3
+      - uses: actions/setup-python@v4
+        with:
+          python-version: "3.9"
+          cache: 'pip' # caching pip dependencies
+      - uses: pre-commit/action@v3.0.0
+
+  pytest:
+    name: PyTest
     runs-on: ubuntu-latest
     strategy:
       fail-fast: false
@@ -24,9 +46,36 @@ jobs:
 
       - name: Install dependencies
         run: |
-          pip install -e .
-          pip install -r requirements-tests.txt
+          pip3 install -U -e .
+          pip3 install -r requirements-tests.txt
 
       - name: Run tests
         run: |
-          pytest tests/
+          pytest --ignore=tests/e2e/ tests/
+
+  e2e-test:
+    name: E2E Tests
+    runs-on: [self-hosted, gpu]
+    timeout-minutes: 20
+    needs: [pre-commit, pytest]
+
+    steps:
+      - name: Check out repository code
+        uses: actions/checkout@v3
+
+      - name: Setup Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: "3.10"
+#          cache: 'pip' # caching pip dependencies
+
+      - name: Install dependencies
+        run: |
+          pip3 install --extra-index-url https://download.pytorch.org/whl/cu118 -U torch==2.0.1
+          pip3 uninstall -y transformers accelerate
+          pip3 install -U -e .[flash-attn,mamba-ssm]
+          pip3 install -r requirements-tests.txt
+
+      - name: Run e2e tests
+        run: |
+          pytest tests/e2e/

.gitignore

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

.isort.cfg

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

.mypy.ini

@@ -8,6 +8,12 @@ ignore_missing_imports = True
 [mypy-axolotl.monkeypatch.*]
 ignore_errors = True
 
+[mypy-axolotl.models.mixtral.*]
+ignore_errors = True
+
+[mypy-axolotl.models.phi.*]
+ignore_errors = True
+
 [mypy-flash_attn.*]
 ignore_missing_imports = True
 
@@ -20,6 +26,9 @@ ignore_missing_imports = True
 [mypy-peft]
 ignore_missing_imports = True
 
+[mypy-wandb]
+ignore_missing_imports = True
+
 [mypy-bitsandbytes]
 ignore_missing_imports = True
 

data/README.md

@@ -1,24 +0,0 @@
-
-## Download some datasets
-```shell
-curl https://raw.githubusercontent.com/tloen/alpaca-lora/main/alpaca_data_gpt4.json -o data/raw/alpaca_data_gpt4.json
-curl https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json -L -o data/raw/vicuna_cleaned.json
-curl https://github.com/teknium1/GPTeacher/blob/main/Instruct/gpt4-instruct-similarity-0.6-dataset.json?raw=true -L -o data/raw/gpt4-instruct-similarity-0.6-dataset.json
-curl https://github.com/teknium1/GPTeacher/blob/main/Roleplay/roleplay-similarity_0.6-instruct-dataset.json?raw=true -L -o data/raw/roleplay-similarity_0.6-instruct-dataset.json
-```
-
-## Convert the JSON data files to JSONL.
-
-```shell
-python3 ./scripts/alpaca_json_to_jsonl.py --file data/alpaca_data_gpt4.json --output data/alpaca_data_gpt4.jsonl
-python3 ./scripts/alpaca_json_to_jsonl.py --file data/raw/vicuna_cleaned.json --output data/vicuna_cleaned.jsonl
-python3 ./scripts/alpaca_json_to_jsonl.py --file data/raw/roleplay-similarity_0.6-instruct-dataset.json --output data/roleplay-similarity_0.6-instruct-dataset.jsonl
-python3 ./scripts/alpaca_json_to_jsonl.py --file data/raw/gpt4-instruct-similarity-0.6-dataset.json --output data/gpt4-instruct-similarity-0.6-dataset.jsonl
-```
----
-
-Using JSONL makes it easier to subset the data if you want a smaller training set, i.e get 2000 random examples.
-
-```shell
-shuf -n2000 data/vicuna_cleaned.jsonl > data/vicuna_cleaned.subset0.jsonl
-```

data/raw/.gitignore

@@ -1 +0,0 @@
-**

deepspeed/zero1.json

@@ -0,0 +1,31 @@
+{
+  "zero_optimization": {
+    "stage": 1,
+    "overlap_comm": true
+  },
+  "bf16": {
+    "enabled": "auto"
+  },
+  "fp16": {
+    "enabled": "auto",
+    "auto_cast": false,
+    "loss_scale": 0,
+    "initial_scale_power": 32,
+    "loss_scale_window": 1000,
+    "hysteresis": 2,
+    "min_loss_scale": 1
+  },
+  "optimizer": {
+    "type": "AdamW",
+    "params": {
+      "lr": "auto",
+      "betas": "auto",
+      "eps": "auto",
+      "weight_decay": "auto"
+    }
+  },
+  "gradient_accumulation_steps": "auto",
+  "train_batch_size": "auto",
+  "train_micro_batch_size_per_gpu": "auto",
+  "wall_clock_breakdown": false
+}

deepspeed/zero2.json

@@ -0,0 +1,35 @@
+{
+  "zero_optimization": {
+    "stage": 2,
+    "offload_optimizer": {
+      "device": "cpu"
+    },
+    "contiguous_gradients": true,
+    "overlap_comm": true
+  },
+  "bf16": {
+    "enabled": "auto"
+  },
+  "fp16": {
+    "enabled": "auto",
+    "auto_cast": false,
+    "loss_scale": 0,
+    "initial_scale_power": 32,
+    "loss_scale_window": 1000,
+    "hysteresis": 2,
+    "min_loss_scale": 1
+  },
+  "optimizer": {
+    "type": "AdamW",
+    "params": {
+      "lr": "auto",
+      "betas": "auto",
+      "eps": "auto",
+      "weight_decay": "auto"
+    }
+  },
+  "gradient_accumulation_steps": "auto",
+  "train_batch_size": "auto",
+  "train_micro_batch_size_per_gpu": "auto",
+  "wall_clock_breakdown": false
+}

deepspeed/zero3.json

@@ -1,14 +1,6 @@
 {
   "zero_optimization": {
     "stage": 3,
-    "offload_optimizer": {
-      "device": "cpu",
-      "pin_memory": true
-    },
-    "offload_param": {
-      "device": "cpu",
-      "pin_memory": true
-    },
     "overlap_comm": true,
     "contiguous_gradients": true,
     "sub_group_size": 0,
@@ -35,22 +27,12 @@
     "type": "AdamW",
     "params": {
       "lr": "auto",
-      "betas": [
-        0.9,
-        0.95
-      ],
-      "eps": 1e-8,
+      "betas": "auto",
+      "eps": "auto",
       "weight_decay": "auto"
     }
   },
-  "scheduler": {
-    "type": "WarmupLR",
-    "params": {
-      "warmup_min_lr": "auto",
-      "warmup_max_lr": "auto",
-      "warmup_num_steps": "auto"
-    }
-  },
+  "gradient_accumulation_steps": "auto",
   "train_batch_size": "auto",
   "train_micro_batch_size_per_gpu": "auto",
   "wall_clock_breakdown": false

deepspeed/zero3_bf16.json

@@ -0,0 +1,39 @@
+{
+  "zero_optimization": {
+    "stage": 3,
+    "overlap_comm": true,
+    "contiguous_gradients": true,
+    "sub_group_size": 0,
+    "reduce_bucket_size": "auto",
+    "stage3_prefetch_bucket_size": "auto",
+    "stage3_param_persistence_threshold": "auto",
+    "stage3_max_live_parameters": 0,
+    "stage3_max_reuse_distance": 0,
+    "stage3_gather_16bit_weights_on_model_save": true
+  },
+  "bf16": {
+    "enabled": true
+  },
+  "fp16": {
+    "enabled": "auto",
+    "auto_cast": false,
+    "loss_scale": 0,
+    "initial_scale_power": 32,
+    "loss_scale_window": 1000,
+    "hysteresis": 2,
+    "min_loss_scale": 1
+  },
+  "optimizer": {
+    "type": "AdamW",
+    "params": {
+      "lr": "auto",
+      "betas": "auto",
+      "eps": "auto",
+      "weight_decay": "auto"
+    }
+  },
+  "gradient_accumulation_steps": "auto",
+  "train_batch_size": "auto",
+  "train_micro_batch_size_per_gpu": "auto",
+  "wall_clock_breakdown": false
+}

docker-compose.yaml

@@ -9,6 +9,11 @@ services:
       - ~/.cache/huggingface/:/root/.cache/huggingface/
     # set environment variables
     environment:
+      # Set environment variables
+      - GIT_AUTHOR_NAME=${GIT_AUTHOR_NAME}
+      - GIT_AUTHOR_EMAIL=${GIT_AUTHOR_EMAIL}
+      - GIT_COMMITTER_NAME=${GIT_COMMITTER_NAME}
+      - GIT_COMMITTER_EMAIL=${GIT_COMMITTER_EMAIL}
       - WANDB_API_KEY=${WANDB_API_KEY}
     deploy:
       resources:

docker/Dockerfile

@@ -5,25 +5,31 @@ ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
 ARG AXOLOTL_EXTRAS=""
 ARG CUDA="118"
 ENV BNB_CUDA_VERSION=$CUDA
+ARG PYTORCH_VERSION="2.0.1"
+
+ENV PYTORCH_VERSION=$PYTORCH_VERSION
 
 RUN apt-get update && \
-    apt-get install -y vim curl
+    apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev
 
 WORKDIR /workspace
 
-RUN pip3 install --force-reinstall "peft @ git+https://github.com/huggingface/peft.git@main"
 RUN git clone --depth=1 https://github.com/OpenAccess-AI-Collective/axolotl.git
+
+WORKDIR /workspace/axolotl
+
 # If AXOLOTL_EXTRAS is set, append it in brackets
-RUN cd axolotl && \
-    if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
-        pip install -e .[$AXOLOTL_EXTRAS]; \
+RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
+        pip install -e .[deepspeed,flash-attn,$AXOLOTL_EXTRAS]; \
     else \
-        pip install -e .; \
+        pip install -e .[deepspeed,flash-attn]; \
     fi
 
+# So we can test the Docker image
+RUN pip install pytest
+
 # fix so that git fetch/pull from remote works
-RUN cd axolotl && \
-    git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
+RUN git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
     git config --get remote.origin.fetch
 
 # helper for huggingface-login cli

docker/Dockerfile-base

@@ -10,95 +10,28 @@ ENV PATH="/root/miniconda3/bin:${PATH}"
 ARG PYTHON_VERSION="3.9"
 ARG PYTORCH_VERSION="2.0.1"
 ARG CUDA="118"
+ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
 
 ENV PYTHON_VERSION=$PYTHON_VERSION
+ENV TORCH_CUDA_ARCH_LIST=$TORCH_CUDA_ARCH_LIST
 
-RUN apt-get update
-RUN apt-get install -y wget git build-essential ninja-build git-lfs libaio-dev && rm -rf /var/lib/apt/lists/*
-
-RUN wget \
+RUN apt-get update \
+    && apt-get install -y wget git build-essential ninja-build git-lfs libaio-dev && rm -rf /var/lib/apt/lists/* \
+    && wget \
     https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh \
     && mkdir /root/.conda \
     && bash Miniconda3-latest-Linux-x86_64.sh -b \
-    && rm -f Miniconda3-latest-Linux-x86_64.sh
-
-RUN conda create -n "py${PYTHON_VERSION}" python="${PYTHON_VERSION}"
+    && rm -f Miniconda3-latest-Linux-x86_64.sh \
+    && conda create -n "py${PYTHON_VERSION}" python="${PYTHON_VERSION}"
 
 ENV PATH="/root/miniconda3/envs/py${PYTHON_VERSION}/bin:${PATH}"
 
 WORKDIR /workspace
 
 RUN python3 -m pip install --upgrade pip && pip3 install packaging && \
-    python3 -m pip install --no-cache-dir -U torch==${PYTORCH_VERSION}+cu${CUDA} --extra-index-url https://download.pytorch.org/whl/cu$CUDA
+    python3 -m pip install --no-cache-dir -U torch==${PYTORCH_VERSION}+cu${CUDA} deepspeed-kernels --extra-index-url https://download.pytorch.org/whl/cu$CUDA
 
-
-FROM base-builder AS flash-attn-builder
-
-WORKDIR /workspace
-
-ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
-
-RUN git clone https://github.com/Dao-AILab/flash-attention.git && \
-    cd flash-attention && \
-    git checkout v2.0.4  && \
-    python3 setup.py bdist_wheel && \
-    cd csrc/fused_dense_lib && \
-    python3 setup.py bdist_wheel && \
-    cd ../xentropy && \
-    python3 setup.py bdist_wheel && \
-    cd ../rotary && \
-    python3 setup.py bdist_wheel && \
-    cd ../layer_norm && \
-    python3 setup.py bdist_wheel
-
-FROM base-builder AS deepspeed-builder
-
-ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
-
-WORKDIR /workspace
-
-RUN git clone https://github.com/microsoft/DeepSpeed.git && \
-    cd DeepSpeed && \
-    MAX_CONCURRENCY=8 DS_BUILD_SPARSE_ATTN=0 DS_BUILD_OPS=1 python3 setup.py bdist_wheel
-
-FROM base-builder AS bnb-builder
-
-WORKDIR /workspace
-ARG CUDA="118"
-ENV CUDA=$CUDA
-
-RUN git clone https://github.com/TimDettmers/bitsandbytes.git && \
-    cd bitsandbytes && \
-    CUDA_VERSION=$CUDA make cuda11x && \
-    python setup.py bdist_wheel
-
-FROM base-builder
-
-ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
-ENV TORCH_CUDA_ARCH_LIST=$TORCH_CUDA_ARCH_LIST
-
-# recompile apex
-RUN python3 -m pip uninstall -y apex
-RUN git clone https://github.com/NVIDIA/apex
-#  `MAX_JOBS=1` disables parallel building to avoid cpu memory OOM when building image on GitHub Action (standard) runners
-RUN cd apex && MAX_JOBS=1 python3 -m pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" ./
-
-RUN mkdir -p /workspace/builds
-COPY --from=bnb-builder /workspace/bitsandbytes /workspace/builds/bitsandbytes
-
-RUN mkdir -p /workspace/wheels/bitsandbytes
-COPY --from=deepspeed-builder /workspace/DeepSpeed/dist/deepspeed-*.whl wheels
-COPY --from=bnb-builder /workspace/bitsandbytes/dist/bitsandbytes-*.whl wheels
-COPY --from=bnb-builder /workspace/bitsandbytes/bitsandbytes/libbitsandbytes*.so wheels/bitsandbytes
-COPY --from=flash-attn-builder /workspace/flash-attention/dist/flash_attn-*.whl wheels
-COPY --from=flash-attn-builder /workspace/flash-attention/csrc/fused_dense_lib/dist/fused_dense_lib-*.whl wheels
-COPY --from=flash-attn-builder /workspace/flash-attention/csrc/xentropy/dist/xentropy_cuda_lib-*.whl wheels
-COPY --from=flash-attn-builder /workspace/flash-attention/csrc/rotary/dist/rotary_emb-*.whl wheels
-COPY --from=flash-attn-builder /workspace/flash-attention/csrc/layer_norm/dist/dropout_layer_norm-*.whl wheels
-
-RUN pip3 install wheels/deepspeed-*.whl wheels/flash_attn-*.whl wheels/fused_dense_lib-*.whl wheels/xentropy_cuda_lib-*.whl wheels/rotary_emb-*.whl wheels/dropout_layer_norm-*.whl
-RUN cd /workspace/builds/bitsandbytes && python3 setup.py install
-RUN git lfs install --skip-repo
-RUN pip3 install awscli && \
+RUN git lfs install --skip-repo && \
+    pip3 install awscli && \
     # The base image ships with `pydantic==1.8.2` which is not working
     pip3 install -U --no-cache-dir pydantic==1.10.10

docker/Dockerfile-runpod

@@ -4,6 +4,7 @@ FROM winglian/axolotl:$BASE_TAG
 ENV HF_DATASETS_CACHE="/workspace/data/huggingface-cache/datasets"
 ENV HUGGINGFACE_HUB_CACHE="/workspace/data/huggingface-cache/hub"
 ENV TRANSFORMERS_CACHE="/workspace/data/huggingface-cache/hub"
+ENV HF_HOME="/workspace/data/huggingface-cache/hub"
 
 COPY scripts/runpod-entrypoint.sh /root/runpod-entrypoint.sh
 

docs/faq.md

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

docs/multi-node.md

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

docs/multipack.md

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

docs/nccl.md

@@ -0,0 +1,46 @@
+# NCCL
+
+NVIDIA NCCL is a library to facilitate and optimize multi-GPU communication operations, such as broadcast, all-gather, reduce, all-reduce, etc. Broadly, NCCL configuration is highly environment-specific and is configured via several [environment variables](https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/env.html). A common NCCL-related problem occurs when a long-running operation times out causing the training process to abort:
+
+```text
+Watchdog caught collective operation timeout: WorkNCCL(SeqNum=42, OpType=ALLGATHER, Timeout(ms)=1800000) ran for 1806948 milliseconds before timing out.
+```
+
+Often, this timeout will happen after 30 minutes (the default setting) and is accompanied by below-average power consumption with near 100% GPU utilization before the error is raised. Nvidia recommends [disabling PCI access control services (ACS)](https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/troubleshooting.html#pci-access-control-services-acs) as a possible solution if this is available to you.
+
+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
+nvidia-smi nvlink --status
+```
+
+To force NCCL to use NVLink, simply set this in the environment:
+
+```shell
+export NCCL_P2P_LEVEL=NVL
+```
+
+If NVLink is not available in your environment there are other options for ``NCCL_P2P_LEVEL`` in the table below:
+
+| NCCL_P2P_LEVEL | Description |
+| -------------- | ----------- |
+| PIX | P2P data transfers through no more than a single PCIe bridge. Faster data transfer rates vs to paths involving multiple bridges, but slower compared to direct GPU-to-GPU communication. |
+| PXB | P2P data transfers through multiple PCIe bridges but not going through the PCIe Host Bridge; this path involves a complex routing process, potentially incurring a moderate level of latency. |
+| PHB | P2P data transfers occur over the PCIe and through a PCIe Host Bridge, typically involving the CPU, which can facilitate direct memory access but might introduce additional latency compared to more direct paths (ex PIX, NVL) |
+
+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
+./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
+export NCCL_DEBUG=INFO
+export NCCL_DEBUG_SUBSYS=ALL
+export TORCH_DISTRIBUTED_DEBUG=INFO
+export TORCHELASTIC_ERROR_FILE=/PATH/TO/torcherror.log
+```
+
+Finally, if you believe your training job needs more time you can increase the timeout past 30 minutes by setting the ``ddp_timeout`` value in the Axolotl configuration. See [PyTorch init_process_group](https://pytorch.org/docs/stable/distributed.html#torch.distributed.init_process_group) for documentation on this value.

docs/rlhf.md

@@ -0,0 +1,35 @@
+# RLHF (Beta)
+
+### 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:
+
+- Proximal Policy Optimization (PPO) (not yet supported in axolotl)
+- Direct Preference Optimization (DPO)
+- Identity Preference Optimization (IPO)
+
+
+### 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.
+
+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
+
+#### DPO
+```yaml
+rl: true
+datasets:
+  - path: Intel/orca_dpo_pairs
+    split: train
+    type: intel_apply_chatml
+  - path: argilla/ultrafeedback-binarized-preferences
+    split: train
+    type: argilla_apply_chatml
+```
+
+#### IPO
+```yaml
+rl: ipo
+```

examples/cerebras/btlm-ft.yml

@@ -0,0 +1,89 @@
+base_model: cerebras/btlm-3b-8k-base
+model_type: AutoModelForCausalLM
+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
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path: last_prepared_run
+val_set_size: 0.05
+
+adapter:
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len:
+sample_packing: false
+sample_packing_eff_est:
+sample_packing_seq_len_multiplier:
+total_num_tokens:
+
+lora_r:
+lora_alpha:
+lora_dropout:
+lora_target_modules:
+lora_target_linear:
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+output_dir: btlm-out
+gradient_accumulation_steps: 1
+micro_batch_size: 1
+num_epochs: 1
+optimizer: adamw_torch
+adam_beta2: 0.95
+adam_eps: 0.000000001
+max_grad_norm: 1.0
+
+torchdistx_path:
+lr_scheduler: cosine
+lr_quadratic_warmup: true
+learning_rate: 0.000085
+train_on_inputs: true
+group_by_length: false
+bf16: true
+fp16: false
+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:
+
+gptq_groupsize:
+gptq_model_v1:
+
+warmup_steps: 32
+evals_per_epoch: 4
+saves_per_epoch: 1
+save_total_limit:
+
+debug:
+deepspeed:
+weight_decay: 0.1
+special_tokens:
+  pad_token: "<|endoftext|>"
+fsdp:
+#  - full_shard
+#  - auto_wrap
+fsdp_config:
+#  fsdp_state_dict_type: FULL_STATE_DICT
+#  fsdp_transformer_layer_cls_to_wrap: BTLMBlock

examples/cerebras/qlora.yml

@@ -1,5 +1,4 @@
 base_model: cerebras/Cerebras-GPT-1.3B
-base_model_config: cerebras/Cerebras-GPT-1.3B
 load_in_8bit: false
 load_in_4bit: true
 strict: false
@@ -7,8 +6,8 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 adapter: qlora
 lora_model_dir:
 sequence_len: 2048
@@ -25,7 +24,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./qlora-out
 batch_size: 4
@@ -50,8 +49,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-eval_steps: 20
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.1

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

@@ -0,0 +1,67 @@
+base_model: codellama/CodeLlama-13b-hf
+model_type: LlamaForCausalLM
+tokenizer_type: CodeLlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: true
+load_in_4bit: false
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./lora-out
+
+sequence_len: 4096
+sample_packing: true
+pad_to_sequence_len: true
+
+adapter: lora
+lora_model_dir:
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 4
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+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>"
+  unk_token: "<unk>"

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

@@ -0,0 +1,69 @@
+base_model: codellama/CodeLlama-13b-hf
+model_type: LlamaForCausalLM
+tokenizer_type: CodeLlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./qlora-out
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 4096
+sample_packing: true
+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:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 4
+optimizer: paged_adamw_32bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+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>"
+  unk_token: "<unk>"

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

@@ -0,0 +1,67 @@
+base_model: codellama/CodeLlama-34b-hf
+model_type: LlamaForCausalLM
+tokenizer_type: CodeLlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: true
+load_in_4bit: false
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./lora-out
+
+sequence_len: 4096
+sample_packing: true
+pad_to_sequence_len: true
+
+adapter: lora
+lora_model_dir:
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 4
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+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>"
+  unk_token: "<unk>"

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

@@ -0,0 +1,69 @@
+base_model: codellama/CodeLlama-34b-hf
+model_type: LlamaForCausalLM
+tokenizer_type: CodeLlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./qlora-out
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 4096
+sample_packing: true
+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:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 4
+optimizer: paged_adamw_32bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+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>"
+  unk_token: "<unk>"

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

@@ -0,0 +1,67 @@
+base_model: codellama/CodeLlama-7b-hf
+model_type: LlamaForCausalLM
+tokenizer_type: CodeLlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: true
+load_in_4bit: false
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./lora-out
+
+sequence_len: 4096
+sample_packing: true
+pad_to_sequence_len: true
+
+adapter: lora
+lora_model_dir:
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 4
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+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>"
+  unk_token: "<unk>"

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

@@ -0,0 +1,69 @@
+base_model: codellama/CodeLlama-7b-hf
+model_type: LlamaForCausalLM
+tokenizer_type: CodeLlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./qlora-out
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 4096
+sample_packing: true
+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:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 4
+optimizer: paged_adamw_32bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+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>"
+  unk_token: "<unk>"

examples/code-llama/README.md

@@ -0,0 +1,22 @@
+# Overview
+
+This is an example of CodeLLaMA configuration for 7b, 13b and 34b.
+
+The 7b variant fits on any 24GB VRAM GPU and will take up about 17 GB of VRAM during training if using qlora and 20 GB if using lora. On a RTX 4090 it trains 3 epochs of the default dataset in about 15 minutes.
+
+The 13b variant will fit if you change these settings to these values:
+gradient_accumulation_steps: 2
+micro_batch_size: 1
+
+The 34b variant does not fit on 24GB of VRAM - you will need something with +40 gb VRAM that also supports flash attention v2 - A6000 or A100 are good choices.
+
+```shell
+accelerate launch scripts/finetune.py examples/code-llama/[MODEL_SIZE]/qlora.yml
+
+```
+or
+
+```shell
+accelerate launch scripts/finetune.py examples/code-llama/[MODEL_SIZE]/lora.yml
+
+```

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

@@ -1,8 +1,8 @@
 base_model: tiiuae/falcon-7b
-base_model_config: tiiuae/falcon-7b
 trust_remote_code: true
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
+is_falcon_derived_model: true
 load_in_8bit: true
 load_in_4bit: false
 gptq: false
@@ -11,8 +11,8 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca:chat
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 adapter: lora
 lora_model_dir:
 sequence_len: 2048
@@ -26,7 +26,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./falcon-7b
 batch_size: 2
@@ -51,8 +51,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 40
-eval_steps: 5
-save_steps: 43
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

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

@@ -1,11 +1,11 @@
 # 1b: tiiuae/falcon-rw-1b
 # 40b: tiiuae/falcon-40b
 base_model: tiiuae/falcon-7b
-base_model_config: tiiuae/falcon-7b
 # required by falcon custom model code: https://huggingface.co/tiiuae/falcon-7b/tree/main
 trust_remote_code: true
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
+is_falcon_derived_model: true
 load_in_8bit: false
 # enable 4bit for QLoRA
 load_in_4bit: true
@@ -17,8 +17,8 @@ datasets:
     data_files:
       - Chain-of-Thought/formatted_cot_data/gsm8k_train.json
     type: "alpaca:chat"
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 # enable QLoRA
 adapter: qlora
 lora_model_dir:
@@ -40,7 +40,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./qlora-out
 
@@ -53,7 +53,7 @@ output_dir: ./qlora-out
 # decrease if OOM, increase for max VRAM utilization
 micro_batch_size: 1
 gradient_accumulation_steps: 2
-num_epochs: 3
+num_epochs: 4
 # Optimizer for QLoRA
 optimizer: paged_adamw_32bit
 torchdistx_path:
@@ -80,8 +80,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-eval_steps: 5
-save_steps: 10
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.000001

examples/falcon/config-7b.yml

@@ -1,8 +1,8 @@
 base_model: tiiuae/falcon-7b
-base_model_config: tiiuae/falcon-7b
 trust_remote_code: true
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
+is_falcon_derived_model: true
 load_in_8bit: false
 load_in_4bit: false
 gptq: false
@@ -11,8 +11,8 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca:chat
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 adapter:
 lora_model_dir:
 sequence_len: 2048
@@ -26,7 +26,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./falcon-7b
 batch_size: 2
@@ -51,8 +51,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 40
-eval_steps: 5
-save_steps: 43
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

examples/gptj/qlora.yml

@@ -1,5 +1,4 @@
 base_model: EleutherAI/gpt-j-6b
-base_model_config: EleutherAI/gpt-j-6b
 load_in_8bit: false
 load_in_4bit: true
 strict: false
@@ -7,8 +6,8 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 adapter: qlora
 lora_model_dir:
 sequence_len: 2048
@@ -22,7 +21,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./qlora-out
 gradient_accumulation_steps: 2
@@ -47,8 +46,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-eval_steps: 20
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.1

examples/gptq-lora-7b/README.md

@@ -1,8 +0,0 @@
-# LLaMa 7B using LoRA
-
-This is a good place to start for beginners. This will run on an NVIDIA RTX4090 with no other changes needed.
-
-```shell
-accelerate launch scripts/finetune.py examples/gptq-lora-7b/config.yml
-
-```

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

@@ -1,63 +0,0 @@
-base_model: Neko-Institute-of-Science/LLaMA-7B-4bit-128g
-base_model_config: Neko-Institute-of-Science/LLaMA-7B-4bit-128g
-model_type: LlamaForCausalLM
-tokenizer_type: LlamaTokenizer
-trust_remote_code:
-load_in_8bit: true
-gptq: true
-datasets:
-  - path: vicgalle/alpaca-gpt4
-    type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.02
-adapter:
-lora_model_dir:
-sequence_len: 2048
-max_packed_sequence_len:
-lora_r: 8
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules:
-  - q_proj
-  - v_proj
-lora_fan_in_fan_out: false
-wandb_project: llama-7b-lora-int4
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-output_dir: ./llama-7b-lora-int4
-gradient_accumulation_steps: 1
-micro_batch_size: 1
-num_epochs: 3
-optimizer: adamw_bnb_8bit
-torchdistx_path:
-lr_scheduler: cosine
-learning_rate: 0.0000002
-train_on_inputs: false
-group_by_length: false
-fp16: true
-bf16: false
-tf32: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 5
-xformers_attention:
-flash_attention:
-gradient_checkpointing: true
-gptq_groupsize: 128
-gptq_model_v1: false
-warmup_steps: 20
-eval_steps: 110
-save_steps: 660
-debug:
-deepspeed:
-weight_decay: 0.0001
-fsdp:
-fsdp_config:
-tokens:
-  pad_token: "[PAD]"
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"

examples/jeopardy-bot/config.yml

@@ -1,12 +1,11 @@
 base_model: huggyllama/llama-7b
-base_model_config: huggyllama/llama-7b
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 load_in_8bit: false
 datasets:
   - path: openaccess-ai-collective/jeopardy
     type: jeopardy
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter:
 lora_model_dir:
@@ -20,12 +19,12 @@ lora_fan_in_fan_out: false
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./jeopardy-bot-7b
 gradient_accumulation_steps: 1
 micro_batch_size: 1
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
@@ -43,8 +42,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-eval_steps: 110
-save_steps: 660
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.1

examples/llama-2/README.md

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

examples/llama-2/fft_optimized.yml

@@ -0,0 +1,72 @@
+base_model: NousResearch/Llama-2-7b-hf
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: false
+load_in_4bit: false
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.05
+output_dir: ./out
+
+sequence_len: 4096
+sample_packing: true
+pad_to_sequence_len: true
+
+adapter:
+lora_model_dir:
+lora_r:
+lora_alpha:
+lora_dropout:
+lora_target_linear:
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 1
+micro_batch_size: 1
+num_epochs: 1
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+flash_attn_cross_entropy: false
+flash_attn_rms_norm: true
+flash_attn_fuse_qkv: false
+flash_attn_fuse_mlp: true
+
+warmup_steps: 100
+evals_per_epoch: 4
+eval_table_size:
+saves_per_epoch: 1
+debug:
+deepspeed: #deepspeed/zero2.json # multi-gpu only
+weight_decay: 0.1
+fsdp:
+fsdp_config:
+special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"

examples/llama-2/gptq-lora.yml

@@ -0,0 +1,73 @@
+base_model: TheBloke/Llama-2-7B-GPTQ
+is_llama_derived_model: false
+gptq: true
+gptq_disable_exllama: true
+model_type: AutoModelForCausalLM
+tokenizer_type: LlamaTokenizer
+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
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+adapter: lora
+lora_model_dir:
+sequence_len: 4096
+sample_packing:
+lora_r: 8
+lora_alpha: 32
+lora_dropout: 0.05
+lora_target_modules:
+  - k_proj
+  - o_proj
+  - q_proj
+  - v_proj
+lora_target_linear:
+lora_fan_in_fan_out:
+wandb_project:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+output_dir: ./model-out
+gradient_accumulation_steps: 1
+micro_batch_size: 1
+num_epochs: 4
+optimizer: adamw_torch
+adam_beta2: 0.95
+adam_eps: 0.00001
+max_grad_norm: 1.0
+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>"
+  eos_token: "</s>"
+  unk_token: "<unk>"

examples/llama-2/lora.yml

@@ -1,5 +1,4 @@
-base_model: meta-llama/Llama-2-7b-hf
-base_model_config: meta-llama/Llama-2-7b-hf
+base_model: NousResearch/Llama-2-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 is_llama_derived_model: true
@@ -11,12 +10,13 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 output_dir: ./lora-out
 
 sequence_len: 4096
 sample_packing: true
+pad_to_sequence_len: true
 
 adapter: lora
 lora_model_dir:
@@ -29,12 +29,12 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -54,8 +54,10 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 20
-save_steps:
+evals_per_epoch: 4
+eval_table_size:
+eval_table_max_new_tokens: 128
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

examples/llama-2/qlora.yml

@@ -1,5 +1,4 @@
-base_model: meta-llama/Llama-2-7b-hf
-base_model_config: meta-llama/Llama-2-7b-hf
+base_model: NousResearch/Llama-2-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 is_llama_derived_model: true
@@ -11,8 +10,8 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 output_dir: ./qlora-out
 
 adapter: qlora
@@ -20,6 +19,7 @@ lora_model_dir:
 
 sequence_len: 4096
 sample_packing: true
+pad_to_sequence_len: true
 
 lora_r: 32
 lora_alpha: 16
@@ -31,12 +31,12 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 3
+num_epochs: 4
 optimizer: paged_adamw_32bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -56,8 +56,9 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 20
-save_steps:
+evals_per_epoch: 4
+eval_table_size:
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

examples/llama-2/relora.yml

@@ -0,0 +1,73 @@
+base_model: NousResearch/Llama-2-7b-hf
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./relora-out
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 4096
+sample_packing: true
+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
+relora_cpu_offload: false
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 4
+num_epochs: 4
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+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>"
+  unk_token: "<unk>"

examples/mamba/config.yml

@@ -0,0 +1,61 @@
+base_model: state-spaces/mamba-2.8b
+model_type: MambaLMHeadModel
+tokenizer_type: AutoTokenizer
+tokenizer_config: EleutherAI/gpt-neox-20b
+
+load_in_8bit: false
+load_in_4bit: false
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.0
+output_dir: ./out
+
+sequence_len: 2048
+sample_packing: false
+pad_to_sequence_len: false
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 1
+num_epochs: 2
+optimizer: paged_adamw_8bit
+lr_scheduler: cosine
+learning_rate: 5e-5
+
+train_on_inputs: false
+group_by_length: true
+
+bf16: true
+fp16: false
+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_table_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/README.md

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

examples/mistral/config.yml

@@ -0,0 +1,62 @@
+base_model: mistralai/Mistral-7B-v0.1
+model_type: MistralForCausalLM
+tokenizer_type: LlamaTokenizer
+is_mistral_derived_model: true
+
+load_in_8bit: false
+load_in_4bit: false
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./out
+
+sequence_len: 8192
+sample_packing: true
+pad_to_sequence_len: true
+eval_sample_packing: false
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 4
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.000005
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+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:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"

examples/mistral/mixtral.yml

@@ -0,0 +1,91 @@
+base_model: mistralai/Mixtral-8x7B-v0.1
+model_type: AutoModelForCausalLM
+tokenizer_type: LlamaTokenizer
+trust_remote_code: true
+
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+
+datasets:
+  - path: tatsu-lab/alpaca
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.0
+output_dir: ./qlora-out
+
+## You can optionally freeze the entire model and unfreeze a subset of parameters
+unfrozen_parameters:
+#  - lm_head.*
+#  - model.embed_tokens.*
+#  - model.layers.2[0-9]+.block_sparse_moe.gate.*
+#  - model.layers.2[0-9]+.block_sparse_moe.experts.*
+#  - model.layers.3[0-9]+.block_sparse_moe.gate.*
+#  - model.layers.3[0-9]+.block_sparse_moe.experts.*
+
+model_config:
+  output_router_logits: true
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 4096
+sample_packing: true
+pad_to_sequence_len: true
+
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_fan_in_fan_out:
+#lora_target_modules:
+#  - gate
+#  - q_proj
+#  - k_proj
+#  - v_proj
+#  - o_proj
+#  - w1
+#  - w2
+#  - w3
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 2
+micro_batch_size: 1
+num_epochs: 1
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+loss_watchdog_threshold: 5.0
+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: deepspeed/zero2.json
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:

examples/mistral/qlora.yml

@@ -0,0 +1,81 @@
+base_model: mistralai/Mistral-7B-v0.1
+model_type: MistralForCausalLM
+tokenizer_type: LlamaTokenizer
+is_mistral_derived_model: true
+
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.1
+output_dir: ./qlora-out
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 8192
+sample_packing: true
+pad_to_sequence_len: true
+
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_fan_in_fan_out:
+lora_target_modules:
+  - gate_proj
+  - down_proj
+  - up_proj
+  - q_proj
+  - v_proj
+  - k_proj
+  - o_proj
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_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
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+loss_watchdog_threshold: 5.0
+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:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"

examples/mpt-7b/config.yml

@@ -1,12 +1,11 @@
 base_model: mosaicml/mpt-7b
-base_model_config: mosaicml/mpt-7b
 tokenizer_type: AutoTokenizer
 trust_remote_code: true  # required for mpt as their model class is not merged into transformers yet
 load_in_8bit: false
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter:
 lora_model_dir:
@@ -22,12 +21,12 @@ lora_fan_in_fan_out: false
 wandb_project: mpt-alpaca-7b
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./mpt-alpaca-7b
 gradient_accumulation_steps: 1
 micro_batch_size: 1
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
@@ -45,8 +44,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-eval_steps: 110
-save_steps: 660
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0001

examples/openllama-3b/config.yml

@@ -1,5 +1,4 @@
-base_model: openlm-research/open_llama_3b
-base_model_config: openlm-research/open_llama_3b
+base_model: openlm-research/open_llama_3b_v2
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 load_in_8bit: false
@@ -9,12 +8,12 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter:
 lora_model_dir:
-sequence_len: 256
-max_packed_sequence_len:
+sequence_len: 1024
+sample_packing: true
 lora_r:
 lora_alpha:
 lora_dropout:
@@ -24,16 +23,16 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./openllama-out
 gradient_accumulation_steps: 1
 micro_batch_size: 1
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
-learning_rate: 0.00001
+learning_rate: 0.000003
 train_on_inputs: false
 group_by_length: false
 float16: true
@@ -45,13 +44,13 @@ early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
 logging_steps: 1
-xformers_attention: true
-flash_attention:
+xformers_attention:
+flash_attention: true
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 10
-eval_steps: 50
-save_steps:
+warmup_steps: 20
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.1

examples/openllama-3b/lora.yml

@@ -1,5 +1,4 @@
-base_model: openlm-research/open_llama_3b
-base_model_config: openlm-research/open_llama_3b
+base_model: openlm-research/open_llama_3b_v2
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 load_in_8bit: true
@@ -9,12 +8,12 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter: lora
 lora_model_dir:
-sequence_len: 256
-max_packed_sequence_len:
+sequence_len: 1024
+sample_packing: true
 lora_r: 8
 lora_alpha: 16
 lora_dropout: 0.0
@@ -30,12 +29,12 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./lora-out
-batch_size: 16
-micro_batch_size: 4
-num_epochs: 3
+gradient_accumulation_steps: 1
+micro_batch_size: 2
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
@@ -50,16 +49,16 @@ early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
 logging_steps: 1
-xformers_attention: true
-flash_attention:
+xformers_attention:
+flash_attention: true
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 10
-eval_steps: 50
-save_steps:
+warmup_steps: 20
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
-weight_decay: 0.0
+weight_decay: 0.1
 fsdp:
 fsdp_config:
 special_tokens:

examples/openllama-3b/qlora.yml

@@ -1,5 +1,4 @@
-base_model: openlm-research/open_llama_3b
-base_model_config: openlm-research/open_llama_3b
+base_model: openlm-research/open_llama_3b_v2
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 load_in_8bit: false
@@ -9,12 +8,12 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 adapter: qlora
 lora_model_dir:
-sequence_len: 2048
-max_packed_sequence_len: 2048
+sequence_len: 1024
+sample_packing: true
 lora_r: 8
 lora_alpha: 32
 lora_dropout: 0.05
@@ -24,36 +23,36 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./qlora-out
-batch_size: 4
-micro_batch_size: 4
-num_epochs: 2
+gradient_accumulation_steps: 1
+micro_batch_size: 2
+num_epochs: 4
 optimizer: paged_adamw_32bit
 torchdistx_path:
 lr_scheduler: cosine
 learning_rate: 0.0002
 train_on_inputs: false
 group_by_length: false
-bf16: true
-fp16: false
-tf32: true
+bf16: false
+fp16: true
+tf32: false
 gradient_checkpointing: true
 early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
 logging_steps: 1
-xformers_attention: true
-flash_attention:
+xformers_attention:
+flash_attention: true
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 10
-eval_steps: 20
-save_steps:
+warmup_steps: 20
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
-weight_decay: 0.0
+weight_decay: 0.1
 fsdp:
 fsdp_config:
 special_tokens:

examples/phi/README.md

@@ -0,0 +1,11 @@
+# Phi
+
+Due to some nuances with the phi code, please use deepspeed when training phi for full finetune.
+
+```shell
+accelerate launch -m axolotl.cli.train examples/phi/phi-ft.yml --deepspeed deepspeed/zero1.json
+
+# OR
+
+python -m axolotl.cli.train examples/phi/phi-qlora.yml
+```

examples/phi/phi-ft.yml

@@ -0,0 +1,74 @@
+base_model: microsoft/phi-1_5
+model_type: PhiForCausalLM
+tokenizer_type: AutoTokenizer
+is_llama_derived_model: false
+trust_remote_code: true
+
+load_in_8bit: false
+load_in_4bit: false
+strict: false
+
+datasets:
+  - path: garage-bAInd/Open-Platypus
+    type: alpaca
+
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./phi-sft-out
+
+sequence_len: 2048
+sample_packing: true
+pad_to_sequence_len:
+
+adapter:
+lora_model_dir:
+lora_r:
+lora_alpha:
+lora_dropout:
+lora_target_linear:
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 1
+micro_batch_size: 1
+num_epochs: 4
+optimizer: adamw_torch
+adam_beta2: 0.95
+adam_epsilon: 0.00001
+max_grad_norm: 1.0
+lr_scheduler: cosine
+learning_rate: 0.000003
+
+train_on_inputs: false
+group_by_length: true
+bf16: true
+fp16: false
+tf32: true
+
+gradient_checkpointing:
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention:
+
+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:
+  bos_token: "<|endoftext|>"
+  eos_token: "<|endoftext|>"
+  unk_token: "<|endoftext|>"
+  pad_token: "<|endoftext|>"

examples/phi/phi-qlora.yml

@@ -0,0 +1,74 @@
+base_model: microsoft/phi-1_5
+model_type: AutoModelForCausalLM
+tokenizer_type: AutoTokenizer
+is_llama_derived_model: false
+trust_remote_code: true
+
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+
+datasets:
+  - path: garage-bAInd/Open-Platypus
+    type: alpaca
+
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./phi-sft-out
+
+sequence_len: 1024
+sample_packing: false  # not CURRENTLY compatible with LoRAs
+pad_to_sequence_len:
+
+adapter: qlora
+lora_model_dir:
+lora_r: 64
+lora_alpha: 32
+lora_dropout: 0.05
+lora_target_linear: true
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 1
+micro_batch_size: 1
+num_epochs: 4
+optimizer: adamw_torch
+adam_beta2: 0.95
+adam_epsilon: 0.00001
+max_grad_norm: 1.0
+lr_scheduler: cosine
+learning_rate: 0.000003
+
+train_on_inputs: false
+group_by_length: true
+bf16: true
+fp16: false
+tf32: true
+
+gradient_checkpointing:
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention:
+
+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:
+  bos_token: "<|endoftext|>"
+  eos_token: "<|endoftext|>"
+  unk_token: "<|endoftext|>"
+  pad_token: "<|endoftext|>"

examples/pythia-12b/config.yml

@@ -1,5 +1,4 @@
 base_model: EleutherAI/pythia-12b-deduped
-base_model_config: EleutherAI/pythia-12b-deduped
 base_model_ignore_patterns: pytorch*  # prefer safetensors
 model_type: GPTNeoXForCausalLM
 tokenizer_type: AutoTokenizer
@@ -10,7 +9,7 @@ device_map: auto
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.05
 adapter:
 lora_model_dir:
@@ -25,7 +24,7 @@ lora_fan_in_fan_out: true  # pythia/GPTNeoX lora specific
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./pythia-12b
 gradient_accumulation_steps: 1
@@ -47,4 +46,3 @@ local_rank:
 gradient_checkpointing: true
 fsdp:
 fsdp_config:
-collator_pad_to_longest: true

examples/pythia/lora.yml

@@ -1,10 +1,9 @@
 base_model: EleutherAI/pythia-1.4b-deduped
-base_model_config: EleutherAI/pythia-1.4b-deduped
 load_in_8bit: true
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.05
 adapter: lora
 lora_model_dir:
@@ -19,20 +18,20 @@ lora_fan_in_fan_out: true  # pythia/GPTNeoX lora specific
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./lora-alpaca-pythia
 gradient_accumulation_steps: 1
 micro_batch_size: 4
-num_epochs: 3
+num_epochs: 4
 learning_rate: 0.00001
 train_on_inputs: false
 group_by_length: false
-bf16: True
-tf32: True
+bf16: true
+tf32: true
 early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
 weight_decay: 0.1
-eval_steps: 20
+evals_per_epoch: 4
 logging_steps: 1

examples/qwen/lora.yml

@@ -0,0 +1,68 @@
+base_model: Qwen/Qwen-7B
+model_type: AutoModelForCausalLM
+tokenizer_type: AutoTokenizer
+
+is_qwen_derived_model: true
+trust_remote_code: true
+
+load_in_8bit: true
+load_in_4bit: false
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./lora-out
+
+sequence_len: 2048  # supports up to 8192
+sample_packing: false
+pad_to_sequence_len:
+
+adapter: lora
+lora_model_dir:
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 4
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: 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_table_max_new_tokens: 128
+saves_per_epoch: 1
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:

examples/qwen/qlora.yml

@@ -0,0 +1,68 @@
+base_model: Qwen/Qwen-7B
+model_type: AutoModelForCausalLM
+tokenizer_type: AutoTokenizer
+
+is_qwen_derived_model: true
+trust_remote_code: true
+
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./lora-out
+
+sequence_len: 2048  # supports up to 8192
+sample_packing: false
+pad_to_sequence_len:
+
+adapter: qlora
+lora_model_dir:
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 4
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: 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_table_max_new_tokens: 128
+saves_per_epoch: 1
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:

examples/redpajama/config-3b.yml

@@ -1,5 +1,4 @@
 base_model: togethercomputer/RedPajama-INCITE-Chat-3B-v1
-base_model_config: togethercomputer/RedPajama-INCITE-Chat-3B-v1
 model_type: GPTNeoXForCausalLM
 tokenizer_type: AutoTokenizer
 trust_remote_code:
@@ -7,7 +6,7 @@ load_in_8bit: false
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter:
 lora_model_dir:
@@ -23,12 +22,12 @@ lora_fan_in_fan_out: false
 wandb_project: redpajama-alpaca-3b
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./redpajama-alpaca-3b
 batch_size: 4
 micro_batch_size: 1
-num_epochs: 3
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
@@ -46,8 +45,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-eval_steps: 110
-save_steps: 660
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0001

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

@@ -1,11 +1,10 @@
 base_model: replit/replit-code-v1-3b
-base_model_config: replit/replit-code-v1-3b
 trust_remote_code: true
 load_in_8bit: false
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.05
 adapter: lora
 lora_model_dir:
@@ -22,12 +21,12 @@ lora_fan_in_fan_out:
 wandb_project: lora-replit
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./lora-replit
 batch_size: 8
 micro_batch_size: 1
-num_epochs: 3
+num_epochs: 4
 optimizer:
 torchdistx_path:
 lr_scheduler:
@@ -46,8 +45,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-eval_steps: 50
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0

examples/tiny-llama/README.md

@@ -0,0 +1,17 @@
+# Overview
+
+This is a simple example of how to finetune TinyLlama1.1B using either lora or qlora:
+
+LoRa:
+
+```
+accelerate launch -m axolotl.cli.train examples/tiny-llama/lora.yml
+```
+
+qLoRa:
+
+```
+accelerate launch -m axolotl.cli.train examples/tiny-llama/qlora.yml
+```
+
+Both take about 10 minutes to complete on a 4090.

examples/tiny-llama/lora.yml

@@ -0,0 +1,64 @@
+base_model: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: true
+load_in_4bit: false
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./lora-out
+
+sequence_len: 4096
+sample_packing: true
+pad_to_sequence_len: true
+
+adapter: lora
+lora_model_dir:
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 4
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+evals_per_epoch: 4
+saves_per_epoch: 1
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:

examples/tiny-llama/pretrain.yml

@@ -0,0 +1,58 @@
+base_model: TinyLlama/TinyLlama-1.1B-Chat-v1.0
+
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: false
+load_in_4bit: false
+strict: false
+
+max_steps: 200
+pretraining_dataset:
+  path: c4
+  name: en
+dataset_prepared_path:
+val_set_size: 0.0
+output_dir: ./model-out
+
+sequence_len: 2048
+sample_packing: true
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 4
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+evals_per_epoch:
+eval_table_size:
+saves_per_epoch: 1
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:

examples/tiny-llama/qlora.yml

@@ -0,0 +1,66 @@
+base_model: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path:
+val_set_size: 0.05
+output_dir: ./qlora-out
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 4096
+sample_packing: true
+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:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 4
+optimizer: paged_adamw_32bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+evals_per_epoch: 4
+saves_per_epoch: 1
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:

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

@@ -1,7 +1,6 @@
 # An example finetuning Saleforce's XGen-7b model with 8k context using qlora
 # on Tim Dettmer's Guanaco dataset.
 base_model: Salesforce/xgen-7b-8k-base
-base_model_config: Salesforce/xgen-7b-8k-base
 trust_remote_code: true
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
@@ -16,8 +15,8 @@ datasets:
     data_files:
       - openassistant_best_replies_train.jsonl
     type: "completion"
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
+dataset_prepared_path:
+val_set_size: 0.05
 # enable QLoRA
 adapter: qlora
 lora_model_dir:
@@ -39,7 +38,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_run_id:
+wandb_name:
 wandb_log_model:
 output_dir: ./qlora-out
 
@@ -52,7 +51,7 @@ output_dir: ./qlora-out
 # decrease if OOM, increase for max VRAM utilization
 micro_batch_size: 1
 gradient_accumulation_steps: 1
-num_epochs: 3
+num_epochs: 4
 # Optimizer for QLoRA
 optimizer: paged_adamw_32bit
 torchdistx_path:
@@ -79,8 +78,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-eval_steps: 50
-save_steps: 50
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

examples/yi-34B-chat/README.md

@@ -0,0 +1,5 @@
+# Overview
+
+This is an example of a Yi-34B-Chat configuration. It demonstrates that it is possible to finetune a 34B model on a GPU with 24GB of VRAM.
+
+Tested on an RTX 4090 with `python -m axolotl.cli.train examples/mistral/qlora.yml`, a single epoch of finetuning on the alpaca dataset using qlora runs in 47 mins, using 97% of available memory.

examples/yi-34B-chat/qlora.yml

@@ -0,0 +1,76 @@
+base_model: 01-ai/Yi-34B-Chat
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+is_mistral_derived_model: false
+is_llama_derived_model: true
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+sequence_len: 1024
+bf16: true
+fp16: false
+tf32: false
+flash_attention: true
+special_tokens:
+  bos_token: "<|startoftext|>"
+  eos_token: "<|endoftext|>"
+  unk_token: "<unk>"
+
+# Data
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+warmup_steps: 10
+
+# Iterations
+num_epochs: 1
+
+# Evaluation
+val_set_size: 0.1
+evals_per_epoch: 5
+eval_table_size:
+eval_table_max_new_tokens: 128
+eval_sample_packing: false
+eval_batch_size: 1
+
+# LoRA
+output_dir: ./qlora-out
+adapter: qlora
+lora_model_dir:
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_fan_in_fan_out:
+lora_target_modules:
+
+# Sampling
+sample_packing: false
+pad_to_sequence_len: false
+
+# Batching
+gradient_accumulation_steps: 4
+micro_batch_size: 1
+gradient_checkpointing: true
+
+# wandb
+wandb_project:
+
+# Optimizer
+optimizer: paged_adamw_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+# Misc
+train_on_inputs: false
+group_by_length: false
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+debug:
+deepspeed:
+weight_decay: 0
+fsdp:
+fsdp_config:

requirements.txt

@@ -1,20 +1,26 @@
-peft @ git+https://github.com/huggingface/peft.git
-transformers @ git+https://github.com/huggingface/transformers.git
+--extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/
+auto-gptq==0.5.1
+packaging
+peft==0.6.0
+transformers @ git+https://github.com/huggingface/transformers.git@3cefac1d974db5e2825a0cb2b842883a628be7a0
+tokenizers==0.15.0
 bitsandbytes>=0.41.1
-accelerate @ git+https://github.com/huggingface/accelerate@2a289f6108e77a77a4efffb3f6316bc98538413b
+accelerate==0.24.1
+deepspeed
 addict
 fire
-PyYAML==6.0
-datasets
-accelerate>=0.19.0
+PyYAML>=6.0
+datasets>=2.15.0
+flash-attn==2.3.3
 sentencepiece
 wandb
 einops
-xformers
-optimum
+xformers==0.0.22
+optimum==1.13.2
 hf_transfer
+colorama
 numba
-numpy==1.24.4
+numpy>=1.24.4
 # qlora things
 bert-score==0.3.13
 evaluate==0.4.0
@@ -22,3 +28,14 @@ rouge-score==0.1.2
 scipy
 scikit-learn==1.2.2
 pynvml
+art
+fschat==0.2.34
+gradio==3.50.2
+tensorboard
+
+# remote filesystems
+s3fs
+gcsfs
+# adlfs
+
+trl @ git+https://github.com/huggingface/trl.git@main

scripts/alpaca_json_to_jsonl.py

@@ -1,52 +0,0 @@
-"""Module to convert json file to jsonl"""
-
-import os
-import sys
-from pathlib import Path
-from typing import Optional, Union
-
-import fire
-
-from axolotl.convert import (
-    FileReader,
-    FileWriter,
-    JsonlSerializer,
-    JsonParser,
-    JsonToJsonlConverter,
-    StdoutWriter,
-)
-from axolotl.logging_config import configure_logging
-
-configure_logging()
-
-# add src to the pythonpath so we don't need to pip install this
-project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
-src_dir = os.path.join(project_root, "src")
-sys.path.insert(0, src_dir)
-
-
-def main(
-    file: Path,
-    output: Optional[Path] = None,
-    to_stdout: Optional[bool] = False,
-):
-    """
-    Convert a json file to jsonl
-    """
-
-    file_reader = FileReader()
-    writer: Union[StdoutWriter, FileWriter]
-    if to_stdout or output is None:
-        writer = StdoutWriter()
-    else:
-        writer = FileWriter(output)
-    json_parser = JsonParser()
-    jsonl_serializer = JsonlSerializer()
-
-    converter = JsonToJsonlConverter(file_reader, writer, json_parser, jsonl_serializer)
-
-    converter.convert(file, output)
-
-
-if __name__ == "__main__":
-    fire.Fire(main)

scripts/finetune.py

@@ -1,315 +1,52 @@
 """Prepare and train a model on a dataset. Can also infer from a model or merge lora"""
-
-import importlib
 import logging
-import os
-import random
-import signal
-import sys
 from pathlib import Path
-from typing import Any, Dict, List, Optional, Union
 
 import fire
-import torch
-import yaml
+import transformers
 
-# add src to the pythonpath so we don't need to pip install this
-from optimum.bettertransformer import BetterTransformer
-from transformers import GenerationConfig, TextStreamer
+from axolotl.cli import (
+    check_accelerate_default_config,
+    check_user_token,
+    do_inference,
+    do_merge_lora,
+    load_cfg,
+    load_datasets,
+    print_axolotl_text_art,
+)
+from axolotl.cli.shard import shard
+from axolotl.common.cli import TrainerCliArgs
+from axolotl.train import train
 
-from axolotl.logging_config import configure_logging
-from axolotl.utils.config import normalize_config, validate_config
-from axolotl.utils.data import prepare_dataset
-from axolotl.utils.dict import DictDefault
-from axolotl.utils.distributed import is_main_process
-from axolotl.utils.models import load_model, load_tokenizer
-from axolotl.utils.tokenization import check_dataset_labels
-from axolotl.utils.trainer import setup_trainer
-from axolotl.utils.wandb import setup_wandb_env_vars
-
-project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
-src_dir = os.path.join(project_root, "src")
-sys.path.insert(0, src_dir)
-
-configure_logging()
-LOG = logging.getLogger("axolotl.scripts")
-
-os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
+LOG = logging.getLogger("axolotl.scripts.finetune")
 
 
-def print_axolotl_text_art():
-    ascii_art = """
-                           dP            dP   dP
-                           88            88   88
-.d8888b. dP.  .dP .d8888b. 88 .d8888b. d8888P 88
-88'  `88  `8bd8'  88'  `88 88 88'  `88   88   88
-88.  .88  .d88b.  88.  .88 88 88.  .88   88   88
-`88888P8 dP'  `dP `88888P' dP `88888P'   dP   dP
-"""
-
-    if is_main_process():
-        print(ascii_art)
-
-
-def get_multi_line_input() -> Optional[str]:
-    print("Give me an instruction (Ctrl + D to finish): ")
-    instruction = ""
-    for line in sys.stdin:
-        instruction += line  # pylint: disable=consider-using-join
-    # instruction = pathlib.Path("/proc/self/fd/0").read_text()
-    return instruction
-
-
-def do_inference(cfg, model, tokenizer, prompter: Optional[str]):
-    default_tokens = {"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>"}
-
-    for token, symbol in default_tokens.items():
-        # If the token isn't already specified in the config, add it
-        if not (cfg.special_tokens and token in cfg.special_tokens):
-            tokenizer.add_special_tokens({token: symbol})
-
-    prompter_module = None
-    if prompter:
-        prompter_module = getattr(
-            importlib.import_module("axolotl.prompters"), prompter
-        )
-
-    if cfg.landmark_attention:
-        from axolotl.monkeypatch.llama_landmark_attn import set_model_mem_id
-
-        set_model_mem_id(model, tokenizer)
-        model.set_mem_cache_args(
-            max_seq_len=255, mem_freq=50, top_k=5, max_cache_size=None
-        )
-
-    while True:
-        print("=" * 80)
-        # support for multiline inputs
-        instruction = get_multi_line_input()
-        if not instruction:
-            return
-        if prompter_module:
-            prompt: str = next(
-                prompter_module().build_prompt(instruction=instruction.strip("\n"))
-            )
-        else:
-            prompt = instruction.strip()
-        batch = tokenizer(prompt, return_tensors="pt", add_special_tokens=True)
-
-        print("=" * 40)
-        model.eval()
-        with torch.no_grad():
-            generation_config = GenerationConfig(
-                repetition_penalty=1.1,
-                max_new_tokens=1024,
-                temperature=0.9,
-                top_p=0.95,
-                top_k=40,
-                bos_token_id=tokenizer.bos_token_id,
-                eos_token_id=tokenizer.eos_token_id,
-                pad_token_id=tokenizer.pad_token_id,
-                do_sample=True,
-                use_cache=True,
-                return_dict_in_generate=True,
-                output_attentions=False,
-                output_hidden_states=False,
-                output_scores=False,
-            )
-            streamer = TextStreamer(tokenizer)
-            generated = model.generate(
-                inputs=batch["input_ids"].to(cfg.device),
-                generation_config=generation_config,
-                streamer=streamer,
-            )
-        print("=" * 40)
-        print(tokenizer.decode(generated["sequences"].cpu().tolist()[0]))
-
-
-def choose_config(path: Path):
-    yaml_files = list(path.glob("*.yml"))
-
-    if not yaml_files:
-        raise ValueError(
-            "No YAML config files found in the specified directory. Are you using a .yml extension?"
-        )
-
-    print("Choose a YAML file:")
-    for idx, file in enumerate(yaml_files):
-        print(f"{idx + 1}. {file}")
-
-    chosen_file = None
-    while chosen_file is None:
-        try:
-            choice = int(input("Enter the number of your choice: "))
-            if 1 <= choice <= len(yaml_files):
-                chosen_file = yaml_files[choice - 1]
-            else:
-                print("Invalid choice. Please choose a number from the list.")
-        except ValueError:
-            print("Invalid input. Please enter a number.")
-
-    return chosen_file
-
-
-def check_not_in(list1: List[str], list2: Union[Dict[str, Any], List[str]]) -> bool:
-    return not any(el in list2 for el in list1)
-
-
-def train(
-    config: Path = Path("configs/"),
-    prepare_ds_only: bool = False,
-    **kwargs,
-):
+def do_cli(config: Path = Path("examples/"), **kwargs):
     print_axolotl_text_art()
-    if Path(config).is_dir():
-        config = choose_config(config)
-
-    # load the config from the yaml file
-    with open(config, encoding="utf-8") as file:
-        cfg: DictDefault = DictDefault(yaml.safe_load(file))
-    # if there are any options passed in the cli, if it is something that seems valid from the yaml,
-    # then overwrite the value
-    cfg_keys = cfg.keys()
-    for k, _ in kwargs.items():
-        # if not strict, allow writing to cfg even if it's not in the yml already
-        if k in cfg_keys or not cfg.strict:
-            # handle booleans
-            if isinstance(cfg[k], bool):
-                cfg[k] = bool(kwargs[k])
-            else:
-                cfg[k] = kwargs[k]
-
-    validate_config(cfg)
-
-    normalize_config(cfg)
-
-    setup_wandb_env_vars(cfg)
-
-    # load the tokenizer first
-    LOG.info(f"loading tokenizer... {cfg.tokenizer_config or cfg.base_model_config}")
-    tokenizer = load_tokenizer(cfg)
-
-    if (
-        check_not_in(["shard", "merge_lora"], kwargs) and not cfg.inference
-    ):  # don't need to load dataset for these
-        train_dataset, eval_dataset, total_num_steps = prepare_dataset(cfg, tokenizer)
-
-    if cfg.debug or "debug" in kwargs:
-        LOG.info("check_dataset_labels...")
-        check_dataset_labels(
-            train_dataset.select(
-                [random.randrange(0, len(train_dataset) - 1) for _ in range(5)]  # nosec
-            ),
-            tokenizer,
-        )
-
-    if prepare_ds_only:
-        LOG.info("Finished preparing dataset. Exiting...")
-        return
-
-    # Load the model and tokenizer
-    LOG.info("loading model and (optionally) peft_config...")
-    model, peft_config = load_model(cfg, tokenizer)
-
-    safe_serialization = cfg.save_safetensors is True
-
-    if "merge_lora" in kwargs and cfg.adapter is not None:
-        LOG.info("running merge of LoRA with base model")
-        model = model.merge_and_unload()
-        model.to(dtype=torch.float16)
-
-        if cfg.local_rank == 0:
-            LOG.info("saving merged model")
-            model.save_pretrained(
-                str(Path(cfg.output_dir) / "merged"),
-                safe_serialization=safe_serialization,
+    LOG.warning(
+        str(
+            PendingDeprecationWarning(
+                "scripts/finetune.py will be replaced with calling axolotl.cli.train"
             )
-            tokenizer.save_pretrained(str(Path(cfg.output_dir) / "merged"))
-        return
-
-    if cfg.inference:
-        LOG.info("calling do_inference function")
-        prompter: Optional[str] = "AlpacaPrompter"
-        if "prompter" in kwargs:
-            if kwargs["prompter"] == "None":
-                prompter = None
-            else:
-                prompter = kwargs["prompter"]
-        do_inference(cfg, model, tokenizer, prompter=prompter)
-        return
-
-    if "shard" in kwargs:
-        model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)
-        return
-
-    trainer = setup_trainer(
-        cfg, train_dataset, eval_dataset, model, tokenizer, total_num_steps
+        )
     )
-
-    model.config.use_cache = False
-
-    if torch.__version__ >= "2" and sys.platform != "win32":
-        LOG.info("Compiling torch model")
-        model = torch.compile(model)
-
-    # go ahead and presave, so we have the adapter config available to inspect
-    if peft_config:
-        LOG.info(f"Pre-saving adapter config to {cfg.output_dir}")
-        peft_config.save_pretrained(cfg.output_dir)
-
-    # In case we want to stop early with ctrl+c, this is a nice to have to save the pretrained model
-    if cfg.local_rank == 0:
-
-        def terminate_handler(_, __, model):
-            if cfg.flash_optimum:
-                model = BetterTransformer.reverse(model)
-            model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)
-            sys.exit(0)
-
-        signal.signal(
-            signal.SIGINT, lambda signum, frame: terminate_handler(signum, frame, model)
-        )
-
-    LOG.info("Starting trainer...")
-    if cfg.group_by_length:
-        LOG.info("hang tight... sorting dataset for group_by_length")
-    resume_from_checkpoint = cfg.resume_from_checkpoint
-    if cfg.resume_from_checkpoint is None and cfg.auto_resume_from_checkpoints:
-        possible_checkpoints = [
-            str(cp) for cp in Path(cfg.output_dir).glob("checkpoint-*")
-        ]
-        if len(possible_checkpoints) > 0:
-            sorted_paths = sorted(
-                possible_checkpoints,
-                key=lambda path: int(path.split("-")[-1]),
-            )
-            resume_from_checkpoint = sorted_paths[-1]
-            LOG.info(
-                f"Using Auto-resume functionality to start with checkpoint at {resume_from_checkpoint}"
-            )
-
-    if not Path(cfg.output_dir).is_dir():
-        os.makedirs(cfg.output_dir, exist_ok=True)
-    tokenizer.save_pretrained(cfg.output_dir)
-    if cfg.flash_optimum:
-        with torch.backends.cuda.sdp_kernel(
-            enable_flash=True, enable_math=True, enable_mem_efficient=True
-        ):
-            trainer.train(resume_from_checkpoint=resume_from_checkpoint)
+    parsed_cfg = load_cfg(config, **kwargs)
+    check_accelerate_default_config()
+    check_user_token()
+    parser = transformers.HfArgumentParser((TrainerCliArgs))
+    parsed_cli_args, _ = parser.parse_args_into_dataclasses(
+        return_remaining_strings=True
+    )
+    if parsed_cli_args.inference:
+        do_inference(cfg=parsed_cfg, cli_args=parsed_cli_args)
+    elif parsed_cli_args.merge_lora:
+        do_merge_lora(cfg=parsed_cfg, cli_args=parsed_cli_args)
+    elif parsed_cli_args.shard:
+        shard(cfg=parsed_cfg, cli_args=parsed_cli_args)
     else:
-        trainer.train(resume_from_checkpoint=resume_from_checkpoint)
-
-    LOG.info(f"Training Completed!!! Saving pre-trained model to {cfg.output_dir}")
-
-    # TODO do we need this fix? https://huggingface.co/docs/accelerate/usage_guides/fsdp#saving-and-loading
-    # only save on rank 0, otherwise it corrupts output on multi-GPU when multiple processes attempt to write the same file
-    if cfg.fsdp:
-        trainer.save_model(cfg.output_dir)
-    elif cfg.local_rank == 0:
-        if cfg.flash_optimum:
-            model = BetterTransformer.reverse(model)
-        model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)
+        dataset_meta = load_datasets(cfg=parsed_cfg, cli_args=parsed_cli_args)
+        train(cfg=parsed_cfg, cli_args=parsed_cli_args, dataset_meta=dataset_meta)
 
 
 if __name__ == "__main__":
-    fire.Fire(train)
+    fire.Fire(do_cli)

setup.py

@@ -1,33 +1,61 @@
 """setup.py for axolotl"""
 
+from importlib.metadata import PackageNotFoundError, version
+
 from setuptools import find_packages, setup
 
-install_requires = []
-with open("./requirements.txt", encoding="utf-8") as requirements_file:
-    # don't include peft yet until we check the int4
-    # need to manually install peft for now...
-    reqs = [r.strip() for r in requirements_file.readlines() if "peft" not in r]
-    reqs = [r for r in reqs if r and r[0] != "#"]
-    for r in reqs:
-        install_requires.append(r)
+
+def parse_requirements():
+    _install_requires = []
+    _dependency_links = []
+    with open("./requirements.txt", encoding="utf-8") as requirements_file:
+        lines = [r.strip() for r in requirements_file.readlines()]
+        for line in lines:
+            if line.startswith("--extra-index-url"):
+                # Handle custom index URLs
+                _, url = line.split()
+                _dependency_links.append(url)
+            elif (
+                "flash-attn" not in line
+                and "deepspeed" not in line
+                and line
+                and line[0] != "#"
+            ):
+                # Handle standard packages
+                _install_requires.append(line)
+
+    try:
+        torch_version = version("torch")
+        if torch_version.startswith("2.1.1"):
+            _install_requires.pop(_install_requires.index("xformers==0.0.22"))
+            _install_requires.append("xformers==0.0.23")
+    except PackageNotFoundError:
+        pass
+
+    return _install_requires, _dependency_links
+
+
+install_requires, dependency_links = parse_requirements()
+
 
 setup(
     name="axolotl",
-    version="0.1",
-    description="You know you're going to axolotl questions",
+    version="0.3.0",
+    description="LLM Trainer",
+    long_description="Axolotl is a tool designed to streamline the fine-tuning of various AI models, offering support for multiple configurations and architectures.",
     package_dir={"": "src"},
     packages=find_packages(),
     install_requires=install_requires,
+    dependency_links=dependency_links,
     extras_require={
-        "gptq": [
-            "alpaca_lora_4bit @ git+https://github.com/winglian/alpaca_lora_4bit.git@setup_pip",
+        "flash-attn": [
+            "flash-attn==2.3.3",
         ],
-        "gptq_triton": [
-            "alpaca_lora_4bit[triton] @ git+https://github.com/winglian/alpaca_lora_4bit.git@setup_pip",
-        ],
-        "extras": [
-            "flash-attn",
+        "deepspeed": [
             "deepspeed",
         ],
+        "mamba-ssm": [
+            "mamba-ssm==1.0.1",
+        ],
     },
 )

src/axolotl/cli/__init__.py

@@ -0,0 +1,436 @@
+"""Prepare and train a model on a dataset. Can also infer from a model or merge lora"""
+
+import importlib
+import logging
+import math
+import os
+import random
+import sys
+from pathlib import Path
+from threading import Thread
+from typing import Any, Dict, List, Optional, Union
+
+import gradio as gr
+import torch
+import yaml
+
+# add src to the pythonpath so we don't need to pip install this
+from accelerate.commands.config import config_args
+from art import text2art
+from datasets import concatenate_datasets, load_dataset
+from huggingface_hub import HfApi
+from huggingface_hub.utils import LocalTokenNotFoundError
+from transformers import GenerationConfig, TextIteratorStreamer, TextStreamer
+
+from axolotl.common.cli import TrainerCliArgs, load_model_and_tokenizer
+from axolotl.logging_config import configure_logging
+from axolotl.train import TrainDatasetMeta
+from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.data import prepare_dataset
+from axolotl.utils.dict import DictDefault
+from axolotl.utils.distributed import is_main_process
+from axolotl.utils.models import load_tokenizer
+from axolotl.utils.tokenization import check_dataset_labels
+from axolotl.utils.trainer import prepare_optim_env
+from axolotl.utils.wandb_ import setup_wandb_env_vars
+
+project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
+src_dir = os.path.join(project_root, "src")
+sys.path.insert(0, src_dir)
+
+configure_logging()
+LOG = logging.getLogger("axolotl.scripts")
+
+os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
+
+
+def print_axolotl_text_art(suffix=None):
+    font = "nancyj"
+    ascii_text = "  axolotl"
+    if suffix:
+        ascii_text += f"  x  {suffix}"
+    ascii_art = text2art(ascii_text, font=font)
+
+    if is_main_process():
+        print(ascii_art)
+
+
+def get_multi_line_input() -> Optional[str]:
+    print("Give me an instruction (Ctrl + D to submit): ")
+    instruction = ""
+    for line in sys.stdin:
+        instruction += line  # pylint: disable=consider-using-join
+    # instruction = pathlib.Path("/proc/self/fd/0").read_text()
+    return instruction
+
+
+def do_merge_lora(
+    *,
+    cfg: DictDefault,
+    cli_args: TrainerCliArgs,
+):
+    model, tokenizer = load_model_and_tokenizer(cfg=cfg, cli_args=cli_args)
+    safe_serialization = cfg.save_safetensors is True
+
+    LOG.info("running merge of LoRA with base model")
+    model = model.merge_and_unload(progressbar=True)
+    model.to(dtype=cfg.torch_dtype)
+
+    if cfg.local_rank == 0:
+        LOG.info(f"saving merged model to: {str(Path(cfg.output_dir) / 'merged')}")
+        model.save_pretrained(
+            str(Path(cfg.output_dir) / "merged"),
+            safe_serialization=safe_serialization,
+            progressbar=True,
+        )
+        tokenizer.save_pretrained(str(Path(cfg.output_dir) / "merged"))
+
+
+def do_inference(
+    *,
+    cfg: DictDefault,
+    cli_args: TrainerCliArgs,
+):
+    model, tokenizer = load_model_and_tokenizer(cfg=cfg, cli_args=cli_args)
+    prompter = cli_args.prompter
+    default_tokens = {"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>"}
+
+    for token, symbol in default_tokens.items():
+        # If the token isn't already specified in the config, add it
+        if not (cfg.special_tokens and token in cfg.special_tokens):
+            tokenizer.add_special_tokens({token: symbol})
+
+    prompter_module = None
+    if prompter:
+        prompter_module = getattr(
+            importlib.import_module("axolotl.prompters"), prompter
+        )
+
+    model = model.to(cfg.device, dtype=cfg.torch_dtype)
+
+    while True:
+        print("=" * 80)
+        # support for multiline inputs
+        instruction = get_multi_line_input()
+        if not instruction:
+            return
+        if prompter_module:
+            prompt: str = next(
+                prompter_module().build_prompt(instruction=instruction.strip("\n"))
+            )
+        else:
+            prompt = instruction.strip()
+        batch = tokenizer(prompt, return_tensors="pt", add_special_tokens=True)
+
+        print("=" * 40)
+        model.eval()
+        with torch.no_grad():
+            generation_config = GenerationConfig(
+                repetition_penalty=1.1,
+                max_new_tokens=1024,
+                temperature=0.9,
+                top_p=0.95,
+                top_k=40,
+                bos_token_id=tokenizer.bos_token_id,
+                eos_token_id=tokenizer.eos_token_id,
+                pad_token_id=tokenizer.pad_token_id,
+                do_sample=True,
+                use_cache=True,
+                return_dict_in_generate=True,
+                output_attentions=False,
+                output_hidden_states=False,
+                output_scores=False,
+            )
+            streamer = TextStreamer(tokenizer)
+            generated = model.generate(
+                inputs=batch["input_ids"].to(cfg.device),
+                generation_config=generation_config,
+                streamer=streamer,
+            )
+        print("=" * 40)
+        print(tokenizer.decode(generated["sequences"].cpu().tolist()[0]))
+
+
+def do_inference_gradio(
+    *,
+    cfg: DictDefault,
+    cli_args: TrainerCliArgs,
+):
+    model, tokenizer = load_model_and_tokenizer(cfg=cfg, cli_args=cli_args)
+    prompter = cli_args.prompter
+    default_tokens = {"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>"}
+
+    for token, symbol in default_tokens.items():
+        # If the token isn't already specified in the config, add it
+        if not (cfg.special_tokens and token in cfg.special_tokens):
+            tokenizer.add_special_tokens({token: symbol})
+
+    prompter_module = None
+    if prompter:
+        prompter_module = getattr(
+            importlib.import_module("axolotl.prompters"), prompter
+        )
+
+    model = model.to(cfg.device, dtype=cfg.torch_dtype)
+
+    def generate(instruction):
+        if not instruction:
+            return
+        if prompter_module:
+            # pylint: disable=stop-iteration-return
+            prompt: str = next(
+                prompter_module().build_prompt(instruction=instruction.strip("\n"))
+            )
+        else:
+            prompt = instruction.strip()
+        batch = tokenizer(prompt, return_tensors="pt", add_special_tokens=True)
+
+        model.eval()
+        with torch.no_grad():
+            generation_config = GenerationConfig(
+                repetition_penalty=1.1,
+                max_new_tokens=1024,
+                temperature=0.9,
+                top_p=0.95,
+                top_k=40,
+                bos_token_id=tokenizer.bos_token_id,
+                eos_token_id=tokenizer.eos_token_id,
+                pad_token_id=tokenizer.pad_token_id,
+                do_sample=True,
+                use_cache=True,
+                return_dict_in_generate=True,
+                output_attentions=False,
+                output_hidden_states=False,
+                output_scores=False,
+            )
+            streamer = TextIteratorStreamer(tokenizer)
+            generation_kwargs = {
+                "inputs": batch["input_ids"].to(cfg.device),
+                "generation_config": generation_config,
+                "streamer": streamer,
+            }
+
+            thread = Thread(target=model.generate, kwargs=generation_kwargs)
+            thread.start()
+
+            all_text = ""
+
+            for new_text in streamer:
+                all_text += new_text
+                yield all_text
+
+    demo = gr.Interface(
+        fn=generate,
+        inputs="textbox",
+        outputs="text",
+        title=cfg.get("gradio_title", "Axolotl Gradio Interface"),
+    )
+    demo.queue().launch(show_api=False, share=True)
+
+
+def choose_config(path: Path):
+    yaml_files = list(path.glob("*.yml"))
+
+    if not yaml_files:
+        raise ValueError(
+            "No YAML config files found in the specified directory. Are you using a .yml extension?"
+        )
+
+    if len(yaml_files) == 1:
+        print(f"Using default YAML file '{yaml_files[0]}'")
+        return yaml_files[0]
+
+    print("Choose a YAML file:")
+    for idx, file in enumerate(yaml_files):
+        print(f"{idx + 1}. {file}")
+
+    chosen_file = None
+    while chosen_file is None:
+        try:
+            choice = int(input("Enter the number of your choice: "))
+            if 1 <= choice <= len(yaml_files):
+                chosen_file = yaml_files[choice - 1]
+            else:
+                print("Invalid choice. Please choose a number from the list.")
+        except ValueError:
+            print("Invalid input. Please enter a number.")
+
+    return chosen_file
+
+
+def check_not_in(list1: List[str], list2: Union[Dict[str, Any], List[str]]) -> bool:
+    return not any(el in list2 for el in list1)
+
+
+def load_cfg(config: Path = Path("examples/"), **kwargs):
+    if Path(config).is_dir():
+        config = choose_config(config)
+
+    # load the config from the yaml file
+    with open(config, encoding="utf-8") as file:
+        cfg: DictDefault = DictDefault(yaml.safe_load(file))
+    cfg.axolotl_config_path = config
+    # if there are any options passed in the cli, if it is something that seems valid from the yaml,
+    # then overwrite the value
+    cfg_keys = cfg.keys()
+    for k, _ in kwargs.items():
+        # if not strict, allow writing to cfg even if it's not in the yml already
+        if k in cfg_keys or not cfg.strict:
+            # handle booleans
+            if isinstance(cfg[k], bool):
+                cfg[k] = bool(kwargs[k])
+            else:
+                cfg[k] = kwargs[k]
+
+    validate_config(cfg)
+
+    prepare_optim_env(cfg)
+
+    normalize_config(cfg)
+
+    setup_wandb_env_vars(cfg)
+    return cfg
+
+
+def load_datasets(
+    *,
+    cfg: DictDefault,
+    cli_args: TrainerCliArgs,
+) -> TrainDatasetMeta:
+    tokenizer = load_tokenizer(cfg)
+
+    train_dataset, eval_dataset, total_num_steps, prompters = prepare_dataset(
+        cfg, tokenizer
+    )
+
+    if cli_args.debug or cfg.debug:
+        LOG.info("check_dataset_labels...")
+        check_dataset_labels(
+            train_dataset.select(
+                [
+                    random.randrange(0, len(train_dataset) - 1)  # nosec
+                    for _ in range(cli_args.debug_num_examples)
+                ]
+            ),
+            tokenizer,
+            num_examples=cli_args.debug_num_examples,
+            text_only=cli_args.debug_text_only,
+        )
+
+        LOG.info("printing prompters...")
+        for prompter in prompters:
+            LOG.info(prompter)
+
+    return TrainDatasetMeta(
+        train_dataset=train_dataset,
+        eval_dataset=eval_dataset,
+        total_num_steps=total_num_steps,
+    )
+
+
+def load_rl_datasets(
+    *,
+    cfg: DictDefault,
+    cli_args: TrainerCliArgs,  # pylint: disable=unused-argument
+) -> TrainDatasetMeta:
+    train_datasets: List[Any] = []
+    for i, ds_cfg in enumerate(cfg.datasets):
+        train_datasets.insert(i, load_dataset(ds_cfg["path"], split=ds_cfg["split"]))
+    # eval_dataset = load_dataset(
+    #     cfg.test_datasets[0]["path"], split=cfg.test_datasets[0]["split"]
+    # )
+    eval_dataset = None
+
+    def argilla_apply_chatml(sample):  # pylint: disable=possibly-unused-variable
+        if "system" in sample and sample["system"]:
+            sample["prompt"] = (
+                f"<|im_start|>system\n{sample['system']}<|im_end|>\n"
+                f"<|im_start|>user\n{sample['instruction']}<|im_end|>\n<|im_start|>assistant\n"
+            )
+        else:
+            sample[
+                "prompt"
+            ] = f"<|im_start|>user\n{sample['instruction']}<|im_end|>\n<|im_start|>assistant\n"
+        sample["chosen"] = f"{sample['chosen_response']}<|im_end|>"
+        sample["rejected"] = f"{sample['rejected_response']}<|im_end|>"
+        return sample
+
+    def intel_apply_chatml(sample):  # pylint: disable=possibly-unused-variable
+        if "system" in sample and sample["system"]:
+            sample["prompt"] = (
+                f"<|im_start|>system\n{sample['system']}<|im_end|>\n"
+                f"<|im_start|>user\n{sample['question']}<|im_end|>\n<|im_start|>assistant\n"
+            )
+        else:
+            sample[
+                "prompt"
+            ] = f"<|im_start|>user\n{sample['question']}<|im_end|>\n<|im_start|>assistant\n"
+        sample["chosen"] = f"{sample['chosen']}<|im_end|>"
+        sample["rejected"] = f"{sample['rejected']}<|im_end|>"
+        return sample
+
+    def apply_chatml(sample):  # pylint: disable=possibly-unused-variable
+        if "system" in sample and sample["system"]:
+            sample["prompt"] = (
+                f"<|im_start|>system\n{sample['system']}<|im_end|>\n"
+                f"<|im_start|>user\n{sample['prompt']}<|im_end|>\n<|im_start|>assistant\n"
+            )
+        else:
+            sample[
+                "prompt"
+            ] = f"<|im_start|>user\n{sample['prompt']}<|im_end|>\n<|im_start|>assistant\n"
+        sample["chosen"] = f"{sample['chosen']}<|im_end|>"
+        sample["rejected"] = f"{sample['rejected']}<|im_end|>"
+        return sample
+
+    def ultra_apply_chatml(sample):  # pylint: disable=possibly-unused-variable
+        if "system" in sample and sample["system"]:
+            sample["prompt"] = (
+                f"<|im_start|>system\n{sample['system']}<|im_end|>\n"
+                f"<|im_start|>user\n{sample['prompt']}<|im_end|>\n<|im_start|>assistant\n"
+            )
+        else:
+            sample[
+                "prompt"
+            ] = f"<|im_start|>user\n{sample['prompt']}<|im_end|>\n<|im_start|>assistant\n"
+        sample["chosen"] = f"{sample['chosen'][1]['content']}<|im_end|>"
+        sample["rejected"] = f"{sample['rejected'][1]['content']}<|im_end|>"
+        return sample
+
+    for i, data_set in enumerate(train_datasets):
+        _type = cfg.datasets[i]["type"]
+        ds_type_fn = locals()[_type]
+        train_datasets[i] = data_set.map(ds_type_fn)
+    train_dataset = concatenate_datasets(train_datasets)
+
+    # eval_dataset = eval_dataset.map(intel_apply_chatml)
+
+    total_num_steps = int(
+        math.ceil(len(train_dataset) * cfg.num_epochs / cfg.batch_size)
+    )
+
+    return TrainDatasetMeta(
+        train_dataset=train_dataset,
+        eval_dataset=eval_dataset,
+        total_num_steps=total_num_steps,
+    )
+
+
+def check_accelerate_default_config():
+    if Path(config_args.default_yaml_config_file).exists():
+        LOG.warning(
+            f"accelerate config file found at {config_args.default_yaml_config_file}. This can lead to unexpected errors"
+        )
+
+
+def check_user_token():
+    # Verify if token is valid
+    api = HfApi()
+    try:
+        user_info = api.whoami()
+        return bool(user_info)
+    except LocalTokenNotFoundError:
+        LOG.warning(
+            "Error verifying HuggingFace token. Remember to log in using `huggingface-cli login` and get your access token from https://huggingface.co/settings/tokens if you want to use gated models or datasets."
+        )
+        return False

src/axolotl/cli/inference.py

@@ -0,0 +1,36 @@
+"""
+CLI to run inference on a trained model
+"""
+from pathlib import Path
+
+import fire
+import transformers
+
+from axolotl.cli import (
+    do_inference,
+    do_inference_gradio,
+    load_cfg,
+    print_axolotl_text_art,
+)
+from axolotl.common.cli import TrainerCliArgs
+
+
+def do_cli(config: Path = Path("examples/"), gradio=False, **kwargs):
+    # pylint: disable=duplicate-code
+    print_axolotl_text_art()
+    parsed_cfg = load_cfg(config, **kwargs)
+    parsed_cfg.sample_packing = False
+    parser = transformers.HfArgumentParser((TrainerCliArgs))
+    parsed_cli_args, _ = parser.parse_args_into_dataclasses(
+        return_remaining_strings=True
+    )
+    parsed_cli_args.inference = True
+
+    if gradio:
+        do_inference_gradio(cfg=parsed_cfg, cli_args=parsed_cli_args)
+    else:
+        do_inference(cfg=parsed_cfg, cli_args=parsed_cli_args)
+
+
+if __name__ == "__main__":
+    fire.Fire(do_cli)

src/axolotl/cli/merge_lora.py

@@ -0,0 +1,42 @@
+"""
+CLI to run merge a trained LoRA into a base model
+"""
+from pathlib import Path
+
+import fire
+import transformers
+
+from axolotl.cli import do_merge_lora, load_cfg, print_axolotl_text_art
+from axolotl.common.cli import TrainerCliArgs
+
+
+def do_cli(config: Path = Path("examples/"), **kwargs):
+    # pylint: disable=duplicate-code
+    print_axolotl_text_art()
+    parser = transformers.HfArgumentParser((TrainerCliArgs))
+    parsed_cli_args, _ = parser.parse_args_into_dataclasses(
+        return_remaining_strings=True
+    )
+    parsed_cli_args.merge_lora = True
+
+    parsed_cfg = load_cfg(
+        config,
+        merge_lora=True,
+        load_in_8bit=False,
+        load_in_4bit=False,
+        flash_attention=False,
+        **kwargs,
+    )
+
+    if not parsed_cfg.lora_model_dir and parsed_cfg.output_dir:
+        parsed_cfg.lora_model_dir = parsed_cfg.output_dir
+    if not Path(parsed_cfg.lora_model_dir).exists():
+        raise ValueError(
+            f"Target directory for merge: `{parsed_cfg.lora_model_dir}` does not exist."
+        )
+
+    do_merge_lora(cfg=parsed_cfg, cli_args=parsed_cli_args)
+
+
+if __name__ == "__main__":
+    fire.Fire(do_cli)

src/axolotl/cli/preprocess.py

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

src/axolotl/cli/shard.py

@@ -0,0 +1,42 @@
+"""
+CLI to shard a trained model into 10GiB chunks
+"""
+import logging
+from pathlib import Path
+
+import fire
+import transformers
+
+from axolotl.cli import load_cfg, print_axolotl_text_art
+from axolotl.common.cli import TrainerCliArgs, load_model_and_tokenizer
+from axolotl.utils.dict import DictDefault
+
+LOG = logging.getLogger("axolotl.scripts")
+
+
+def shard(
+    *,
+    cfg: DictDefault,
+    cli_args: TrainerCliArgs,
+):
+    model, _ = load_model_and_tokenizer(cfg=cfg, cli_args=cli_args)
+    safe_serialization = cfg.save_safetensors is True
+    LOG.debug("Re-saving model w/ sharding")
+    model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)
+
+
+def do_cli(config: Path = Path("examples/"), **kwargs):
+    # pylint: disable=duplicate-code
+    print_axolotl_text_art()
+    parsed_cfg = load_cfg(config, **kwargs)
+    parser = transformers.HfArgumentParser((TrainerCliArgs))
+    parsed_cli_args, _ = parser.parse_args_into_dataclasses(
+        return_remaining_strings=True
+    )
+    parsed_cli_args.shard = True
+
+    shard(cfg=parsed_cfg, cli_args=parsed_cli_args)
+
+
+if __name__ == "__main__":
+    fire.Fire(do_cli)

src/axolotl/cli/train.py

@@ -0,0 +1,42 @@
+"""
+CLI to run training on a model
+"""
+import logging
+from pathlib import Path
+
+import fire
+import transformers
+
+from axolotl.cli import (
+    check_accelerate_default_config,
+    check_user_token,
+    load_cfg,
+    load_datasets,
+    load_rl_datasets,
+    print_axolotl_text_art,
+)
+from axolotl.common.cli import TrainerCliArgs
+from axolotl.train import train
+
+LOG = logging.getLogger("axolotl.cli.train")
+
+
+def do_cli(config: Path = Path("examples/"), **kwargs):
+    # pylint: disable=duplicate-code
+    parsed_cfg = load_cfg(config, **kwargs)
+    print_axolotl_text_art()
+    check_accelerate_default_config()
+    check_user_token()
+    parser = transformers.HfArgumentParser((TrainerCliArgs))
+    parsed_cli_args, _ = parser.parse_args_into_dataclasses(
+        return_remaining_strings=True
+    )
+    if parsed_cfg.rl:
+        dataset_meta = load_rl_datasets(cfg=parsed_cfg, cli_args=parsed_cli_args)
+    else:
+        dataset_meta = load_datasets(cfg=parsed_cfg, cli_args=parsed_cli_args)
+    train(cfg=parsed_cfg, cli_args=parsed_cli_args, dataset_meta=dataset_meta)
+
+
+if __name__ == "__main__":
+    fire.Fire(do_cli)

src/axolotl/common/cli.py

@@ -0,0 +1,54 @@
+"""
+shared module for cli specific things
+"""
+
+import logging
+from dataclasses import dataclass, field
+from typing import Optional
+
+from axolotl.logging_config import configure_logging
+from axolotl.utils.dict import DictDefault
+from axolotl.utils.models import load_model, load_tokenizer
+
+configure_logging()
+LOG = logging.getLogger("axolotl.common.cli")
+
+
+@dataclass
+class TrainerCliArgs:
+    """
+    dataclass representing the various non-training arguments
+    """
+
+    debug: bool = field(default=False)
+    debug_text_only: bool = field(default=False)
+    debug_num_examples: int = field(default=5)
+    inference: bool = field(default=False)
+    merge_lora: bool = field(default=False)
+    prompter: Optional[str] = field(default=None)
+    shard: bool = field(default=False)
+
+
+@dataclass
+class PreprocessCliArgs:
+    """
+    dataclass representing arguments for preprocessing only
+    """
+
+    debug: bool = field(default=False)
+    debug_text_only: bool = field(default=False)
+    debug_num_examples: int = field(default=1)
+    prompter: Optional[str] = field(default=None)
+
+
+def load_model_and_tokenizer(
+    *,
+    cfg: DictDefault,
+    cli_args: TrainerCliArgs,
+):
+    LOG.info(f"loading tokenizer... {cfg.tokenizer_config or cfg.base_model_config}")
+    tokenizer = load_tokenizer(cfg)
+    LOG.info("loading model and (optionally) peft_config...")
+    model, _ = load_model(cfg, tokenizer, inference=cli_args.inference)
+
+    return model, tokenizer

src/axolotl/common/const.py

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

src/axolotl/core/trainer_builder.py

@@ -0,0 +1,943 @@
+"""
+Builder for the training args and trainer
+"""
+
+import abc
+import importlib
+import logging
+import math
+import sys
+from abc import abstractmethod
+from dataclasses import dataclass, field
+from functools import wraps
+from pathlib import Path
+from typing import Optional
+
+import torch
+import transformers
+from datasets import Dataset
+from torch.optim.lr_scheduler import OneCycleLR
+from torch.utils.data import BatchSampler, DataLoader, RandomSampler, SequentialSampler
+from transformers import EarlyStoppingCallback, Trainer, TrainingArguments
+from transformers.trainer_utils import seed_worker
+from trl import DPOTrainer
+
+from axolotl.monkeypatch.relora import ReLoRACallback, ReLoRAScheduler
+from axolotl.utils.callbacks import (
+    EvalFirstStepCallback,
+    GPUStatsCallback,
+    LossWatchDogCallback,
+    SaveAxolotlConfigtoWandBCallback,
+    SaveBetterTransformerModelCallback,
+    bench_eval_callback_factory,
+    log_prediction_callback_factory,
+)
+from axolotl.utils.collators import (
+    BatchSamplerDataCollatorForSeq2Seq,
+    MambaDataCollator,
+)
+from axolotl.utils.samplers import MultipackBatchSampler
+from axolotl.utils.schedulers import get_cosine_schedule_with_quadratic_warmup
+
+try:
+    import torch._dynamo  # pylint: disable=ungrouped-imports
+except ImportError:
+    pass
+
+LOG = logging.getLogger("axolotl.core.trainer_builder")
+
+
+@dataclass
+class AxolotlTrainingArguments(TrainingArguments):
+    """
+    Extend the base TrainingArguments for axolotl helpers
+    """
+
+    model_type: Optional[str] = field(
+        default=None, metadata={"help": "HF model configuration model_type."}
+    )
+    lr_quadratic_warmup: bool = field(
+        default=False,
+        metadata={"help": "Use quadratic warmup for cosine scheduling."},
+    )
+    pretraining: bool = field(
+        default=False,
+        metadata={
+            "help": "Indicates to trainer whether we are doing continued pretraining."
+        },
+    )
+    sample_packing: bool = field(
+        default=False,
+        metadata={"help": "Use sample packing for efficient training."},
+    )
+    eval_sample_packing: Optional[bool] = field(
+        default=None,
+        metadata={"help": "Use sample packing for efficient evals."},
+    )
+    sample_packing_efficiency: float = field(
+        default=1.0,
+        metadata={"help": "Sample packing efficiency for calculating batch length."},
+    )
+    max_seq_length: int = field(
+        default=2048,
+        metadata={"help": "The maximum sequence length the model can handle"},
+    )
+    sample_packing_seq_len_multiplier: int = field(
+        default=1,
+        metadata={"help": "the multiplier for the max len for packed sequences"},
+    )
+    relora_steps: Optional[int] = field(
+        default=None,
+        metadata={"help": "how often to reset for ReLoRA"},
+    )
+    relora_warmup_steps: Optional[int] = field(
+        default=None,
+        metadata={"help": "how many warmup steps to take after reset for ReLoRA"},
+    )
+    bench_split: Optional[str] = field(
+        default="eval", metadata={"help": "The benchmark split to run on"}
+    )
+    bench_dataset: Optional[str] = field(
+        default="pharaouk/dharma-1/dharma_1_mini.json",
+        metadata={
+            "help": "Benchmark dataset to use: options are `mmlu-zs`, `mmlu-fs`, or the full path to the dataset file"
+        },
+    )
+    do_bench_eval: Optional[bool] = field(
+        default=False, metadata={"help": "Whether to run the Benchmark evaluation."}
+    )
+    max_bench_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": "If set, only evaluates on `max_bench_samples` of the benchmark dataset."
+        },
+    )
+    bench_source_max_len: int = field(
+        default=2048, metadata={"help": "Maximum source sequence length for bench."}
+    )
+    dataloader_prefetch_factor: Optional[int] = field(
+        default=None,
+        metadata={"help": "prefetch_factor argument to the dataloader"},
+    )
+
+
+class AxolotlTrainer(Trainer):
+    """
+    Extend the base Trainer for axolotl helpers
+    """
+
+    args = None  # type: AxolotlTrainingArguments
+    tag_names = ["axolotl"]
+
+    def __init__(self, *args, num_epochs=1, bench_data_collator=None, **kwargs):
+        self.num_epochs = num_epochs
+        self.bench_data_collator = bench_data_collator
+        super().__init__(*args, **kwargs)
+
+    def create_scheduler(
+        self, num_training_steps: int, optimizer: torch.optim.Optimizer = None
+    ):
+        """
+        Setup the scheduler. The optimizer of the trainer must have been set up either before this method is called or
+        passed as an argument.
+
+        Args:
+            num_training_steps (int): The number of training steps to do.
+            optimizer (torch.optim.Optimizer): The training optimizer
+        """
+
+        # fmt: off
+        if self.lr_scheduler is None:  # type: ignore  # pylint: disable=access-member-before-definition
+            # fmt: on
+            if (
+                self.args.lr_scheduler_type == "cosine"
+                and self.args.lr_quadratic_warmup is True
+            ):
+                self.lr_scheduler = get_cosine_schedule_with_quadratic_warmup(  # pylint: disable=attribute-defined-outside-init
+                    optimizer,
+                    num_warmup_steps=self.args.get_warmup_steps(num_training_steps),
+                    num_training_steps=num_training_steps,
+                )
+            else:
+                return super().create_scheduler(num_training_steps, optimizer)
+        return self.lr_scheduler
+
+    def _get_train_sampler(self) -> Optional[torch.utils.data.Sampler]:
+        if self.args.sample_packing and not self.args.pretraining:
+            return MultipackBatchSampler(
+                RandomSampler(self.train_dataset),
+                self.args.train_batch_size,
+                drop_last=True,
+                batch_max_len=self._train_batch_size * self.args.max_seq_length,
+                lengths=(
+                    self.train_dataset.data.column("position_ids")
+                    .to_pandas()
+                    .apply(lambda x: x[-1] + 1)
+                    .values
+                ),
+                packing_efficiency_estimate=self.args.sample_packing_efficiency,
+            )
+        return super()._get_train_sampler()
+
+    def _get_eval_sampler(
+        self, eval_dataset: Dataset
+    ) -> Optional[torch.utils.data.Sampler]:
+        if self.args.sample_packing and self.args.eval_sample_packing is not False:
+            return MultipackBatchSampler(
+                SequentialSampler(eval_dataset),
+                self.args.per_device_eval_batch_size,
+                drop_last=True,
+                batch_max_len=self.args.eval_batch_size * self.args.max_seq_length,
+                lengths=(
+                    eval_dataset.data.column("position_ids")
+                    .to_pandas()
+                    .apply(lambda x: x[-1] + 1)
+                    .values
+                ),
+                packing_efficiency_estimate=self.args.sample_packing_efficiency,
+            )
+        return super()._get_eval_sampler(eval_dataset)
+
+    def get_train_dataloader(self) -> DataLoader:
+        if self.args.sample_packing and not self.args.pretraining:
+            train_dataset = self.train_dataset
+            train_dataset = train_dataset.remove_columns(["length"])
+            data_collator = self.data_collator
+            dataloader_params = {
+                "batch_size": self._train_batch_size,
+                "collate_fn": data_collator,
+                "num_workers": self.args.dataloader_num_workers,
+                "pin_memory": self.args.dataloader_pin_memory,
+            }
+            if self.args.dataloader_prefetch_factor:
+                dataloader_params[
+                    "prefetch_factor"
+                ] = self.args.dataloader_prefetch_factor
+
+            sampler = self._get_train_sampler()
+            if isinstance(sampler, BatchSampler):
+                dataloader_params["batch_sampler"] = sampler
+                del dataloader_params["batch_size"]
+            else:
+                dataloader_params["sampler"] = sampler
+                dataloader_params["drop_last"] = self.args.dataloader_drop_last
+            dataloader_params["worker_init_fn"] = seed_worker
+
+            self.accelerator.even_batches = False
+            return self.accelerator.prepare_data_loader(
+                DataLoader(train_dataset, **dataloader_params)
+            )
+        return super().get_train_dataloader()
+
+    def get_eval_dataloader(self, eval_dataset: Optional[Dataset] = None) -> DataLoader:
+        if self.args.sample_packing and self.args.eval_sample_packing is not False:
+            eval_dataset = (
+                eval_dataset if eval_dataset is not None else self.eval_dataset
+            )
+
+            eval_sampler = self._get_eval_sampler(eval_dataset)
+            eval_dataset = eval_dataset.remove_columns(["length"])
+            data_collator = self.data_collator
+            dataloader_params = {
+                "batch_size": self.args.eval_batch_size,
+                "collate_fn": data_collator,
+                "num_workers": self.args.dataloader_num_workers,
+                "pin_memory": self.args.dataloader_pin_memory,
+            }
+            if self.args.dataloader_prefetch_factor:
+                dataloader_params[
+                    "prefetch_factor"
+                ] = self.args.dataloader_prefetch_factor
+
+            if isinstance(eval_sampler, BatchSampler):
+                dataloader_params["batch_sampler"] = eval_sampler
+                del dataloader_params["batch_size"]
+            else:
+                dataloader_params["sampler"] = eval_sampler
+                dataloader_params["drop_last"] = self.args.dataloader_drop_last
+
+            self.accelerator.even_batches = False
+            return self.accelerator.prepare_data_loader(
+                DataLoader(eval_dataset, **dataloader_params)
+            )
+        return super().get_eval_dataloader(eval_dataset)
+
+    def _get_bench_sampler(
+        self, bench_dataset: Dataset
+    ) -> Optional[torch.utils.data.Sampler]:
+        if self.args.world_size <= 1:
+            return SequentialSampler(bench_dataset)
+        return None
+
+    def get_bench_dataloader(
+        self,
+        bench_dataset: Dataset,
+    ) -> DataLoader:
+        dataloader_params = {
+            "batch_size": self.args.eval_batch_size,
+            "collate_fn": self.bench_data_collator,
+            "num_workers": self.args.dataloader_num_workers,
+            "pin_memory": self.args.dataloader_pin_memory,
+        }
+        if self.args.dataloader_prefetch_factor:
+            dataloader_params["prefetch_factor"] = self.args.dataloader_prefetch_factor
+
+        if not isinstance(bench_dataset, torch.utils.data.IterableDataset):
+            dataloader_params["sampler"] = self._get_bench_sampler(bench_dataset)
+            dataloader_params["drop_last"] = self.args.dataloader_drop_last
+
+        return DataLoader(bench_dataset, **dataloader_params)
+        # return self.accelerator.prepare(DataLoader(bench_dataset, **dataloader_params))
+
+    def compute_loss(self, model, inputs, return_outputs=False):
+        # use one's weighted cross entropy loss calc
+        # if self.args.sample_packing:
+        #     labels = inputs.pop("labels")
+        #     outputs = model(**inputs)
+        #     loss = trainer_weighted_loss(outputs, labels, shift_labels=True)
+        #     return (loss, outputs) if return_outputs else loss
+        return super().compute_loss(model, inputs, return_outputs=return_outputs)
+
+    def _sanitize_kwargs_for_tagging(self, tag_names, kwargs=None):
+        if isinstance(tag_names, str):
+            tag_names = [tag_names]
+
+        if kwargs is not None:
+            if "tags" not in kwargs:
+                kwargs["tags"] = tag_names
+            elif "tags" in kwargs and isinstance(kwargs["tags"], list):
+                kwargs["tags"].extend(tag_names)
+            elif "tags" in kwargs and isinstance(kwargs["tags"], str):
+                tag_names.append(kwargs["tags"])
+                kwargs["tags"] = tag_names
+
+        return kwargs
+
+    @wraps(Trainer.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 = self._sanitize_kwargs_for_tagging(
+            tag_names=self.tag_names, kwargs=kwargs
+        )
+
+        return super().push_to_hub(*args, **kwargs)
+
+
+class AxolotlMambaTrainer(AxolotlTrainer):
+    """
+    Mamba specific trainer to handle loss calculation
+    """
+
+    tag_names = ["axolotl", "mamba"]
+
+    def compute_loss(
+        self,
+        model,
+        inputs,
+        return_outputs=False,  # pylint: disable=unused-argument
+    ):
+        input_ids = inputs.pop("input_ids")
+        lm_logits = model(input_ids).logits
+
+        labels = input_ids.to(lm_logits.device)
+        shift_logits = lm_logits[:, :-1, :].contiguous()
+        labels = labels[:, 1:].contiguous()
+
+        loss_fct = torch.nn.CrossEntropyLoss()
+        lm_loss = loss_fct(
+            shift_logits.view(-1, shift_logits.size(-1)), labels.view(-1)
+        )
+
+        return lm_loss
+
+
+class OneCycleLRSchedulerTrainer(AxolotlTrainer):
+    """
+    Trainer subclass that uses the OneCycleLR scheduler
+    """
+
+    tag_names = ["axolotl", "onecycle"]
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.lr_scheduler = None
+
+    def create_scheduler(
+        self,
+        num_training_steps: int,
+        optimizer: Optional[torch.optim.Optimizer] = None,
+    ):
+        optimizer = self.optimizer if optimizer is None else optimizer
+        num_warmup_steps = self.args.get_warmup_steps(num_training_steps)
+        pct_start = num_warmup_steps / num_training_steps
+
+        self.lr_scheduler = OneCycleLR(
+            optimizer,
+            max_lr=self.args.learning_rate,
+            total_steps=num_training_steps,
+            pct_start=pct_start,
+            div_factor=6,
+        )
+
+        return self.lr_scheduler
+
+
+class ReLoRATrainer(AxolotlTrainer):
+    """
+    Trainer subclass that uses the OneCycleLR scheduler
+    """
+
+    tag_names = ["axolotl", "relora"]
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.lr_scheduler = None
+
+    def create_scheduler(
+        self,
+        num_training_steps: int,
+        optimizer: Optional[torch.optim.Optimizer] = None,
+    ):
+        optimizer = self.optimizer if optimizer is None else optimizer
+        lr_scheduler = super().create_scheduler(num_training_steps, optimizer)
+
+        if self.args.relora_steps:
+            warmup_steps = (
+                self.args.relora_warmup_steps if self.args.relora_warmup_steps else 10
+            )
+            self.lr_scheduler = ReLoRAScheduler(
+                optimizer,
+                lr_scheduler,
+                self.args.relora_steps,
+                warmup_steps,
+            )
+        else:
+            self.lr_scheduler = lr_scheduler
+
+        return self.lr_scheduler
+
+
+class TrainerBuilderBase(abc.ABC):
+    """
+    Base class for trainer builder
+    """
+
+    _train_dataset = None
+    _eval_dataset = None
+    _model_ref = None
+
+    def __init__(self, cfg, model, tokenizer):
+        self.cfg = cfg
+        self.model = model
+        self.tokenizer = tokenizer
+
+    @property
+    def model_ref(self):
+        return self._model_ref
+
+    @model_ref.setter
+    def model_ref(self, model):
+        self._model_ref = model
+
+    @property
+    def train_dataset(self):
+        return self._train_dataset
+
+    @train_dataset.setter
+    def train_dataset(self, dataset):
+        self._train_dataset = dataset
+
+    @property
+    def eval_dataset(self):
+        return self._eval_dataset
+
+    @eval_dataset.setter
+    def eval_dataset(self, dataset):
+        self._eval_dataset = dataset
+
+    @abstractmethod
+    def build(self, total_num_steps):
+        pass
+
+    @abstractmethod
+    def get_callbacks(self):
+        pass
+
+    @abstractmethod
+    def get_post_trainer_create_callbacks(self, trainer):
+        """
+        Callbacks added after the trainer is created, usually b/c these need access to the trainer
+        """
+
+
+class HFCausalTrainerBuilder(TrainerBuilderBase):
+    """
+    Build the HuggingFace training args/trainer for Causal models
+    """
+
+    def hook_pre_create_training_args(self, training_arguments_kwargs):
+        # TODO
+        return training_arguments_kwargs
+
+    def hook_post_create_training_args(self, training_arguments):
+        # TODO
+        return training_arguments
+
+    def hook_pre_create_trainer(self, trainer_kwargs, trainer_cls):
+        # TODO
+        return trainer_kwargs, trainer_cls
+
+    def hook_post_create_trainer(self, trainer):
+        # TODO
+        return trainer
+
+    def get_callbacks(self):
+        callbacks = []
+        callbacks.append(GPUStatsCallback(self.cfg))
+        callbacks.append(EvalFirstStepCallback)
+
+        if self.cfg.relora_steps:
+            callbacks.append(ReLoRACallback(self.cfg))
+
+        if (
+            hasattr(self.model, "use_bettertransformer")
+            and self.model.use_bettertransformer is True
+        ):
+            callbacks.append(SaveBetterTransformerModelCallback)
+
+        if self.cfg.use_wandb:
+            callbacks.append(
+                SaveAxolotlConfigtoWandBCallback(self.cfg.axolotl_config_path)
+            )
+
+        if self.cfg.loss_watchdog_threshold is not None:
+            callbacks.append(LossWatchDogCallback(self.cfg))
+
+        return callbacks
+
+    def get_post_trainer_create_callbacks(self, trainer):
+        callbacks = []
+        if self.cfg.use_wandb and self.cfg.eval_table_size > 0:
+            LogPredictionCallback = log_prediction_callback_factory(
+                trainer, self.tokenizer
+            )
+            callbacks.append(LogPredictionCallback(self.cfg))
+
+        if self.cfg.do_bench_eval:
+            callbacks.append(bench_eval_callback_factory(trainer, self.tokenizer))
+
+        if self.cfg.early_stopping_patience:
+            early_stop_cb = EarlyStoppingCallback(
+                self.cfg.early_stopping_patience,
+            )
+            callbacks.append(early_stop_cb)
+
+        return callbacks
+
+    def _get_trainer_cls(self):
+        if self.cfg.lr_scheduler == "one_cycle" and (
+            self.cfg.fsdp or self.cfg.adapter == "qlora"
+        ):
+            return OneCycleLRSchedulerTrainer
+        if self.cfg.relora_steps:
+            return ReLoRATrainer
+        if self.cfg.model_config_type == "mamba":
+            return AxolotlMambaTrainer
+        return AxolotlTrainer
+
+    def build(self, total_num_steps):
+        warmup_steps = None
+        if self.cfg.warmup_steps is not None:
+            warmup_steps = self.cfg.warmup_steps
+        elif self.cfg.warmup_ratio is not None:
+            warmup_steps = max(int(self.cfg.warmup_ratio * total_num_steps), 0)
+        else:
+            warmup_steps = min(int(0.03 * total_num_steps), 100)
+
+        logging_steps = (
+            self.cfg.logging_steps
+            if self.cfg.logging_steps is not None
+            else max(min(int(0.005 * total_num_steps), 10), 1)
+        )
+
+        training_arguments_kwargs = {}
+        if self.cfg.bf16 == "full":
+            training_arguments_kwargs["bf16_full_eval"] = True
+        else:
+            training_arguments_kwargs["bf16"] = self.cfg.bf16
+        training_arguments_kwargs["fp16"] = (
+            self.cfg.fp16 and not self.cfg.bf16
+        ) or False
+        training_arguments_kwargs["tf32"] = self.cfg.tf32
+        training_arguments_kwargs["warmup_steps"] = warmup_steps
+        training_arguments_kwargs["logging_steps"] = logging_steps
+
+        if self.cfg.seed:
+            training_arguments_kwargs["seed"] = self.cfg.seed
+
+        if self.cfg.gradient_checkpointing:
+            training_arguments_kwargs[
+                "gradient_checkpointing"
+            ] = self.cfg.gradient_checkpointing
+            if self.cfg.gradient_checkpointing_kwargs:
+                training_arguments_kwargs[
+                    "gradient_checkpointing_kwargs"
+                ] = self.cfg.gradient_checkpointing_kwargs
+            else:
+                training_arguments_kwargs["gradient_checkpointing_kwargs"] = {
+                    "use_reentrant": False
+                }
+        if self.cfg.fsdp:
+            training_arguments_kwargs["fsdp"] = self.cfg.fsdp
+            if self.cfg.fsdp_config:
+                training_arguments_kwargs["fsdp_config"] = dict(self.cfg.fsdp_config)
+
+        # deepspeed
+        if self.cfg.deepspeed:
+            training_arguments_kwargs["deepspeed"] = self.cfg.deepspeed
+
+        if self.cfg.lr_quadratic_warmup is not None:
+            training_arguments_kwargs[
+                "lr_quadratic_warmup"
+            ] = self.cfg.lr_quadratic_warmup
+
+        if self.cfg.adam_beta1:
+            training_arguments_kwargs["adam_beta1"] = self.cfg.adam_beta1
+        if self.cfg.adam_beta2:
+            training_arguments_kwargs["adam_beta2"] = self.cfg.adam_beta2
+        if self.cfg.adam_epsilon:
+            training_arguments_kwargs["adam_epsilon"] = self.cfg.adam_epsilon
+        if self.cfg.max_grad_norm:
+            training_arguments_kwargs["max_grad_norm"] = self.cfg.max_grad_norm
+
+        if self.cfg.hub_model_id:
+            training_arguments_kwargs["hub_model_id"] = self.cfg.hub_model_id
+            training_arguments_kwargs["push_to_hub"] = True
+            training_arguments_kwargs["hub_private_repo"] = True
+            training_arguments_kwargs["hub_always_push"] = True
+
+            if self.cfg.hub_strategy:
+                training_arguments_kwargs["hub_strategy"] = self.cfg.hub_strategy
+
+        if self.cfg.save_safetensors is not None:
+            training_arguments_kwargs["save_safetensors"] = self.cfg.save_safetensors
+
+        if self.cfg.sample_packing_eff_est:
+            training_arguments_kwargs[
+                "sample_packing_efficiency"
+            ] = self.cfg.sample_packing_eff_est
+
+        if self.cfg.dataloader_pin_memory is not None:
+            training_arguments_kwargs[
+                "dataloader_pin_memory"
+            ] = self.cfg.dataloader_pin_memory
+        if self.cfg.dataloader_num_workers is not None:
+            training_arguments_kwargs[
+                "dataloader_num_workers"
+            ] = self.cfg.dataloader_num_workers
+        if self.cfg.dataloader_prefetch_factor is not None:
+            training_arguments_kwargs[
+                "dataloader_prefetch_factor"
+            ] = self.cfg.dataloader_prefetch_factor
+
+        if self.cfg.val_set_size == 0:
+            # no eval set, so don't eval
+            training_arguments_kwargs["evaluation_strategy"] = "no"
+        elif self.cfg.eval_steps:
+            training_arguments_kwargs["evaluation_strategy"] = "steps"
+            training_arguments_kwargs["eval_steps"] = self.cfg.eval_steps
+        elif self.cfg.evaluation_strategy:
+            training_arguments_kwargs[
+                "evaluation_strategy"
+            ] = self.cfg.evaluation_strategy
+        else:
+            # we have an eval set, but no steps defined, default to use epoch
+            training_arguments_kwargs["evaluation_strategy"] = "epoch"
+
+        if self.cfg.save_steps:
+            training_arguments_kwargs["save_strategy"] = "steps"
+            training_arguments_kwargs["save_steps"] = self.cfg.save_steps
+        elif self.cfg.save_strategy:
+            training_arguments_kwargs["save_strategy"] = self.cfg.save_strategy
+        else:
+            # default to saving each epoch if not defined
+            training_arguments_kwargs["save_strategy"] = "epoch"
+
+        if self.cfg.do_bench_eval:
+            training_arguments_kwargs["do_bench_eval"] = self.cfg.do_bench_eval
+            if self.cfg.bench_dataset:
+                training_arguments_kwargs["bench_dataset"] = self.cfg.bench_dataset
+        if self.cfg.metric_for_best_model:
+            training_arguments_kwargs[
+                "metric_for_best_model"
+            ] = self.cfg.metric_for_best_model
+        if self.cfg.greater_is_better:
+            training_arguments_kwargs["greater_is_better"] = self.cfg.greater_is_better
+
+        if self.cfg.torch_compile:
+            if torch.__version__ < "2.1.0":  # pylint: disable=protected-access
+                LOG.warning("torch>=2.1.0 required for torch_compile to work properly")
+            elif torch._dynamo:  # pylint: disable=protected-access
+                torch._dynamo.config.suppress_errors = (  # pylint: disable=protected-access
+                    True
+                )
+                training_arguments_kwargs["torch_compile"] = self.cfg.torch_compile
+                if self.cfg.torch_compile_backend:
+                    training_arguments_kwargs[
+                        "torch_compile_backend"
+                    ] = self.cfg.torch_compile_backend
+
+        # DDP Config
+        if self.cfg.ddp_timeout:
+            training_arguments_kwargs["ddp_timeout"] = self.cfg.ddp_timeout
+        # see https://pytorch.org/docs/stable/generated/torch.nn.parallel.DistributedDataParallel.html
+        if self.cfg.ddp_bucket_cap_mb:
+            training_arguments_kwargs["ddp_bucket_cap_mb"] = self.cfg.ddp_bucket_cap_mb
+        if self.cfg.ddp_broadcast_buffers is not None:
+            training_arguments_kwargs[
+                "ddp_broadcast_buffers"
+            ] = self.cfg.ddp_broadcast_buffers
+
+        # these are all the "standard" kwargs that are def used
+        training_arguments_kwargs["max_steps"] = (
+            total_num_steps if self.cfg.max_steps else -1
+        )
+        training_arguments_kwargs["max_seq_length"] = self.cfg.sequence_len
+        training_arguments_kwargs[
+            "per_device_train_batch_size"
+        ] = self.cfg.micro_batch_size
+        training_arguments_kwargs[
+            "per_device_eval_batch_size"
+        ] = self.cfg.eval_batch_size
+        training_arguments_kwargs[
+            "gradient_accumulation_steps"
+        ] = self.cfg.gradient_accumulation_steps
+        training_arguments_kwargs[
+            "eval_accumulation_steps"
+        ] = self.cfg.gradient_accumulation_steps
+        training_arguments_kwargs["num_train_epochs"] = self.cfg.num_epochs
+        training_arguments_kwargs["learning_rate"] = self.cfg.learning_rate
+        training_arguments_kwargs["output_dir"] = self.cfg.output_dir
+        training_arguments_kwargs["save_total_limit"] = (
+            self.cfg.save_total_limit if self.cfg.save_total_limit else 4
+        )
+        training_arguments_kwargs["load_best_model_at_end"] = (
+            (
+                self.cfg.load_best_model_at_end is not False
+                or self.cfg.early_stopping_patience
+            )
+            and self.cfg.val_set_size > 0
+            and self.cfg.save_steps
+            and self.cfg.eval_steps
+            and self.cfg.save_steps % self.cfg.eval_steps == 0
+        ) or False
+        training_arguments_kwargs["ddp_find_unused_parameters"] = (
+            False if self.cfg.ddp else None
+        )
+        training_arguments_kwargs["group_by_length"] = self.cfg.group_by_length
+        training_arguments_kwargs["report_to"] = "wandb" if self.cfg.use_wandb else None
+        training_arguments_kwargs["run_name"] = (
+            self.cfg.wandb_name if self.cfg.use_wandb else None
+        )
+        training_arguments_kwargs["optim"] = (
+            self.cfg.optimizer if self.cfg.optimizer else "adamw_hf"
+        )
+        training_arguments_kwargs["lr_scheduler_type"] = (
+            self.cfg.lr_scheduler
+            if self.cfg.lr_scheduler
+            and self.cfg.lr_scheduler not in ("one_cycle", "log_sweep")
+            else "cosine"
+        )
+        training_arguments_kwargs["lr_scheduler_kwargs"] = (
+            self.cfg.lr_scheduler_kwargs if self.cfg.lr_scheduler_kwargs else {}
+        )
+        training_arguments_kwargs["weight_decay"] = (
+            self.cfg.weight_decay if self.cfg.weight_decay is not None else 0.0
+        )
+        training_arguments_kwargs["sample_packing"] = (
+            self.cfg.sample_packing if self.cfg.sample_packing else False
+        )
+        training_arguments_kwargs["eval_sample_packing"] = (
+            self.cfg.sample_packing
+            if self.cfg.eval_sample_packing is not False
+            else False
+        )
+        training_arguments_kwargs[
+            "sample_packing_seq_len_multiplier"
+        ] = self.cfg.micro_batch_size
+        training_arguments_kwargs["relora_steps"] = self.cfg.relora_steps
+        training_arguments_kwargs["relora_warmup_steps"] = self.cfg.relora_warmup_steps
+        training_arguments_kwargs = self.hook_pre_create_training_args(
+            training_arguments_kwargs
+        )
+        training_arguments_kwargs["model_type"] = self.cfg.model_config_type
+        training_arguments_kwargs["pretraining"] = bool(self.cfg.pretraining_dataset)
+
+        if self.cfg.neftune_noise_alpha is not None:
+            training_arguments_kwargs[
+                "neftune_noise_alpha"
+            ] = self.cfg.neftune_noise_alpha
+
+        training_args = (
+            AxolotlTrainingArguments(  # pylint: disable=unexpected-keyword-arg
+                **training_arguments_kwargs,
+            )
+        )
+        training_args = self.hook_post_create_training_args(training_args)
+        trainer_kwargs = {}
+
+        if self.cfg.optimizer == "adamw_anyprecision":
+            if Path(self.cfg.torchdistx_path).exists():
+                sys.path.append(self.cfg.torchdistx_path)
+                importlib.import_module("torchdistx")
+
+        data_collator_kwargs = {
+            "padding": True,  # True/"longest" is the default
+        }
+        if self.cfg.pad_to_sequence_len:
+            data_collator_kwargs["pad_to_multiple_of"] = 64 * math.ceil(
+                self.cfg.sequence_len / 64
+            )
+        else:
+            # A100 is best at 64, while others at 8. Let's use the larger so we don't have to check
+            # https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html
+            data_collator_kwargs["pad_to_multiple_of"] = 64
+
+        trainer_cls = self._get_trainer_cls()
+        trainer_kwargs, trainer_cls = self.hook_pre_create_trainer(
+            trainer_kwargs, trainer_cls
+        )
+        trainer = trainer_cls(
+            model=self.model,
+            train_dataset=self.train_dataset,
+            eval_dataset=self.eval_dataset,
+            args=training_args,
+            data_collator=self.build_collator(training_args, **data_collator_kwargs),
+            bench_data_collator=transformers.DataCollatorForSeq2Seq(
+                self.tokenizer,
+                return_tensors="pt",
+                **data_collator_kwargs,
+            ),
+            callbacks=self.get_callbacks(),
+            num_epochs=self.cfg.num_epochs,
+            **trainer_kwargs,
+        )
+        trainer = self.hook_post_create_trainer(trainer)
+        for callback in self.get_post_trainer_create_callbacks(trainer):
+            trainer.add_callback(callback)
+
+        if self.cfg.deepspeed and self.cfg.sample_packing:
+            trainer.accelerator.state.deepspeed_plugin.deepspeed_config[
+                "train_micro_batch_size_per_gpu"
+            ] = self.cfg.micro_batch_size
+
+        return trainer
+
+    def build_collator(self, training_args: AxolotlTrainingArguments, **kwargs):
+        if training_args.pretraining:
+            return None
+
+        if self.cfg.model_config_type == "mamba":
+            return MambaDataCollator(tokenizer=self.tokenizer)
+
+        return BatchSamplerDataCollatorForSeq2Seq(
+            self.tokenizer,
+            return_tensors="pt",
+            **kwargs,
+        )
+
+
+class HFDPOTrainerBuilder(TrainerBuilderBase):
+    """
+    Trainer factory class for DPO Trainer
+    """
+
+    def get_callbacks(self):
+        callbacks = []
+        return callbacks
+
+    def get_post_trainer_create_callbacks(self, trainer):
+        callbacks = []
+        return callbacks
+
+    def build_training_arguments(self, total_num_steps):
+        training_args_kwargs = {}
+        for arg in [
+            "adam_beta1",
+            "adam_beta2",
+            "adam_epsilon",
+            "dataloader_num_workers",
+            "dataloader_pin_memory",
+        ]:
+            if hasattr(self.cfg, arg) and getattr(self.cfg, arg) is not None:
+                training_args_kwargs[arg] = getattr(self.cfg, arg)
+        training_args = TrainingArguments(
+            per_device_train_batch_size=self.cfg.micro_batch_size,
+            max_steps=total_num_steps,
+            remove_unused_columns=False,
+            gradient_accumulation_steps=self.cfg.gradient_accumulation_steps,
+            learning_rate=self.cfg.learning_rate,
+            evaluation_strategy="no",
+            # eval_steps=self.cfg.eval_steps,
+            save_strategy="steps",
+            save_steps=self.cfg.save_steps,
+            output_dir=self.cfg.output_dir,
+            warmup_steps=self.cfg.warmup_steps,
+            bf16=True,
+            gradient_checkpointing=self.cfg.gradient_checkpointing,
+            gradient_checkpointing_kwargs={"use_reentrant": False},
+            logging_first_step=True,
+            logging_steps=1,
+            optim=self.cfg.optimizer,
+            save_total_limit=self.cfg.save_total_limit or 5,
+            **training_args_kwargs,
+        )
+
+        return training_args
+
+    def build(self, total_num_steps):
+        training_args = self.build_training_arguments(total_num_steps)
+        dpo_trainer_kwargs = {}
+        if self.cfg.rl == "ipo":
+            dpo_trainer_kwargs["loss_type"] = "ipo"
+            if self.cfg.dpo_label_smoothing:
+                dpo_trainer_kwargs["label_smoothing"] = self.cfg.dpo_label_smoothing
+
+        dpo_trainer = DPOTrainer(
+            self.model,
+            self.model_ref,
+            args=training_args,
+            beta=self.cfg.dpo_beta or 0.1,
+            train_dataset=self.train_dataset,
+            # eval_dataset=self.eval_dataset,
+            eval_dataset=None,
+            tokenizer=self.tokenizer,
+            max_length=self.cfg.sequence_len,
+            max_target_length=None,
+            max_prompt_length=self.cfg.sequence_len,
+            generate_during_eval=True,
+            **dpo_trainer_kwargs,
+        )
+
+        return dpo_trainer
+
+
+class HFPPOTrainerBuilder(TrainerBuilderBase):
+    """
+    HF Factory class for PPO Trainer
+    """
+
+    def get_callbacks(self):
+        callbacks = []
+        return callbacks
+
+    def get_post_trainer_create_callbacks(self, trainer):
+        callbacks = []
+        return callbacks
+
+    def build(self, total_num_steps):
+        # build PPOConfig
+        pass

src/axolotl/core/trainers/trl.py

@@ -0,0 +1,66 @@
+"""
+module for TRL PPO training
+"""
+import torch
+from tqdm import tqdm
+from trl import PPOTrainer
+
+
+class TRLPPOTrainer(PPOTrainer):
+    """
+    wrapper for ppo trainer to handle customizations
+    """
+
+    def train(
+        self,
+        reward_pipe,
+        resume_from_checkpoint=None,  # pylint: disable=unused-argument
+    ):
+        generation_kwargs = {
+            "min_length": -1,
+            "top_k": 0.0,
+            "top_p": 1.0,
+            "do_sample": True,
+            "pad_token_id": self.tokenizer.eos_token_id,
+            "max_new_tokens": 32,
+        }
+        sent_kwargs = {
+            "return_all_scores": True,
+            "function_to_apply": "none",
+            "batch_size": 16,
+        }
+
+        for epoch, batch in tqdm(  # pylint: disable=unused-variable
+            enumerate(self.dataloader)
+        ):
+            query_tensors = batch["input_ids"]
+
+            # generate model response
+            response_tensors, ref_response_tensors = self.generate(
+                query_tensors,
+                return_prompt=False,
+                generate_ref_response=True,
+                **generation_kwargs
+            )
+            batch["response"] = self.tokenizer.batch_decode(response_tensors)
+            batch["ref_response"] = self.tokenizer.batch_decode(ref_response_tensors)
+
+            # Compute sentiment score
+            texts = [q + r for q, r in zip(batch["query"], batch["response"])]
+            pipe_outputs = reward_pipe(texts, **sent_kwargs)
+            rewards = [torch.tensor(output[1]["score"]) for output in pipe_outputs]
+            ref_texts = [q + r for q, r in zip(batch["query"], batch["ref_response"])]
+            ref_pipe_outputs = reward_pipe(ref_texts, **sent_kwargs)
+            ref_rewards = [
+                torch.tensor(output[1]["score"]) for output in ref_pipe_outputs
+            ]
+            batch["ref_rewards"] = ref_rewards
+
+            # Run PPO step
+            stats = self.step(query_tensors, response_tensors, rewards)
+            self.log_stats(
+                stats,
+                batch,
+                rewards,
+                columns_to_log=["query", "response", "ref_response", "ref_rewards"],
+            )

src/axolotl/datasets.py

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

src/axolotl/logging_config.py

@@ -1,16 +1,43 @@
-"""Logging configuration settings"""
+"""
+Common logging module for axolotl
+"""
 
 import os
 import sys
+from logging import Formatter
 from logging.config import dictConfig
 from typing import Any, Dict
 
+from colorama import Fore, Style, init
+
+
+class ColorfulFormatter(Formatter):
+    """
+    Formatter to add coloring to log messages by log type
+    """
+
+    COLORS = {
+        "WARNING": Fore.YELLOW,
+        "ERROR": Fore.RED,
+        "CRITICAL": Fore.RED + Style.BRIGHT,
+    }
+
+    def format(self, record):
+        record.rank = int(os.getenv("LOCAL_RANK", "0"))
+        log_message = super().format(record)
+        return self.COLORS.get(record.levelname, "") + log_message + Fore.RESET
+
+
 DEFAULT_LOGGING_CONFIG: Dict[str, Any] = {
     "version": 1,
     "formatters": {
         "simple": {
             "format": "[%(asctime)s] [%(levelname)s] [%(name)s.%(funcName)s:%(lineno)d] [PID:%(process)d] %(message)s",
         },
+        "colorful": {
+            "()": ColorfulFormatter,
+            "format": "[%(asctime)s] [%(levelname)s] [%(name)s.%(funcName)s:%(lineno)d] [PID:%(process)d] [RANK:%(rank)d] %(message)s",
+        },
     },
     "filters": {},
     "handlers": {
@@ -20,14 +47,25 @@ DEFAULT_LOGGING_CONFIG: Dict[str, Any] = {
             "filters": [],
             "stream": sys.stdout,
         },
+        "color_console": {
+            "class": "logging.StreamHandler",
+            "formatter": "colorful",
+            "filters": [],
+            "stream": sys.stdout,
+        },
     },
     "root": {"handlers": ["console"], "level": os.getenv("LOG_LEVEL", "INFO")},
     "loggers": {
-        "axolotl": {"handlers": ["console"], "level": "DEBUG", "propagate": False},
+        "axolotl": {
+            "handlers": ["color_console"],
+            "level": "DEBUG",
+            "propagate": False,
+        },
     },
 }
 
 
 def configure_logging():
     """Configure with default logging"""
+    init()  # Initialize colorama
     dictConfig(DEFAULT_LOGGING_CONFIG)

src/axolotl/models/mamba/__init__.py

@@ -0,0 +1,12 @@
+"""
+Modeling module for Mamba models
+"""
+
+
+def fix_mamba_attn_for_loss():
+    from mamba_ssm.models import mixer_seq_simple
+
+    from .modeling_mamba import MambaLMHeadModel as MambaLMHeadModelFixed
+
+    mixer_seq_simple.MambaLMHeadModel = MambaLMHeadModelFixed
+    return mixer_seq_simple.MambaLMHeadModel  # pylint: disable=invalid-name

src/axolotl/models/mamba/configuration_mamba.py

@@ -0,0 +1,42 @@
+"""
+HF Transformers MambaConfig
+"""
+from transformers import PretrainedConfig
+
+
+class MambaConfig(PretrainedConfig):
+    """
+    modeling configuration for state space model/mamba
+    """
+
+    model_type = "mamba"
+
+    def __init__(
+        self,
+        vocab_size=50280,
+        d_model=2560,
+        n_layer=64,
+        rms_norm=True,
+        residual_in_fp32=True,
+        fused_add_norm=True,
+        pad_vocab_size_multiple=8,
+        pad_token_id=50277,
+        bos_token_id=0,
+        eos_token_id=0,
+        tie_word_embeddings=False,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.d_model = d_model
+        self.n_layer = n_layer
+        self.rms_norm = rms_norm
+        self.residual_in_fp32 = residual_in_fp32
+        self.fused_add_norm = fused_add_norm
+        self.pad_vocab_size_multiple = pad_vocab_size_multiple
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )

src/axolotl/models/mamba/modeling_mamba.py

@@ -0,0 +1,128 @@
+# pylint: skip-file
+import os
+from collections import namedtuple
+from functools import partial
+from typing import Optional, Union
+
+import torch
+from mamba_ssm.models.mixer_seq_simple import MixerModel, _init_weights
+from mamba_ssm.utils.generation import GenerationMixin
+from mamba_ssm.utils.hf import load_config_hf, load_state_dict_hf
+from torch import nn
+from torch.nn import CrossEntropyLoss
+
+from axolotl.models.mamba.configuration_mamba import MambaConfig
+
+
+class MambaLMHeadModel(nn.Module, GenerationMixin):
+    def __init__(
+        self,
+        d_model: int,
+        n_layer: int,
+        vocab_size: int,
+        initializer_cfg=None,
+        pad_vocab_size_multiple: int = 1,
+        device=None,
+        dtype=None,
+        **backbone_kwargs,
+    ) -> None:
+        factory_kwargs = {"device": device, "dtype": dtype}
+        super().__init__()
+        if vocab_size % pad_vocab_size_multiple != 0:
+            vocab_size += pad_vocab_size_multiple - (
+                vocab_size % pad_vocab_size_multiple
+            )
+        self.config = MambaConfig(
+            vocab_size=vocab_size,
+            d_model=d_model,
+            n_layer=n_layer,
+            pad_vocab_size_multiple=pad_vocab_size_multiple,
+        )
+        self.backbone = MixerModel(
+            d_model=d_model,
+            n_layer=n_layer,
+            vocab_size=vocab_size,
+            initializer_cfg=initializer_cfg,
+            **backbone_kwargs,
+            **factory_kwargs,
+        )
+        self.lm_head = nn.Linear(d_model, vocab_size, bias=False, **factory_kwargs)
+
+        # Initialize weights and apply final processing
+        self.apply(
+            partial(
+                _init_weights,
+                n_layer=n_layer,
+                **(initializer_cfg if initializer_cfg is not None else {}),
+            )
+        )
+        self.tie_weights()
+
+    def tie_weights(self):
+        self.lm_head.weight = self.backbone.embedding.weight
+
+    def allocate_inference_cache(self, batch_size, max_seqlen, dtype=None, **kwargs):
+        return self.backbone.allocate_inference_cache(
+            batch_size, max_seqlen, dtype=dtype, **kwargs
+        )
+
+    def forward(
+        self,
+        input_ids,
+        position_ids=None,
+        inference_params=None,
+        num_last_tokens=0,
+        labels=None,
+        **kwargs,
+    ):
+        """
+        "position_ids" is just to be compatible with Transformer generation. We don't use it.
+        num_last_tokens: if > 0, only return the logits for the last n tokens
+        """
+        hidden_states = self.backbone(input_ids, inference_params=inference_params)
+        if num_last_tokens > 0:
+            hidden_states = hidden_states[:, -num_last_tokens:]
+        lm_logits = self.lm_head(hidden_states)
+
+        CausalLMOutput = namedtuple("CausalLMOutput", ["logits"])
+        return CausalLMOutput(logits=lm_logits)
+
+        loss = None
+        if labels is not None:
+            logits = lm_logits
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            # Enable model parallelism
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+            CausalLMOutput = namedtuple("CausalLMOutput", ["logits", "loss"])
+            print(loss)
+            return CausalLMOutput(logits=lm_logits, loss=loss)
+
+        else:
+            CausalLMOutput = namedtuple("CausalLMOutput", ["logits"])
+            return CausalLMOutput(logits=lm_logits)
+
+    def save_pretrained(
+        self,
+        save_directory: Union[str, os.PathLike],
+        state_dict: Optional[dict] = None,
+        safe_serialization: Optional[bool] = None,  # pylint: disable=unused-argument
+    ):
+        if state_dict is None:
+            state_dict = self.state_dict()
+        torch.save(state_dict, os.path.join(save_directory, "pytorch_model.bin"))
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_name, device=None, dtype=None, **kwargs):
+        config = load_config_hf(pretrained_model_name)
+        model = cls(**config, device=device, dtype=dtype, **kwargs)
+        model.load_state_dict(
+            load_state_dict_hf(pretrained_model_name, device={"": device}, dtype=dtype)
+        )
+        return model

src/axolotl/models/phi/__init__.py

@@ -0,0 +1,8 @@
+"""
+MixFormers model architecture used for phi models
+"""
+
+from .configuration_mixformer_sequential import MixFormerSequentialConfig  # noqa
+from .configuration_phi import PhiConfig  # noqa
+from .modeling_mixformer_sequential import MixFormerSequentialForCausalLM  # noqa
+from .modeling_phi import PhiForCausalLM  # noqa

src/axolotl/models/phi/configuration_mixformer_sequential.py

@@ -0,0 +1,63 @@
+# pylint: skip-file
+
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT license.
+
+import math
+from typing import Any, Dict, List, Optional, Union
+
+from transformers import PretrainedConfig
+
+
+class MixFormerSequentialConfig(PretrainedConfig):
+    """MixFormer (sequential for DeepSpeed) configuration."""
+
+    model_type = "mixformer-sequential"
+
+    attribute_map = {
+        "max_position_embeddings": "n_positions",
+        "hidden_size": "n_embd",
+        "num_attention_heads": "n_head",
+        "num_hidden_layers": "n_layer",
+        "input_emb_layer": "embd_layer",  # `input_emb_layer` key is for backward compatibility
+        "blocks": "architecture",  # `blocks` key is for backward compatibility
+    }
+
+    def __init__(
+        self,
+        vocab_size: Optional[int] = 50304,
+        n_positions: Optional[int] = 2048,
+        n_embd: Optional[int] = 1024,
+        n_layer: Optional[int] = 20,
+        n_inner: Optional[int] = None,
+        n_head: Optional[int] = 16,
+        rotary_dim: Optional[int] = 32,
+        activation_function: Optional[str] = "gelu_new",
+        embd_layer: Optional[str] = "default",
+        architecture: Union[Dict[str, Any], List[Dict[str, Any]]] = None,
+        embd_pdrop: Optional[float] = 0.0,
+        resid_pdrop: Optional[float] = 0.0,
+        layer_norm_epsilon: Optional[float] = 1e-5,
+        initializer_range: Optional[float] = 0.02,
+        tie_word_embeddings: Optional[bool] = False,
+        pad_vocab_size_multiple: Optional[int] = 64,
+        **kwargs
+    ) -> None:
+        self.vocab_size = int(
+            math.ceil(vocab_size / pad_vocab_size_multiple) * pad_vocab_size_multiple
+        )
+        self.n_positions = n_positions
+        self.n_embd = n_embd
+        self.n_layer = n_layer
+        self.n_inner = n_inner
+        self.n_head = n_head
+        self.rotary_dim = min(rotary_dim, n_embd // n_head)
+        self.activation_function = activation_function
+        self.embd_layer = embd_layer
+        self.architecture = architecture
+        self.embd_pdrop = embd_pdrop
+        self.resid_pdrop = resid_pdrop
+        self.layer_norm_epsilon = layer_norm_epsilon
+        self.initializer_range = initializer_range
+
+        super().__init__(tie_word_embeddings=tie_word_embeddings, **kwargs)

src/axolotl/models/phi/configuration_phi.py

@@ -0,0 +1,65 @@
+# pylint: skip-file
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT license.
+
+import math
+from typing import Optional
+
+from transformers import PretrainedConfig
+
+
+class PhiConfig(PretrainedConfig):
+    """Phi configuration."""
+
+    model_type = "phi"
+    attribute_map = {
+        "max_position_embeddings": "n_positions",
+        "hidden_size": "n_embd",
+        "num_attention_heads": "n_head",
+        "num_hidden_layers": "n_layer",
+    }
+
+    def __init__(
+        self,
+        vocab_size: int = 50304,
+        n_positions: int = 2048,
+        n_embd: int = 1024,
+        n_layer: int = 20,
+        n_inner: Optional[int] = None,
+        n_head: int = 16,
+        n_head_kv: Optional[int] = None,
+        rotary_dim: Optional[int] = 32,
+        activation_function: Optional[str] = "gelu_new",
+        flash_attn: bool = False,
+        flash_rotary: bool = False,
+        fused_dense: bool = False,
+        attn_pdrop: float = 0.0,
+        embd_pdrop: float = 0.0,
+        resid_pdrop: float = 0.0,
+        layer_norm_epsilon: float = 1e-5,
+        initializer_range: float = 0.02,
+        tie_word_embeddings: bool = False,
+        pad_vocab_size_multiple: int = 64,
+        **kwargs
+    ) -> None:
+        self.vocab_size = int(
+            math.ceil(vocab_size / pad_vocab_size_multiple) * pad_vocab_size_multiple
+        )
+        self.n_positions = n_positions
+        self.n_embd = n_embd
+        self.n_layer = n_layer
+        self.n_inner = n_inner
+        self.n_head = n_head
+        self.n_head_kv = n_head_kv
+        self.rotary_dim = min(rotary_dim, n_embd // n_head)
+        self.activation_function = activation_function
+        self.flash_attn = flash_attn
+        self.flash_rotary = flash_rotary
+        self.fused_dense = fused_dense
+        self.attn_pdrop = attn_pdrop
+        self.embd_pdrop = embd_pdrop
+        self.resid_pdrop = resid_pdrop
+        self.layer_norm_epsilon = layer_norm_epsilon
+        self.initializer_range = initializer_range
+
+        super().__init__(tie_word_embeddings=tie_word_embeddings, **kwargs)