add guassian dropout support

2023-09-25 14:50:39 -04:00
158 changed files with 3694 additions and 11358 deletions
--- a/.github/FUNDING.yml
+++ b/.github/FUNDING.yml
@@ -3,11 +3,11 @@
 github: OpenAccess-AI-Collective # Replace with up to 4 GitHub Sponsors-enabled usernames e.g., [user1, user2]
 patreon: # Replace with a single Patreon username
 open_collective: # Replace with a single Open Collective username
-ko_fi: axolotl_ai # Replace with a single Ko-fi username
+ko_fi: # Replace with a single Ko-fi username
 tidelift: # Replace with a single Tidelift platform-name/package-name e.g., npm/babel
 community_bridge: # Replace with a single Community Bridge project-name e.g., cloud-foundry
 liberapay: # Replace with a single Liberapay username
 issuehunt: # Replace with a single IssueHunt username
 otechie: # Replace with a single Otechie username
 lfx_crowdfunding: # Replace with a single LFX Crowdfunding project-name e.g., cloud-foundry
-custom: ['https://quickchart.io/qr?text=bitcoin%3Abc1qxlgwlqwfea5s2cxm42xqsfmwjct0rj8w8ea5np&size=480&centerImageUrl=https%3A%2F%2Fupload.wikimedia.org%2Fwikipedia%2Fcommons%2Fthumb%2F4%2F46%2FBitcoin.svg%2F64px-Bitcoin.svg.png'] # Replace with up to 4 custom sponsorship URLs e.g., ['link1', 'link2']
+custom: # Replace with up to 4 custom sponsorship URLs e.g., ['link1', 'link2']
--- a/.github/ISSUE_TEMPLATE/bug-report.yaml
+++ b/.github/ISSUE_TEMPLATE/bug-report.yaml
@@ -53,13 +53,6 @@ body:
    validations:
      required: true

-  - type: textarea
-    id: config
-    attributes:
-      label: Config yaml
-      description: |
-        Please attach the config yaml!
-
  - type: textarea
    id: possible-solution
    attributes:
--- a/.github/PULL_REQUEST_TEMPLATE/pull_request_template_simple.md
+++ b/.github/PULL_REQUEST_TEMPLATE/pull_request_template_simple.md
@@ -20,8 +20,3 @@
 ## Types of changes

 <!--- What types of changes does your code introduce? Put an `x` in all the boxes that apply: -->
-
-## Social Handles (Optional)
-
-<!-- Thanks for submitting a bugfix or enhancement. -->
-<!-- We'd love to show our thanks to you on Twitter & Discord if you provide your handle -->
--- a/.github/workflows/base.yml
+++ b/.github/workflows/base.yml
@@ -25,16 +25,6 @@ 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
--- a/.github/workflows/lint.yml
+++ b/.github/workflows/lint.yml
@@ -1,22 +0,0 @@
-name: lint
-on:
-  # check on PRs, and manual triggers
-  pull_request:
-      paths:
-       - '**.py'
-       - 'requirements.txt'
-       - '.github/workflows/*.yml'
-       - "*.md"
-  workflow_dispatch:
-
-jobs:
-  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
--- a/.github/workflows/main.yml
+++ b/.github/workflows/main.yml
@@ -23,60 +23,33 @@ jobs:
            python_version: "3.10"
            pytorch: 2.0.1
            axolotl_extras:
-            is_latest: true
-          - cuda: 118
-            cuda_version: 11.8.0
-            python_version: "3.10"
-            pytorch: 2.1.1
-            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@v4
+        uses: actions/checkout@v3
      - name: Docker metadata
        id: metadata
-        uses: docker/metadata-action@v5
+        uses: docker/metadata-action@v3
        with:
          images: winglian/axolotl
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
      - name: Login to Docker Hub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v2
        with:
          username: ${{ secrets.DOCKERHUB_USERNAME }}
          password: ${{ secrets.DOCKERHUB_TOKEN }}
-      # guidance for testing before pushing: https://docs.docker.com/build/ci/github-actions/test-before-push/
-      - name: Build and export to Docker
-        uses: docker/build-push-action@v5
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v2
+      - name: Build
+        uses: docker/build-push-action@v4
        with:
          context: .
-          load: true
          build-args: |
            BASE_TAG=${{ github.ref_name }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}
            CUDA=${{ matrix.cuda }}
-            PYTORCH_VERSION=${{ matrix.pytorch }}
          file: ./docker/Dockerfile
-          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) || '' }}
+          push: ${{ github.event_name != 'pull_request' }}
+          tags: ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
          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'
@@ -95,34 +68,24 @@ jobs:
            pytorch: 2.0.1
            axolotl_extras:
            is_latest: true
-          - cuda: 118
-            cuda_version: 11.8.0
-            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@v4
+        uses: actions/checkout@v3
      - name: Docker metadata
        id: metadata
-        uses: docker/metadata-action@v5
+        uses: docker/metadata-action@v3
        with:
          images: winglian/axolotl-runpod
      - name: Login to Docker Hub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v2
        with:
          username: ${{ secrets.DOCKERHUB_USERNAME }}
          password: ${{ secrets.DOCKERHUB_TOKEN }}
      - name: Set up Docker Buildx
        uses: docker/setup-buildx-action@v2
      - name: Build
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v4
        with:
          context: .
          build-args: |
--- a/.github/workflows/pypi.yml
+++ b/.github/workflows/pypi.yml
@@ -34,11 +34,11 @@ jobs:
        run: echo ::set-output name=TAG_NAME::$(echo $GITHUB_REF | cut -d / -f 3)

      - name: Update version in setup.py
-        run: |
+        run: >-
          sed -i -E 's/version="([0-9.]+)",/version="${{ steps.tag.outputs.TAG_NAME }}",/g' setup.py

      - name: Build a binary wheel
-        run: |
+        run: >-
          python setup.py sdist bdist_wheel

      - name: Publish package distributions to PyPI
--- a/.github/workflows/tests.yml
+++ b/.github/workflows/tests.yml
@@ -6,13 +6,9 @@ on:
      - "main"
    paths:
      - '**.py'
-      - 'requirements.txt'
-      - '.github/workflows/*.yml'
  pull_request:
      paths:
       - '**.py'
-       - 'requirements.txt'
-       - '.github/workflows/*.yml'
  workflow_dispatch:

 jobs:
@@ -33,7 +29,7 @@ jobs:
    strategy:
      fail-fast: false
      matrix:
-        python_version: ["3.9", "3.10", "3.11"]
+        python_version: ["3.9", "3.10"]
    timeout-minutes: 10

    steps:
@@ -48,61 +44,35 @@ jobs:

      - name: Install dependencies
        run: |
-          pip3 install -U -e .
+          pip3 install -e .
          pip3 install -r requirements-tests.txt

      - name: Run tests
        run: |
          pytest --ignore=tests/e2e/ tests/

-  docker-e2e-tests:
-    if: github.repository_owner == 'OpenAccess-AI-Collective'
-    # this job needs to be run on self-hosted GPU runners...
-    runs-on: [self-hosted, gpu, docker]
-    timeout-minutes: 30
+  e2e-test:
+    name: E2E Tests
+    runs-on: [self-hosted, gpu]
+    timeout-minutes: 20
    needs: [pre-commit, pytest]

-    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
    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-      - name: Docker metadata
-        id: metadata
-        uses: docker/metadata-action@v5
+      - name: Check out repository code
+        uses: actions/checkout@v3
+
+      - name: Setup Python
+        uses: actions/setup-python@v4
        with:
-          images: winglian/axolotl-tests
-      - name: Build Docker image
+          python-version: "3.10"
+#          cache: 'pip' # caching pip dependencies
+
+      - name: Install dependencies
        run: |
-          # Set up build arguments
-          BASE_TAG="main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}"
-          CUDA="${{ matrix.cuda }}"
-          PYTORCH_VERSION="${{ matrix.pytorch }}"
-          # Build the Docker image
-          docker build . \
-            --file ./docker/Dockerfile-tests \
-            --build-arg BASE_TAG=$BASE_TAG \
-            --build-arg CUDA=$CUDA \
-            --build-arg GITHUB_REF=$GITHUB_REF \
-            --build-arg PYTORCH_VERSION=$PYTORCH_VERSION \
-            --tag ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }} \
-            --no-cache
-      - name: Unit Tests w docker image
+          pip3 install -e .
+          pip3 install flash-attn
+          pip3 install -r requirements-tests.txt
+
+      - name: Run e2e tests
        run: |
-          docker run --rm ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }} pytest --ignore=tests/e2e/ /workspace/axolotl/tests/
-      - name: GPU Unit Tests w docker image
-        run: |
-          docker run --privileged --gpus "all" --env WANDB_DISABLED=true --rm ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }} pytest --ignore=tests/e2e/patched/ /workspace/axolotl/tests/e2e/
-      - name: GPU Unit Tests monkeypatched w docker image
-        run: |
-          docker run --privileged --gpus "all" --env WANDB_DISABLED=true --rm ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }} pytest /workspace/axolotl/tests/e2e/patched/
+          pytest tests/e2e/
--- a/.gitignore
+++ b/.gitignore
@@ -1,7 +1,5 @@
 **/axolotl.egg-info
 configs
-last_run_prepared/
-.vscode

 # Byte-compiled / optimized / DLL files
 __pycache__/
--- a/.mypy.ini
+++ b/.mypy.ini
@@ -8,9 +8,6 @@ ignore_missing_imports = True
 [mypy-axolotl.monkeypatch.*]
 ignore_errors = True

-[mypy-axolotl.models.mixtral.*]
-ignore_errors = True
-
 [mypy-axolotl.models.phi.*]
 ignore_errors = True

--- a/.vscode/README.md
+++ b/.vscode/README.md
@@ -1 +0,0 @@
-See [docs/debugging.md](../docs/debugging.md) for guidance on how to modify these files to debug axolotl with VSCode.
--- a/.vscode/launch.json
+++ b/.vscode/launch.json
@@ -1,34 +0,0 @@
-{
-    // Use IntelliSense to learn about possible attributes.
-    // Hover to view descriptions of existing attributes.
-    // For more information, visit: https://go.microsoft.com/fwlink/?linkid=830387
-    "version": "0.2.0",
-    "configurations": [
-        {
-            "name": "Debug axolotl prompt - sharegpt",
-            "type": "python",
-            "module": "accelerate.commands.launch",
-            "request": "launch",
-            "args": [
-                "-m", "axolotl.cli.train", "dev_sharegpt.yml",
-                // The flags below simplify debugging by overriding the axolotl config
-                // with the debugging tips above.  Modify as needed.
-                "--dataset_processes=1",      // limits data preprocessing to one process
-                "--max_steps=1",              // limits training to just one step
-                "--batch_size=1",             // minimizes batch size
-                "--micro_batch_size=1",       // minimizes batch size
-                "--val_set_size=0",           // disables validation
-                "--sample_packing=False",     // disables sample packing which is necessary for small datasets
-                "--eval_sample_packing=False",// disables sample packing on eval set
-                "--dataset_prepared_path=temp_debug/axolotl_outputs/data", // send data outputs to a temp folder
-                "--output_dir=temp_debug/axolotl_outputs/model" // send model outputs to a temp folder
-                ],
-            "console": "integratedTerminal",      // show output in the integrated terminal
-            "cwd": "${workspaceFolder}/devtools", // set working directory to devtools from the root of the project
-            "justMyCode": true,                   // step through only axolotl code
-            "env": {"CUDA_VISIBLE_DEVICES": "0",  // Since we aren't doing distributed training, we need to limit to one GPU
-                    "HF_HOME": "${workspaceFolder}/devtools/temp_debug/.hf-cache"}, // send HF cache to a temp folder
-            "preLaunchTask": "cleanup-for-dataprep", // delete temp folders (see below)
-        }
-    ]
-}
--- a/.vscode/tasks.json
+++ b/.vscode/tasks.json
@@ -1,27 +0,0 @@
-//this file is used by launch.json
-{
-  "version": "2.0.0",
-  "tasks": [
-    // this task changes into the devtools directory and deletes the temp_debug/axolotl_outputs folder
-    {
-      "label": "delete-outputs",
-      "type": "shell",
-      "command": "rm -rf temp_debug/axolotl_outputs",
-      "options":{ "cwd": "${workspaceFolder}/devtools"},
-      "problemMatcher": []
-    },
-    // this task changes into the devtools directory and deletes the `temp_debug/.hf-cache/datasets` folder
-    {
-      "label": "delete-temp-hf-dataset-cache",
-      "type": "shell",
-      "command": "rm -rf temp_debug/.hf-cache/datasets",
-      "options":{ "cwd": "${workspaceFolder}/devtools"},
-      "problemMatcher": []
-    },
-      // this task combines the two tasks above
-    {
-     "label": "cleanup-for-dataprep",
-     "dependsOn": ["delete-outputs", "delete-temp-hf-dataset-cache"],
-    }
-  ]
-}
--- a/README.md
+++ b/README.md
--- a/deepspeed/zero1.json
+++ b/deepspeed/zero1.json
@@ -24,6 +24,16 @@
      "weight_decay": "auto"
    }
  },
+  "scheduler": {
+    "type": "WarmupDecayLR",
+    "params": {
+      "warmup_min_lr": "auto",
+      "warmup_max_lr": "auto",
+      "warmup_num_steps": "auto",
+      "warmup_type": "linear",
+      "total_num_steps": "auto"
+    }
+  },
  "gradient_accumulation_steps": "auto",
  "train_batch_size": "auto",
  "train_micro_batch_size_per_gpu": "auto",
--- a/deepspeed/zero2.json
+++ b/deepspeed/zero2.json
@@ -28,6 +28,16 @@
      "weight_decay": "auto"
    }
  },
+  "scheduler": {
+    "type": "WarmupDecayLR",
+    "params": {
+      "warmup_min_lr": "auto",
+      "warmup_max_lr": "auto",
+      "warmup_num_steps": "auto",
+      "warmup_type": "linear",
+      "total_num_steps": "auto"
+    }
+  },
  "gradient_accumulation_steps": "auto",
  "train_batch_size": "auto",
  "train_micro_batch_size_per_gpu": "auto",
--- a/deepspeed/zero3.json
+++ b/deepspeed/zero3.json
@@ -1,6 +1,14 @@
 {
  "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,
@@ -32,6 +40,15 @@
      "weight_decay": "auto"
    }
  },
+  "scheduler": {
+    "type": "WarmupLR",
+    "params": {
+      "warmup_min_lr": "auto",
+      "warmup_max_lr": "auto",
+      "warmup_num_steps": "auto",
+      "warmup_type": "linear"
+    }
+  },
  "gradient_accumulation_steps": "auto",
  "train_batch_size": "auto",
  "train_micro_batch_size_per_gpu": "auto",
--- a/deepspeed/zero3_bf16.json
+++ b/deepspeed/zero3_bf16.json
@@ -1,39 +0,0 @@
-{
-  "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
-}
--- a/devtools/README.md
+++ b/devtools/README.md
@@ -1 +0,0 @@
-This directory contains example config files that might be useful for debugging. Please see [docs/debugging.md](../docs/debugging.md) for more information.
--- a/devtools/dev_sharegpt.yml
+++ b/devtools/dev_sharegpt.yml
@@ -1,49 +0,0 @@
-# Example config for debugging the sharegpt prompt format
-base_model: TinyLlama/TinyLlama-1.1B-Chat-v1.0
-model_type: LlamaForCausalLM
-tokenizer_type: LlamaTokenizer
-is_llama_derived_model: true
-
-load_in_8bit: true
-load_in_4bit: false
-
-datasets:
-  - path: philschmid/guanaco-sharegpt-style
-    type: sharegpt
-    shards: 10
-val_set_size: 0
-output_dir: temp_debug/axolotl_outputs/model
-dataset_prepared_path: temp_debug/axolotl_outputs/data
-dataset_processes: 1
-
-sequence_len: 4096
-sample_packing: false
-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:
-
-micro_batch_size: 1
-num_epochs: 1
-max_steps: 10
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: false
-fp16: true
-tf32: false
-
-gradient_checkpointing: true
-logging_steps: 1
-flash_attention: true
-
-warmup_steps: 10
-weight_decay: 0.0
--- a/docker/Dockerfile
+++ b/docker/Dockerfile
@@ -5,31 +5,24 @@ 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 --allow-change-held-packages vim curl nano libnccl2 libnccl-dev
+    apt-get install -y vim curl

 WORKDIR /workspace

 RUN git clone --depth=1 https://github.com/OpenAccess-AI-Collective/axolotl.git
-
-WORKDIR /workspace/axolotl
-
 # If AXOLOTL_EXTRAS is set, append it in brackets
-RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
-        pip install -e .[deepspeed,flash-attn,mamba-ssm,$AXOLOTL_EXTRAS]; \
+RUN cd axolotl && \
+    if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
+        pip install -e .[flash-attn,$AXOLOTL_EXTRAS]; \
    else \
-        pip install -e .[deepspeed,flash-attn,mamba-ssm]; \
+        pip install -e .[flash-attn]; \
    fi

-# So we can test the Docker image
-RUN pip install pytest
-
 # fix so that git fetch/pull from remote works
-RUN git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
+RUN cd axolotl && \
+    git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
    git config --get remote.origin.fetch

 # helper for huggingface-login cli
--- a/docker/Dockerfile-base
+++ b/docker/Dockerfile-base
@@ -10,13 +10,11 @@ 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 \
-    && apt-get install -y wget git build-essential ninja-build git-lfs libaio-dev && rm -rf /var/lib/apt/lists/* \
+    && 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 \
@@ -29,9 +27,52 @@ 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} deepspeed-kernels --extra-index-url https://download.pytorch.org/whl/cu$CUDA
+    python3 -m pip install --no-cache-dir -U torch==${PYTORCH_VERSION}+cu${CUDA} --extra-index-url https://download.pytorch.org/whl/cu$CUDA

-RUN git lfs install --skip-repo && \
-    pip3 install awscli && \
+FROM base-builder AS deepspeed-builder
+
+ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
+
+WORKDIR /workspace
+
+RUN git clone https://github.com/microsoft/DeepSpeed.git && \
+    cd DeepSpeed && \
+    MAX_CONCURRENCY=8 DS_BUILD_SPARSE_ATTN=0 DS_BUILD_OPS=1 DS_BUILD_EVOFORMER_ATTN=0 python3 setup.py bdist_wheel
+
+FROM base-builder AS bnb-builder
+
+WORKDIR /workspace
+ARG CUDA="118"
+ENV CUDA=$CUDA
+ARG MAX_JOBS="-1"
+ENV MAX_JOBS=$MAX_JOBS
+
+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
+RUN cd apex && python3 -m pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" ./
+
+RUN mkdir -p /workspace/builds
+COPY --from=bnb-builder /workspace/bitsandbytes /workspace/builds/bitsandbytes
+
+RUN mkdir -p /workspace/wheels/bitsandbytes
+COPY --from=deepspeed-builder /workspace/DeepSpeed/dist/deepspeed-*.whl wheels
+COPY --from=bnb-builder /workspace/bitsandbytes/dist/bitsandbytes-*.whl wheels
+COPY --from=bnb-builder /workspace/bitsandbytes/bitsandbytes/libbitsandbytes*.so wheels/bitsandbytes
+
+RUN pip3 install wheels/deepspeed-*.whl
+RUN cd /workspace/builds/bitsandbytes && python3 setup.py install
+RUN git lfs install --skip-repo
+RUN pip3 install awscli && \
    # The base image ships with `pydantic==1.8.2` which is not working
    pip3 install -U --no-cache-dir pydantic==1.10.10
--- a/docker/Dockerfile-runpod
+++ b/docker/Dockerfile-runpod
@@ -4,8 +4,6 @@ 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"
-ENV HF_HUB_ENABLE_HF_TRANSFER="1"

 COPY scripts/runpod-entrypoint.sh /root/runpod-entrypoint.sh

--- a/docker/Dockerfile-tests
+++ b/docker/Dockerfile-tests
@@ -1,40 +0,0 @@
-ARG BASE_TAG=main-base
-FROM winglian/axolotl-base:$BASE_TAG
-
-ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
-ARG AXOLOTL_EXTRAS=""
-ARG CUDA="118"
-ENV BNB_CUDA_VERSION=$CUDA
-ARG PYTORCH_VERSION="2.0.1"
-ARG GITHUB_REF="main"
-
-ENV PYTORCH_VERSION=$PYTORCH_VERSION
-
-RUN apt-get update && \
-    apt-get install -y --allow-change-held-packages vim curl nano libnccl2 libnccl-dev
-
-WORKDIR /workspace
-
-RUN git clone --depth=1 https://github.com/OpenAccess-AI-Collective/axolotl.git
-
-WORKDIR /workspace/axolotl
-
-RUN git fetch origin +$GITHUB_REF && \
-    git checkout FETCH_HEAD
-
-# If AXOLOTL_EXTRAS is set, append it in brackets
-RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
-        pip install -e .[deepspeed,flash-attn,mamba-ssm,$AXOLOTL_EXTRAS]; \
-    else \
-        pip install -e .[deepspeed,flash-attn,mamba-ssm]; \
-    fi
-
-# So we can test the Docker image
-RUN pip install pytest
-
-# fix so that git fetch/pull from remote works
-RUN git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
-    git config --get remote.origin.fetch
-
-# helper for huggingface-login cli
-RUN git config --global credential.helper store
--- a/docs/debugging.md
+++ b/docs/debugging.md
@@ -1,242 +0,0 @@
-# Debugging Axolotl
-
-This document provides some tips and tricks for debugging Axolotl.  It also provides an example configuration for debugging with VSCode.  A good debugging setup is essential to understanding how Axolotl code works behind the scenes.
-
-## Table of Contents
-
- [General Tips](#general-tips)
- [Debugging with VSCode](#debugging-with-vscode)
-    - [Background](#background)
-    - [Configuration](#configuration)
-    - [Customizing your debugger](#customizing-your-debugger)
-    - [Video Tutorial](#video-tutorial)
- [Debugging With Docker](#debugging-with-docker)
-    - [Setup](#setup)
-    - [Attach To Container](#attach-to-container)
-    - [Video - Attaching To Docker On Remote Host](#video---attaching-to-docker-on-remote-host)
-
-## General Tips
-
-While debugging it's helpful to simplify your test scenario as much as possible.  Here are some tips for doing so:
-
-> [!Important]
-> All of these tips are incorporated into the [example configuration](#configuration) for debugging with VSCode below.
-
-1. **Make sure you are using the latest version of axolotl**:  This project changes often and bugs get fixed fast.  Check your git branch and make sure you have pulled the latest changes from `main`.
-1. **Eliminate concurrency**: Restrict the number of processes to 1 for both training and data preprocessing:
-    - Set `CUDA_VISIBLE_DEVICES` to a single GPU, ex: `export CUDA_VISIBLE_DEVICES=0`.
-    - Set `dataset_processes: 1` in your axolotl config or run the training command with `--dataset_processes=1`.
-2. **Use a small dataset**: Construct or use a small dataset from HF Hub. When using a small dataset, you will often have to make sure `sample_packing: False` and `eval_sample_packing: False` to avoid errors.  If you are in a pinch and don't have time to construct a small dataset but want to use from the HF Hub, you can shard the data (this will still tokenize the entire dataset, but will only use a fraction of the data for training.  For example, to shard the dataset into 20 pieces, add the following to your axolotl config):
-    ```yaml
-    dataset:
-        ...
-        shards: 20
-    ```
-3. **Use a small model**: A good example of a small model is [TinyLlama/TinyLlama-1.1B-Chat-v1.0](https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v1.0).
-4. **Minimize iteration time**: Make sure the training loop finishes as fast as possible, with these settings.
-    - `micro_batch_size: 1`
-    - `max_steps: 1`
-    - `val_set_size: 0`
-5. **Clear Caches:** Axolotl caches certain steps and so does the underlying HuggingFace trainer.  You may want to clear some of these caches when debugging.
-    - Data preprocessing: When debugging data preprocessing, which includes prompt template formation, you may want to delete the directory set in `dataset_prepared_path:` in your axolotl config.  If you didn't set this value, the default is `last_run_prepared`.
-    - HF Hub: If you are debugging data preprocessing, you should clear the relevant HF cache [HuggingFace cache](https://huggingface.co/docs/datasets/cache), by deleting the appropriate `~/.cache/huggingface/datasets/...` folder(s).
-    - **The recommended approach is to redirect all outputs and caches to a temporary folder and delete selected subfolders before each run.  This is demonstrated in the example configuration below.**
-
-
-## Debugging with VSCode
-
-### Background
-
-The below example shows how to configure VSCode to debug data preprocessing of the `sharegpt` format.  This is the format used when you have the following in your axolotl config:
-
-```yaml
-datasets:
-  - path: <path to your sharegpt formatted dataset> # example on HF Hub: philschmid/guanaco-sharegpt-style
-    type: sharegpt
-```
-
->[!Important]
-> If you are already familiar with advanced VSCode debugging, you can skip the below explanation and look at the files [.vscode/launch.json](../.vscode/launch.json) and [.vscode/tasks.json](../.vscode/tasks.json) for an example configuration.
-
->[!Tip]
-> If you prefer to watch a video, rather than read, you can skip to the [video tutorial](#video-tutorial) below (but doing both is recommended).
-
-### Setup
-
-Make sure you have an [editable install](https://setuptools.pypa.io/en/latest/userguide/development_mode.html) of Axolotl, which ensures that changes you make to the code are reflected at runtime.  Run the following commands from the root of this project:
-
-```bash
-pip3 install packaging
-pip3 install -e '.[flash-attn,deepspeed]'
-```
-
-#### Remote Hosts
-
-If you developing on a remote host, you can easily use VSCode to debug remotely.  To do so, you will need to follow this [remote - SSH guide](https://code.visualstudio.com/docs/remote/ssh).  You can also see the video below on [Docker and Remote SSH debugging](#video---attaching-to-docker-on-remote-host).
-
-```bash
-
-### Configuration
-
-The easiest way to get started is to modify the [.vscode/launch.json](../.vscode/launch.json) file in this project.  This is just an example configuration, so you may need to modify or copy it to suit your needs.
-
-For example, to mimic the command `cd devtools && CUDA_VISIBLE_DEVICES=0 accelerate launch -m axolotl.cli.train dev_sharegpt.yml`, you would use the below configuration[^1].  Note that we add additional flags that override the axolotl config and incorporate the tips above (see the comments). We also set the working directory to `devtools` and set the `env` variable `HF_HOME` to a temporary folder that is later partially deleted.  This is because we want to delete the HF dataset cache before each run in order to ensure that the data preprocessing code is run from scratch.
-
-```jsonc
-// .vscode/launch.json
-{
-    "version": "0.2.0",
-    "configurations": [
-        {
-            "name": "Debug axolotl prompt - sharegpt",
-            "type": "python",
-            "module": "accelerate.commands.launch",
-            "request": "launch",
-            "args": [
-                "-m", "axolotl.cli.train", "dev_sharegpt.yml",
-                // The flags below simplify debugging by overriding the axolotl config
-                // with the debugging tips above.  Modify as needed.
-                "--dataset_processes=1",      // limits data preprocessing to one process
-                "--max_steps=1",              // limits training to just one step
-                "--batch_size=1",             // minimizes batch size
-                "--micro_batch_size=1",       // minimizes batch size
-                "--val_set_size=0",           // disables validation
-                "--sample_packing=False",     // disables sample packing which is necessary for small datasets
-                "--eval_sample_packing=False",// disables sample packing on eval set
-                "--dataset_prepared_path=temp_debug/axolotl_outputs/data", // send data outputs to a temp folder
-                "--output_dir=temp_debug/axolotl_outputs/model" // send model outputs to a temp folder
-                ],
-            "console": "integratedTerminal",      // show output in the integrated terminal
-            "cwd": "${workspaceFolder}/devtools", // set working directory to devtools from the root of the project
-            "justMyCode": true,                   // step through only axolotl code
-            "env": {"CUDA_VISIBLE_DEVICES": "0",  // Since we aren't doing distributed training, we need to limit to one GPU
-                    "HF_HOME": "${workspaceFolder}/devtools/temp_debug/.hf-cache"}, // send HF cache to a temp folder
-            "preLaunchTask": "cleanup-for-dataprep", // delete temp folders (see below)
-        }
-    ]
-}
-```
-
-**Additional notes about this configuration:**
-
- The argument `justMyCode` is set to `true` such that you step through only the axolotl code.  If you want to step into dependencies, set this to `false`.
- The `preLaunchTask`: `cleanup-for-dataprep` is defined in [.vscode/tasks.json](../.vscode/tasks.json) and is used to delete the following folders before debugging, which is essential to ensure that the data pre-processing code is run from scratch:
-    -  `./devtools/temp_debug/axolotl_outputs`
-    - `./devtools/temp_debug/.hf-cache/datasets`
-
->[!Tip]
-> You may not want to delete these folders. For example, if you are debugging model training instead of data pre-processing, you may NOT want to delete the cache or output folders. You may also need to add additional tasks to the `tasks.json` file depending on your use case.
-
-Below is the [./vscode/tasks.json](../.vscode/tasks.json) file that defines the `cleanup-for-dataprep` task.  This task is run before each debugging session when you use the above configuration.  Note how there are two tasks that delete the two folders mentioned above.  The third task `cleanup-for-dataprep` is a composite task that combines the two tasks.  A composite task is necessary because VSCode does not allow you to specify multiple tasks in the `preLaunchTask` argument of the `launch.json` file.
-
-```jsonc
-// .vscode/tasks.json
-// this file is used by launch.json
-{
-    "version": "2.0.0",
-    "tasks": [
-      // this task changes into the devtools directory and deletes the temp_debug/axolotl_outputs folder
-      {
-        "label": "delete-outputs",
-        "type": "shell",
-        "command": "rm -rf temp_debug/axolotl_outputs",
-        "options":{ "cwd": "${workspaceFolder}/devtools"},
-        "problemMatcher": []
-      },
-      // this task changes into the devtools directory and deletes the `temp_debug/.hf-cache/datasets` folder
-      {
-        "label": "delete-temp-hf-dataset-cache",
-        "type": "shell",
-        "command": "rm -rf temp_debug/.hf-cache/datasets",
-        "options":{ "cwd": "${workspaceFolder}/devtools"},
-        "problemMatcher": []
-      },
-        // this task combines the two tasks above
-      {
-       "label": "cleanup-for-dataprep",
-       "dependsOn": ["delete-outputs", "delete-temp-hf-dataset-cache"],
-      }
-    ]
-}
-```
-
-### Customizing your debugger
-
-Your debugging use case may differ from the example above.  The easiest thing to do is to put your own axolotl config in the `devtools` folder and modify the `launch.json` file to use your config.  You may also want to modify the `preLaunchTask` to delete different folders or not delete anything at all.
-
-### Video Tutorial
-
-The following video tutorial walks through the above configuration and demonstrates how to debug with VSCode, (click the image below to watch):
-
-<div style="text-align: center; line-height: 0;">
-
-<a href="https://youtu.be/xUUB11yeMmc" target="_blank"
-title="How to debug Axolotl (for fine tuning LLMs)"><img
-src="https://i.ytimg.com/vi/xUUB11yeMmc/maxresdefault.jpg"
-style="border-radius: 10px; display: block; margin: auto;" width="560" height="315" /></a>
-
-<figcaption style="font-size: smaller;"><a href="https://hamel.dev">Hamel Husain's</a> tutorial: <a href="https://www.youtube.com/watch?v=xUUB11yeMmc">Debugging Axolotl w/VSCode</a></figcaption>
-
-</div>
-<br>
-
-## Debugging With Docker
-
-Using [official Axolotl Docker images](https://hub.docker.com/r/winglian/axolotl/tags) is a great way to debug your code, and is a very popular way to use Axolotl.  Attaching VSCode to Docker takes a few more steps.
-
-### Setup
-
-On the host that is running axolotl (ex: if you are using a remote host), clone the axolotl repo and change your current directory to the root:
-
-```bash
-git clone https://github.com/OpenAccess-AI-Collective/axolotl
-cd axolotl
-```
-
->[!Tip]
-> If you already have axolotl cloned on your host, make sure you have the latest changes and change into the root of the project.
-
-Next, run the desired docker image and mount the current directory. Below is a docker command you can run to do this:[^2]
-
-```bash
-docker run --privileged --gpus '"all"' --shm-size 10g --rm -it --name axolotl --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 --mount type=bind,src="${PWD}",target=/workspace/axolotl -v ${HOME}/.cache/huggingface:/root/.cache/huggingface winglian/axolotl:main-py3.10-cu118-2.0.1
-```
-
->[!Tip]
-> To understand which containers are available, see the [Docker section of the README](../README.md#docker) and the [DockerHub repo](https://hub.docker.com/r/winglian/axolotl/tags).  For details of how the Docker containers are built, see axolotl's [Docker CI builds](../.github/workflows/main.yml).
-
-You will now be in the container.  Next, perform an editable install of Axolotl:
-
-```bash
-pip3 install packaging
-pip3 install -e '.[flash-attn,deepspeed]'
-```
-
-### Attach To Container
-
-Next, if you are using a remote host, [Remote into this host with VSCode](https://code.visualstudio.com/docs/remote/ssh).  If you are using a local host, you can skip this step.
-
-Next, select `Dev Containers: Attach to Running Container...` using the command palette (`CMD + SHIFT + P`) in VSCode.  You will be prompted to select a container to attach to.  Select the container you just created.  You will now be in the container with a working directory that is at the root of the project.  Any changes you make to the code will be reflected both in the container and on the host.
-
-Now you are ready to debug as described above (see [Debugging with VSCode](#debugging-with-vscode)).
-
-### Video - Attaching To Docker On Remote Host
-
-Here is a short video that demonstrates how to attach to a Docker container on a remote host:
-
-<div style="text-align: center; line-height: 0;">
-
-<a href="https://youtu.be/0AuoR7QnHR0" target="_blank"
-title="Debugging Axolotl Part 2: Attaching to Docker on a Remote Host"><img
-src="https://i.ytimg.com/vi/0AuoR7QnHR0/hqdefault.jpg"
-style="border-radius: 10px; display: block; margin: auto;" width="560" height="315" /></a>
-
-<figcaption style="font-size: smaller;"><a href="https://hamel.dev">Hamel Husain's</a> tutorial: <a href="https://youtu.be/0AuoR7QnHR0">Debugging Axolotl Part 2: Attaching to Docker on a Remote Host
-</a></figcaption>
-
-</div>
-<br>
-
-[^1]: The config actually mimics the command `CUDA_VISIBLE_DEVICES=0 python -m accelerate.commands.launch -m axolotl.cli.train devtools/sharegpt.yml`, but this is the same thing.
-
-[^2]: Many of the below flags are recommended best practices by Nvidia when using nvidia-container-toolkit.  You can read more about these flags [here](https://docs.nvidia.com/deeplearning/frameworks/user-guide/index.html).
--- a/docs/faq.md
+++ b/docs/faq.md
@@ -1,18 +0,0 @@
-# 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.
--- a/docs/multipack.md
+++ b/docs/multipack.md
@@ -1,51 +0,0 @@
-# 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 ]]
-```
--- a/docs/rlhf.md
+++ b/docs/rlhf.md
@@ -1,44 +0,0 @@
-# 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
-```
-
-#### Trl autounwrap for peft
-
-Trl supports autounwrapping peft models, so that a ref model does not need to be additionally loaded, leading to less VRAM needed. This is on by default. To turn it off, pass the following config.
-
-```yaml
-# load ref model when adapter training.
-rl_adapter_ref_model: true
-```
--- a/examples/cerebras/btlm-ft.yml
+++ b/examples/cerebras/btlm-ft.yml
@@ -1,4 +1,5 @@
 base_model: cerebras/btlm-3b-8k-base
+base_model_config: cerebras/btlm-3b-8k-base
 model_type: AutoModelForCausalLM
 tokenizer_type: GPT2Tokenizer
 trust_remote_code: true
@@ -14,7 +15,7 @@ datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
 dataset_prepared_path: last_prepared_run
-val_set_size: 0.05
+val_set_size: 0.01

 adapter:
 lora_model_dir:
@@ -35,7 +36,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:

 output_dir: btlm-out
@@ -72,8 +73,8 @@ gptq_groupsize:
 gptq_model_v1:

 warmup_steps: 32
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps:
+save_steps:
 save_total_limit:

 debug:
--- a/examples/cerebras/qlora.yml
+++ b/examples/cerebras/qlora.yml
@@ -1,4 +1,5 @@
 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
@@ -6,8 +7,8 @@ push_dataset_to_hub:
 datasets:
  - path: teknium/GPT4-LLM-Cleaned
    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 adapter: qlora
 lora_model_dir:
 sequence_len: 2048
@@ -24,7 +25,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./qlora-out
 batch_size: 4
@@ -49,8 +50,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 20
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.1
--- a/examples/code-llama/13b/lora.yml
+++ b/examples/code-llama/13b/lora.yml
@@ -1,4 +1,5 @@
 base_model: codellama/CodeLlama-13b-hf
+base_model_config: codellama/CodeLlama-13b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
 is_llama_derived_model: true
@@ -10,8 +11,8 @@ strict: false
 datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 output_dir: ./lora-out

 sequence_len: 4096
@@ -29,12 +30,12 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:

 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -54,8 +55,8 @@ xformers_attention:
 flash_attention: true

 warmup_steps: 10
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 20
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.0
--- a/examples/code-llama/13b/qlora.yml
+++ b/examples/code-llama/13b/qlora.yml
@@ -1,4 +1,5 @@
 base_model: codellama/CodeLlama-13b-hf
+base_model_config: codellama/CodeLlama-13b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
 is_llama_derived_model: true
@@ -10,8 +11,8 @@ strict: false
 datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 output_dir: ./qlora-out

 adapter: qlora
@@ -31,12 +32,12 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:

 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 4
+num_epochs: 3
 optimizer: paged_adamw_32bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -56,8 +57,8 @@ xformers_attention:
 flash_attention: true

 warmup_steps: 10
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 20
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.0
--- a/examples/code-llama/34b/lora.yml
+++ b/examples/code-llama/34b/lora.yml
@@ -1,4 +1,5 @@
 base_model: codellama/CodeLlama-34b-hf
+base_model_config: codellama/CodeLlama-34b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
 is_llama_derived_model: true
@@ -10,8 +11,8 @@ strict: false
 datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 output_dir: ./lora-out

 sequence_len: 4096
@@ -29,12 +30,12 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:

 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -54,8 +55,8 @@ xformers_attention:
 flash_attention: true

 warmup_steps: 10
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 20
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.0
--- a/examples/code-llama/34b/qlora.yml
+++ b/examples/code-llama/34b/qlora.yml
@@ -1,4 +1,5 @@
 base_model: codellama/CodeLlama-34b-hf
+base_model_config: codellama/CodeLlama-34b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
 is_llama_derived_model: true
@@ -10,8 +11,8 @@ strict: false
 datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 output_dir: ./qlora-out

 adapter: qlora
@@ -31,12 +32,12 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:

 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 4
+num_epochs: 3
 optimizer: paged_adamw_32bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -56,8 +57,8 @@ xformers_attention:
 flash_attention: true

 warmup_steps: 10
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 20
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.0
--- a/examples/code-llama/7b/lora.yml
+++ b/examples/code-llama/7b/lora.yml
@@ -1,4 +1,5 @@
 base_model: codellama/CodeLlama-7b-hf
+base_model_config: codellama/CodeLlama-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
 is_llama_derived_model: true
@@ -10,8 +11,8 @@ strict: false
 datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 output_dir: ./lora-out

 sequence_len: 4096
@@ -29,12 +30,12 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:

 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -54,8 +55,8 @@ xformers_attention:
 flash_attention: true

 warmup_steps: 10
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 20
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.0
--- a/examples/code-llama/7b/qlora.yml
+++ b/examples/code-llama/7b/qlora.yml
@@ -1,4 +1,5 @@
 base_model: codellama/CodeLlama-7b-hf
+base_model_config: codellama/CodeLlama-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
 is_llama_derived_model: true
@@ -10,8 +11,8 @@ strict: false
 datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 output_dir: ./qlora-out

 adapter: qlora
@@ -31,12 +32,12 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:

 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 4
+num_epochs: 3
 optimizer: paged_adamw_32bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -56,8 +57,8 @@ xformers_attention:
 flash_attention: true

 warmup_steps: 10
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 20
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.0
--- a/examples/falcon/config-7b-lora.yml
+++ b/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:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 adapter: lora
 lora_model_dir:
 sequence_len: 2048
@@ -26,7 +26,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./falcon-7b
 batch_size: 2
@@ -51,8 +51,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 40
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 5
+save_steps: 43
 debug:
 deepspeed:
 weight_decay: 0.0
--- a/examples/falcon/config-7b-qlora.yml
+++ b/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:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 # enable QLoRA
 adapter: qlora
 lora_model_dir:
@@ -40,7 +40,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 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: 4
+num_epochs: 3
 # Optimizer for QLoRA
 optimizer: paged_adamw_32bit
 torchdistx_path:
@@ -80,8 +80,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 5
+save_steps: 10
 debug:
 deepspeed:
 weight_decay: 0.000001
--- a/examples/falcon/config-7b.yml
+++ b/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:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 adapter:
 lora_model_dir:
 sequence_len: 2048
@@ -26,7 +26,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./falcon-7b
 batch_size: 2
@@ -51,8 +51,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 40
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 5
+save_steps: 43
 debug:
 deepspeed:
 weight_decay: 0.0
--- a/examples/gptj/qlora.yml
+++ b/examples/gptj/qlora.yml
@@ -1,4 +1,5 @@
 base_model: EleutherAI/gpt-j-6b
+base_model_config: EleutherAI/gpt-j-6b
 load_in_8bit: false
 load_in_4bit: true
 strict: false
@@ -6,8 +7,8 @@ push_dataset_to_hub:
 datasets:
  - path: teknium/GPT4-LLM-Cleaned
    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 adapter: qlora
 lora_model_dir:
 sequence_len: 2048
@@ -21,7 +22,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./qlora-out
 gradient_accumulation_steps: 2
@@ -46,8 +47,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 20
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.1
--- a/examples/jeopardy-bot/config.yml
+++ b/examples/jeopardy-bot/config.yml
@@ -1,11 +1,12 @@
 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:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.02
 adapter:
 lora_model_dir:
@@ -19,12 +20,12 @@ lora_fan_in_fan_out: false
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./jeopardy-bot-7b
 gradient_accumulation_steps: 1
 micro_batch_size: 1
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
@@ -42,8 +43,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 110
+save_steps: 660
 debug:
 deepspeed:
 weight_decay: 0.1
--- a/examples/llama-2/README.md
+++ b/examples/llama-2/README.md
@@ -9,16 +9,12 @@ gradient_accumulation_steps: 2
 micro_batch_size: 1

 ```shell
-accelerate launch -m axolotl.cli.train examples/llama-2/qlora.yml
+accelerate launch scripts/finetune.py examples/llama-2/qlora.yml
+
 ```
 or

 ```shell
-accelerate launch -m axolotl.cli.train examples/llama-2/lora.yml
-```
+accelerate launch scripts/finetune.py 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
 ```
--- a/examples/llama-2/fft_optimized.yml
+++ b/examples/llama-2/fft_optimized.yml
@@ -1,72 +0,0 @@
-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>"
--- a/examples/llama-2/gptq-lora.yml
+++ b/examples/llama-2/gptq-lora.yml
@@ -1,4 +1,5 @@
 base_model: TheBloke/Llama-2-7B-GPTQ
+base_model_config: TheBloke/Llama-2-7B-GPTQ
 is_llama_derived_model: false
 gptq: true
 gptq_disable_exllama: true
@@ -14,8 +15,8 @@ hf_use_auth_token: true
 datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 adapter: lora
 lora_model_dir:
 sequence_len: 4096
@@ -32,12 +33,12 @@ lora_target_linear:
 lora_fan_in_fan_out:
 wandb_project:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./model-out
 gradient_accumulation_steps: 1
 micro_batch_size: 1
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_torch
 adam_beta2: 0.95
 adam_eps: 0.00001
@@ -62,8 +63,8 @@ flash_attention:
 sdp_attention:
 flash_optimum:
 warmup_steps: 100
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps:
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.1
--- a/examples/llama-2/lora.yml
+++ b/examples/llama-2/lora.yml
@@ -1,4 +1,5 @@
 base_model: NousResearch/Llama-2-7b-hf
+base_model_config: NousResearch/Llama-2-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 is_llama_derived_model: true
@@ -10,8 +11,8 @@ strict: false
 datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 output_dir: ./lora-out

 sequence_len: 4096
@@ -29,12 +30,12 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:

 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -54,10 +55,10 @@ xformers_attention:
 flash_attention: true

 warmup_steps: 10
-evals_per_epoch: 4
-eval_table_size:
+eval_steps: 20
+eval_table_size: 5
 eval_table_max_new_tokens: 128
-saves_per_epoch: 1
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.0
--- a/examples/llama-2/qlora.yml
+++ b/examples/llama-2/qlora.yml
@@ -1,4 +1,5 @@
 base_model: NousResearch/Llama-2-7b-hf
+base_model_config: NousResearch/Llama-2-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 is_llama_derived_model: true
@@ -10,8 +11,8 @@ strict: false
 datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 output_dir: ./qlora-out

 adapter: qlora
@@ -31,12 +32,12 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:

 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 4
+num_epochs: 3
 optimizer: paged_adamw_32bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -56,9 +57,9 @@ xformers_attention:
 flash_attention: true

 warmup_steps: 10
-evals_per_epoch: 4
-eval_table_size:
-saves_per_epoch: 1
+eval_steps: 20
+eval_table_size: 5
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.0
--- a/examples/llama-2/relora.yml
+++ b/examples/llama-2/relora.yml
@@ -1,4 +1,5 @@
 base_model: NousResearch/Llama-2-7b-hf
+base_model_config: NousResearch/Llama-2-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 is_llama_derived_model: true
@@ -10,8 +11,8 @@ strict: false
 datasets:
  - path: teknium/GPT4-LLM-Cleaned
    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 output_dir: ./relora-out

 adapter: qlora
@@ -35,12 +36,12 @@ relora_cpu_offload: false
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:

 gradient_accumulation_steps: 4
 micro_batch_size: 4
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -60,8 +61,8 @@ xformers_attention:
 flash_attention: true

 warmup_steps: 10
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 20
+save_steps: 50
 debug:
 deepspeed:
 weight_decay: 0.0
--- a/examples/llama-2/tiny-llama.yml
+++ b/examples/llama-2/tiny-llama.yml
@@ -1,4 +1,6 @@
-base_model: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T
+base_model: PY007/TinyLlama-1.1B-step-50K-105b
+base_model_config: PY007/TinyLlama-1.1B-step-50K-105b
+
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 is_llama_derived_model: true
@@ -10,13 +12,12 @@ strict: false
 datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 output_dir: ./lora-out

 sequence_len: 4096
 sample_packing: true
-pad_to_sequence_len: true

 adapter: lora
 lora_model_dir:
@@ -29,12 +30,12 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:

 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -54,11 +55,15 @@ xformers_attention:
 flash_attention: true

 warmup_steps: 10
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 20
+eval_table_size: 5
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.0
 fsdp:
 fsdp_config:
 special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"
--- a/examples/mamba/config.yml
+++ b/examples/mamba/config.yml
@@ -1,61 +0,0 @@
-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
--- a/examples/mistral/README.md
+++ b/examples/mistral/README.md
@@ -1,12 +0,0 @@
-**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
-```
--- a/examples/mistral/config.yml
+++ b/examples/mistral/config.yml
@@ -1,62 +0,0 @@
-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>"
--- a/examples/mistral/mixtral.yml
+++ b/examples/mistral/mixtral.yml
@@ -1,91 +0,0 @@
-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:
--- a/examples/mistral/qlora.yml
+++ b/examples/mistral/qlora.yml
@@ -1,81 +0,0 @@
-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>"
--- a/examples/mpt-7b/config.yml
+++ b/examples/mpt-7b/config.yml
@@ -1,11 +1,12 @@
 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:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.02
 adapter:
 lora_model_dir:
@@ -21,12 +22,12 @@ lora_fan_in_fan_out: false
 wandb_project: mpt-alpaca-7b
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./mpt-alpaca-7b
 gradient_accumulation_steps: 1
 micro_batch_size: 1
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
@@ -44,8 +45,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 110
+save_steps: 660
 debug:
 deepspeed:
 weight_decay: 0.0001
--- a/examples/openllama-3b/config.yml
+++ b/examples/openllama-3b/config.yml
@@ -1,4 +1,5 @@
 base_model: openlm-research/open_llama_3b_v2
+base_model_config: openlm-research/open_llama_3b_v2
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 load_in_8bit: false
@@ -8,7 +9,7 @@ push_dataset_to_hub:
 datasets:
  - path: teknium/GPT4-LLM-Cleaned
    type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.02
 adapter:
 lora_model_dir:
@@ -23,7 +24,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./openllama-out
 gradient_accumulation_steps: 1
@@ -49,8 +50,8 @@ flash_attention: true
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 0.05
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.1
--- a/examples/openllama-3b/lora.yml
+++ b/examples/openllama-3b/lora.yml
@@ -1,4 +1,5 @@
 base_model: openlm-research/open_llama_3b_v2
+base_model_config: openlm-research/open_llama_3b_v2
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 load_in_8bit: true
@@ -8,7 +9,7 @@ push_dataset_to_hub:
 datasets:
  - path: teknium/GPT4-LLM-Cleaned
    type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.02
 adapter: lora
 lora_model_dir:
@@ -29,7 +30,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./lora-out
 gradient_accumulation_steps: 1
@@ -54,8 +55,8 @@ flash_attention: true
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 0.05
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.1
--- a/examples/openllama-3b/qlora.yml
+++ b/examples/openllama-3b/qlora.yml
@@ -1,4 +1,5 @@
 base_model: openlm-research/open_llama_3b_v2
+base_model_config: openlm-research/open_llama_3b_v2
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 load_in_8bit: false
@@ -8,8 +9,8 @@ push_dataset_to_hub:
 datasets:
  - path: teknium/GPT4-LLM-Cleaned
    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 adapter: qlora
 lora_model_dir:
 sequence_len: 1024
@@ -23,7 +24,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./qlora-out
 gradient_accumulation_steps: 1
@@ -48,8 +49,8 @@ flash_attention: true
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 0.05
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.1
--- a/examples/phi/phi-ft.yml
+++ b/examples/phi/phi-ft.yml
@@ -1,5 +1,6 @@
 base_model: microsoft/phi-1_5
-model_type: PhiForCausalLM
+base_model_config: microsoft/phi-1_5
+model_type: MixFormerSequentialForCausalLM
 tokenizer_type: AutoTokenizer
 is_llama_derived_model: false
 trust_remote_code: true
@@ -12,7 +13,7 @@ datasets:
  - path: garage-bAInd/Open-Platypus
    type: alpaca

-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.05
 output_dir: ./phi-sft-out

@@ -31,7 +32,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:

 gradient_accumulation_steps: 1
@@ -59,8 +60,8 @@ xformers_attention:
 flash_attention:

 warmup_steps: 100
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 0.05
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.1
--- a/examples/phi/phi-qlora.yml
+++ b/examples/phi/phi-qlora.yml
@@ -1,4 +1,5 @@
 base_model: microsoft/phi-1_5
+base_model_config: microsoft/phi-1_5
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
 is_llama_derived_model: false
@@ -12,7 +13,7 @@ datasets:
  - path: garage-bAInd/Open-Platypus
    type: alpaca

-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.05
 output_dir: ./phi-sft-out

@@ -31,7 +32,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:

 gradient_accumulation_steps: 1
@@ -59,8 +60,8 @@ xformers_attention:
 flash_attention:

 warmup_steps: 100
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 0.05
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0.1
--- a/examples/phi/phi2-ft.yml
+++ b/examples/phi/phi2-ft.yml
@@ -1,74 +0,0 @@
-base_model: microsoft/phi-2
-model_revision:  834565c  # pin model repo to the previous architecture
-model_type: AutoModelForCausalLM
-tokenizer_type: AutoTokenizer
-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: false  # currently unsupported
-pad_to_sequence_len:
-
-adapter:
-lora_model_dir:
-lora_r: 16
-lora_alpha: 32
-lora_dropout: 0.1
-lora_target_linear: true
-lora_fan_in_fan_out:
-lora_modules_to_save:
-  - embd
-  - lm_head
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 1
-micro_batch_size: 1
-num_epochs: 4
-optimizer: paged_adamw_8bit
-adam_beta2: 0.95
-adam_epsilon: 0.00001
-max_grad_norm: 1.0
-lr_scheduler: cosine
-learning_rate: 1e-5
-
-train_on_inputs: false
-group_by_length: false
-bf16: true
-fp16: false
-tf32: true
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_steps: 100
-evals_per_epoch: 4
-saves_per_epoch: 1
-debug:
-deepspeed:
-weight_decay: 0.1
-fsdp:
-fsdp_config:
-resize_token_embeddings_to_32x: true
-special_tokens:
-  pad_token: "<|endoftext|>"
--- a/examples/pythia-12b/config.yml
+++ b/examples/pythia-12b/config.yml
@@ -1,4 +1,5 @@
 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
@@ -9,7 +10,7 @@ device_map: auto
 datasets:
  - path: vicgalle/alpaca-gpt4
    type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.05
 adapter:
 lora_model_dir:
@@ -24,7 +25,7 @@ lora_fan_in_fan_out: true  # pythia/GPTNeoX lora specific
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./pythia-12b
 gradient_accumulation_steps: 1
--- a/examples/pythia/lora.yml
+++ b/examples/pythia/lora.yml
@@ -1,9 +1,10 @@
 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:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.05
 adapter: lora
 lora_model_dir:
@@ -18,20 +19,20 @@ lora_fan_in_fan_out: true  # pythia/GPTNeoX lora specific
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./lora-alpaca-pythia
 gradient_accumulation_steps: 1
 micro_batch_size: 4
-num_epochs: 4
+num_epochs: 3
 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
-evals_per_epoch: 4
+eval_steps: 20
 logging_steps: 1
--- a/examples/qwen/lora.yml
+++ b/examples/qwen/lora.yml
@@ -1,68 +0,0 @@
-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:
--- a/examples/qwen/qlora.yml
+++ b/examples/qwen/qlora.yml
@@ -1,68 +0,0 @@
-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:
--- a/examples/redpajama/config-3b.yml
+++ b/examples/redpajama/config-3b.yml
@@ -1,4 +1,5 @@
 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:
@@ -6,7 +7,7 @@ load_in_8bit: false
 datasets:
  - path: vicgalle/alpaca-gpt4
    type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.02
 adapter:
 lora_model_dir:
@@ -22,12 +23,12 @@ lora_fan_in_fan_out: false
 wandb_project: redpajama-alpaca-3b
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./redpajama-alpaca-3b
 batch_size: 4
 micro_batch_size: 1
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
@@ -45,8 +46,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 110
+save_steps: 660
 debug:
 deepspeed:
 weight_decay: 0.0001
--- a/examples/replit-3b/config-lora.yml
+++ b/examples/replit-3b/config-lora.yml
@@ -1,10 +1,11 @@
 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:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.05
 adapter: lora
 lora_model_dir:
@@ -21,12 +22,12 @@ lora_fan_in_fan_out:
 wandb_project: lora-replit
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./lora-replit
 batch_size: 8
 micro_batch_size: 1
-num_epochs: 4
+num_epochs: 3
 optimizer:
 torchdistx_path:
 lr_scheduler:
@@ -45,8 +46,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 50
+save_steps:
 debug:
 deepspeed:
 weight_decay: 0
--- a/examples/tiny-llama/README.md
+++ b/examples/tiny-llama/README.md
@@ -1,17 +0,0 @@
-# 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.
--- a/examples/tiny-llama/pretrain.yml
+++ b/examples/tiny-llama/pretrain.yml
@@ -1,58 +0,0 @@
-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:
--- a/examples/tiny-llama/qlora.yml
+++ b/examples/tiny-llama/qlora.yml
@@ -1,66 +0,0 @@
-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:
--- a/examples/xgen-7b/xgen-7b-8k-qlora.yml
+++ b/examples/xgen-7b/xgen-7b-8k-qlora.yml
@@ -1,6 +1,7 @@
 # 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
@@ -15,8 +16,8 @@ datasets:
    data_files:
      - openassistant_best_replies_train.jsonl
    type: "completion"
-dataset_prepared_path:
-val_set_size: 0.05
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
 # enable QLoRA
 adapter: qlora
 lora_model_dir:
@@ -38,7 +39,7 @@ lora_fan_in_fan_out:
 wandb_project:
 wandb_entity:
 wandb_watch:
-wandb_name:
+wandb_run_id:
 wandb_log_model:
 output_dir: ./qlora-out

@@ -51,7 +52,7 @@ output_dir: ./qlora-out
 # decrease if OOM, increase for max VRAM utilization
 micro_batch_size: 1
 gradient_accumulation_steps: 1
-num_epochs: 4
+num_epochs: 3
 # Optimizer for QLoRA
 optimizer: paged_adamw_32bit
 torchdistx_path:
@@ -78,8 +79,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-evals_per_epoch: 4
-saves_per_epoch: 1
+eval_steps: 50
+save_steps: 50
 debug:
 deepspeed:
 weight_decay: 0.0
--- a/examples/yi-34B-chat/README.md
+++ b/examples/yi-34B-chat/README.md
@@ -1,5 +0,0 @@
-# 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.
--- a/examples/yi-34B-chat/qlora.yml
+++ b/examples/yi-34B-chat/qlora.yml
@@ -1,76 +0,0 @@
-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:
--- a/image/sticker_fixed.png
+++ b/image/sticker_fixed.png
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,26 +1,28 @@
+--extra-index-url https://download.pytorch.org/whl/cu118
 --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/
-packaging==23.2
-peft==0.7.0
-transformers @ git+https://github.com/huggingface/transformers.git@3cefac1d974db5e2825a0cb2b842883a628be7a0
-tokenizers==0.15.0
+torch==2.0.1
+auto-gptq
+packaging
+peft @ git+https://github.com/huggingface/peft.git
+transformers @ git+https://github.com/huggingface/transformers.git
 bitsandbytes>=0.41.1
-accelerate @ git+https://github.com/huggingface/accelerate.git@0d2280dadc6a93413a5496613b7fdda3a4d2551b
+accelerate @ git+https://github.com/huggingface/accelerate
 deepspeed
 addict
+evaluate
 fire
 PyYAML>=6.0
-datasets>=2.15.0
-flash-attn==2.3.3
+datasets
+flash-attn>=2.2.1
 sentencepiece
 wandb
 einops
-xformers==0.0.22
-optimum==1.13.2
+xformers
+optimum
 hf_transfer
 colorama
 numba
 numpy>=1.24.4
-mlflow
 # qlora things
 bert-score==0.3.13
 evaluate==0.4.0
@@ -29,15 +31,3 @@ scipy
 scikit-learn==1.2.2
 pynvml
 art
-fschat==0.2.34
-gradio==3.50.2
-tensorboard
-
-mamba-ssm==1.1.1
-
-# remote filesystems
-s3fs
-gcsfs
-# adlfs
-
-trl>=0.7.9
--- a/scripts/finetune.py
+++ b/scripts/finetune.py
@@ -7,7 +7,6 @@ import transformers

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


--- a/setup.py
+++ b/setup.py
@@ -1,7 +1,5 @@
 """setup.py for axolotl"""

-from importlib.metadata import PackageNotFoundError, version
-
 from setuptools import find_packages, setup


@@ -11,28 +9,18 @@ def parse_requirements():
    with open("./requirements.txt", encoding="utf-8") as requirements_file:
        lines = [r.strip() for r in requirements_file.readlines()]
        for line in lines:
-            is_extras = (
-                "flash-attn" in line
-                or "flash-attention" in line
-                or "deepspeed" in line
-                or "mamba-ssm" in line
-            )
            if line.startswith("--extra-index-url"):
                # Handle custom index URLs
                _, url = line.split()
                _dependency_links.append(url)
-            elif not is_extras and line and line[0] != "#":
+            elif (
+                "flash-attn" not in line
+                and "deepspeed" not in line
+                and line
+                and line[0] != "#"
+            ):
                # Handle standard packages
                _install_requires.append(line)
-
-    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


@@ -50,19 +38,10 @@ setup(
    dependency_links=dependency_links,
    extras_require={
        "flash-attn": [
-            "flash-attn==2.3.3",
-        ],
-        "fused-dense-lib": [
-            "fused-dense-lib  @ git+https://github.com/Dao-AILab/flash-attention@v2.3.3#subdirectory=csrc/fused_dense_lib",
+            "flash-attn>=2.2.1",
        ],
        "deepspeed": [
            "deepspeed",
        ],
-        "mamba-ssm": [
-            "mamba-ssm==1.0.1",
-        ],
-        "auto-gptq": [
-            "auto-gptq==0.5.1",
-        ],
    },
 )
--- a/src/axolotl/cli/init.py
+++ b/src/axolotl/cli/init.py
@@ -2,41 +2,29 @@

 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 transformers import GenerationConfig, 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_cfg_datasets,
-    normalize_config,
-    validate_config,
-)
+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.mlflow_ import setup_mlflow_env_vars
 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__), ".."))
@@ -54,14 +42,14 @@ def print_axolotl_text_art(suffix=None):
    ascii_text = "  axolotl"
    if suffix:
        ascii_text += f"  x  {suffix}"
-    ascii_art = text2art(ascii_text, font=font)
+    ascii_art = text2art(" axolotl", 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): ")
+    print("Give me an instruction (Ctrl + D to finish): ")
    instruction = ""
    for line in sys.stdin:
        instruction += line  # pylint: disable=consider-using-join
@@ -78,15 +66,14 @@ def do_merge_lora(
    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)
+    model = model.merge_and_unload()
+    model.to(dtype=torch.float16)

    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"))

@@ -111,7 +98,15 @@ def do_inference(
            importlib.import_module("axolotl.prompters"), prompter
        )

-    model = model.to(cfg.device, dtype=cfg.torch_dtype)
+    if cfg.landmark_attention:
+        from axolotl.monkeypatch.llama_landmark_attn import set_model_mem_id
+
+        set_model_mem_id(model, tokenizer)
+        model.set_mem_cache_args(
+            max_seq_len=255, mem_freq=50, top_k=5, max_cache_size=None
+        )
+
+    model = model.to(cfg.device)

    while True:
        print("=" * 80)
@@ -156,83 +151,6 @@ def do_inference(
        print(tokenizer.decode(generated["sequences"].cpu().tolist()[0]))


-def do_inference_gradio(
-    *,
-    cfg: DictDefault,
-    cli_args: TrainerCliArgs,
-):
-    model, tokenizer = load_model_and_tokenizer(cfg=cfg, cli_args=cli_args)
-    prompter = cli_args.prompter
-    default_tokens = {"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>"}
-
-    for token, symbol in default_tokens.items():
-        # If the token isn't already specified in the config, add it
-        if not (cfg.special_tokens and token in cfg.special_tokens):
-            tokenizer.add_special_tokens({token: symbol})
-
-    prompter_module = None
-    if prompter:
-        prompter_module = getattr(
-            importlib.import_module("axolotl.prompters"), prompter
-        )
-
-    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"))

@@ -274,7 +192,6 @@ def load_cfg(config: Path = Path("examples/"), **kwargs):
    # load the config from the yaml file
    with open(config, encoding="utf-8") as file:
        cfg: DictDefault = DictDefault(yaml.safe_load(file))
-    cfg.axolotl_config_path = config
    # if there are any options passed in the cli, if it is something that seems valid from the yaml,
    # then overwrite the value
    cfg_keys = cfg.keys()
@@ -289,16 +206,9 @@ def load_cfg(config: Path = Path("examples/"), **kwargs):

    validate_config(cfg)

-    prepare_optim_env(cfg)
-
    normalize_config(cfg)

-    normalize_cfg_datasets(cfg)
-
    setup_wandb_env_vars(cfg)
-
-    setup_mlflow_env_vars(cfg)
-
    return cfg


@@ -309,9 +219,7 @@ def load_datasets(
 ) -> TrainDatasetMeta:
    tokenizer = load_tokenizer(cfg)

-    train_dataset, eval_dataset, total_num_steps, prompters = prepare_dataset(
-        cfg, tokenizer
-    )
+    train_dataset, eval_dataset, total_num_steps = prepare_dataset(cfg, tokenizer)

    if cli_args.debug or cfg.debug:
        LOG.info("check_dataset_labels...")
@@ -327,98 +235,6 @@ def load_datasets(
            text_only=cli_args.debug_text_only,
        )

-        LOG.info("printing prompters...")
-        for prompter in prompters:
-            LOG.info(prompter)
-
-    return TrainDatasetMeta(
-        train_dataset=train_dataset,
-        eval_dataset=eval_dataset,
-        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,
@@ -431,16 +247,3 @@ def check_accelerate_default_config():
        LOG.warning(
            f"accelerate config file found at {config_args.default_yaml_config_file}. This can lead to unexpected errors"
        )
-
-
-def check_user_token():
-    # Verify if token is valid
-    api = HfApi()
-    try:
-        user_info = api.whoami()
-        return bool(user_info)
-    except LocalTokenNotFoundError:
-        LOG.warning(
-            "Error verifying HuggingFace token. Remember to log in using `huggingface-cli login` and get your access token from https://huggingface.co/settings/tokens if you want to use gated models or datasets."
-        )
-        return False
--- a/src/axolotl/cli/inference.py
+++ b/src/axolotl/cli/inference.py
@@ -6,30 +6,21 @@ 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.cli import do_inference, load_cfg, print_axolotl_text_art
 from axolotl.common.cli import TrainerCliArgs


-def do_cli(config: Path = Path("examples/"), gradio=False, **kwargs):
+def do_cli(config: Path = Path("examples/"), **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)
+    do_inference(cfg=parsed_cfg, cli_args=parsed_cli_args)


 if __name__ == "__main__":
--- a/src/axolotl/cli/merge_lora.py
+++ b/src/axolotl/cli/merge_lora.py
@@ -18,26 +18,7 @@ def do_cli(config: Path = Path("examples/"), **kwargs):
        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."
-        )
-
-    parsed_cfg.load_in_4bit = False
-    parsed_cfg.load_in_8bit = False
-    parsed_cfg.flash_attention = False
+    parsed_cfg = load_cfg(config, merge_lora=True, **kwargs)

    do_merge_lora(cfg=parsed_cfg, cli_args=parsed_cli_args)

--- a/src/axolotl/cli/preprocess.py
+++ b/src/axolotl/cli/preprocess.py
@@ -1,54 +0,0 @@
-"""
-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)
--- a/src/axolotl/cli/train.py
+++ b/src/axolotl/cli/train.py
@@ -1,7 +1,6 @@
 """
 CLI to run training on a model
 """
-import logging
 from pathlib import Path

 import fire
@@ -9,33 +8,27 @@ 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()
+    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_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)
+    dataset_meta = load_datasets(cfg=parsed_cfg, cli_args=parsed_cli_args)
+    if parsed_cli_args.prepare_ds_only:
+        return
    train(cfg=parsed_cfg, cli_args=parsed_cli_args, dataset_meta=dataset_meta)


--- a/src/axolotl/common/cli.py
+++ b/src/axolotl/common/cli.py
@@ -25,22 +25,11 @@ class TrainerCliArgs:
    debug_num_examples: int = field(default=5)
    inference: bool = field(default=False)
    merge_lora: bool = field(default=False)
+    prepare_ds_only: bool = field(default=False)
    prompter: Optional[str] = field(default=None)
    shard: bool = field(default=False)


-@dataclass
-class PreprocessCliArgs:
-    """
-    dataclass representing arguments for preprocessing only
-    """
-
-    debug: bool = field(default=False)
-    debug_text_only: bool = field(default=False)
-    debug_num_examples: int = field(default=1)
-    prompter: Optional[str] = field(default=None)
-
-
 def load_model_and_tokenizer(
    *,
    cfg: DictDefault,
--- a/src/axolotl/common/const.py
+++ b/src/axolotl/common/const.py
@@ -1,5 +0,0 @@
-"""
-Various shared constants
-"""
-
-DEFAULT_DATASET_PREPARED_PATH = "last_run_prepared"
--- a/src/axolotl/core/init.py
+++ b/src/axolotl/core/init.py
--- a/src/axolotl/core/trainer_builder.py
+++ b/src/axolotl/core/trainer_builder.py
--- a/src/axolotl/core/trainers/init.py
+++ b/src/axolotl/core/trainers/init.py
--- a/src/axolotl/core/trainers/trl.py
+++ b/src/axolotl/core/trainers/trl.py
@@ -1,66 +0,0 @@
-"""
-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"],
-            )
--- a/src/axolotl/datasets.py
+++ b/src/axolotl/datasets.py
@@ -2,7 +2,7 @@

 import logging
 import os
-from typing import List, Optional
+from typing import List

 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 processing the data.
+            prompt_tokenizer (PromptTokenizingStrategy): The prompt tokenizing method for proccessing the data.
            dataset (dataset.Dataset): Dataset with text files.
    """

@@ -30,20 +30,14 @@ 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, self.process_count)
-            if self.process_count
-            else min(64, os.cpu_count())
-        )
+        num_proc = min(64, os.cpu_count())
        map_kwargs = {}
        if self.prompt_tokenizer.supports_batched:
            map_kwargs["batched"] = True
@@ -61,7 +55,7 @@ class ConstantLengthDataset(IterableDataset):
    """
    Iterable dataset that returns constant length chunks of tokens from stream of text files.
        Args:
-            tokenizer (Tokenizer): The processor used for processing the data.
+            tokenizer (Tokenizer): The processor used for proccessing the data.
            dataset (dataset.Dataset): Dataset with text files.
            seq_length (int): Length of token sequences to return.
    """
--- a/src/axolotl/models/mamba/init.py
+++ b/src/axolotl/models/mamba/init.py
@@ -1,24 +0,0 @@
-"""
-Modeling module for Mamba models
-"""
-
-import importlib
-
-
-def check_mamba_ssm_installed():
-    mamba_ssm_spec = importlib.util.find_spec("mamba_ssm")
-    if mamba_ssm_spec is None:
-        raise ImportError(
-            "MambaLMHeadModel requires mamba_ssm. Please install it with `pip install -e .[mamba-ssm]`"
-        )
-
-
-def fix_mamba_attn_for_loss():
-    check_mamba_ssm_installed()
-
-    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
--- a/src/axolotl/models/mamba/configuration_mamba.py
+++ b/src/axolotl/models/mamba/configuration_mamba.py
@@ -1,42 +0,0 @@
-"""
-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,
-        )
--- a/src/axolotl/models/mamba/modeling_mamba.py
+++ b/src/axolotl/models/mamba/modeling_mamba.py
@@ -1,128 +0,0 @@
-# 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
--- a/src/axolotl/models/phi/init.py
+++ b/src/axolotl/models/phi/init.py
@@ -3,6 +3,4 @@ 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
--- a/src/axolotl/models/phi/configuration_phi.py
+++ b/src/axolotl/models/phi/configuration_phi.py
@@ -1,65 +0,0 @@
-# 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)
--- a/src/axolotl/models/phi/modeling_mixformer_sequential.py
+++ b/src/axolotl/models/phi/modeling_mixformer_sequential.py
@@ -711,8 +711,12 @@ class ParallelBlock(nn.Module):
        self.resid_dropout = nn.Dropout(config.resid_pdrop)
        self.block_idx = block_idx

-        self.mixer = MHA(config, layer_idx=block_idx)
-        self.mlp = MLP(config)
+        self.mixer = MHA(config=config, **mixer, layer_idx=block_idx)
+        mlp_cls = mlp.pop("mlp_cls")
+        if mlp_cls == "fused_mlp":
+            self.mlp = FusedMLP(config=config, **mlp)
+        else:
+            self.mlp = MLP(config=config, **mlp)

    def forward(
        self,
--- a/src/axolotl/models/phi/modeling_phi.py
+++ b/src/axolotl/models/phi/modeling_phi.py
--- a/src/axolotl/monkeypatch/btlm_attn_hijack_flash.py
+++ b/src/axolotl/monkeypatch/btlm_attn_hijack_flash.py
@@ -7,7 +7,6 @@ import logging
 from typing import Optional, Tuple

 import torch
-from accelerate import init_empty_weights
 from flash_attn.flash_attn_interface import flash_attn_func
 from transformers import AutoConfig, AutoModelForCausalLM

@@ -18,8 +17,7 @@ def replace_btlm_attn_with_flash_attn(model_name="cerebras/btlm-3b-8k-base"):
    # this is a wonky hack to get the remotely loaded module
    model_config = AutoConfig.from_pretrained(model_name, trust_remote_code=True)
    # we need to load the model here in order for modeling_btlm to be available
-    with init_empty_weights():
-        AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True)
+    AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True)
    module_name = model_config.__class__.__module__.replace(
        ".configuration_btlm", ".modeling_btlm"
    )
--- a/src/axolotl/monkeypatch/falcon_attn_hijack_flash.py
+++ b/src/axolotl/monkeypatch/falcon_attn_hijack_flash.py
@@ -0,0 +1,101 @@
+"""
+Flash Attention monkey patch for Falcon
+
+copied from https://github.com/pacman100/DHS-LLM-Workshop/blob/main/chat_assistant/training/falcon_flash_attn_monkey_patch.py
+"""
+
+from typing import Optional, Tuple
+
+import torch
+import transformers
+from flash_attn import flash_attn_func
+
+
+def forward(
+    self,
+    hidden_states: torch.Tensor,
+    alibi: Optional[torch.Tensor],
+    attention_mask: torch.Tensor,  # pylint: disable=unused-argument
+    layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+    head_mask: Optional[torch.Tensor] = None,  # pylint: disable=unused-argument
+    use_cache: bool = False,
+    output_attentions: bool = False,  # pylint: disable=unused-argument
+):
+    fused_qkv = self.query_key_value(
+        hidden_states
+    )  # [batch_size, seq_length, 3 x hidden_size]
+    num_kv_heads = (
+        self.num_heads if self.new_decoder_architecture else self.num_kv_heads
+    )
+    # 3 x [batch_size, seq_length, num_heads, head_dim]
+    (
+        query_layer,
+        key_layer,
+        value_layer,
+    ) = self._split_heads(  # pylint: disable=protected-access
+        fused_qkv
+    )
+
+    batch_size, query_length, _, _ = query_layer.shape
+
+    query_layer = query_layer.transpose(1, 2).reshape(
+        batch_size * self.num_heads, query_length, self.head_dim
+    )
+    key_layer = key_layer.transpose(1, 2).reshape(
+        batch_size * num_kv_heads,
+        query_length,
+        self.head_dim,
+    )
+    value_layer = value_layer.transpose(1, 2).reshape(
+        batch_size * num_kv_heads, query_length, self.head_dim
+    )
+
+    past_kv_length = 0 if layer_past is None else layer_past[0].shape[1]
+    query_layer, key_layer = self.maybe_rotary(query_layer, key_layer, past_kv_length)
+
+    if layer_past is not None:
+        past_key, past_value = layer_past
+        # concatenate along seq_length dimension:
+        #  - key: [batch_size * self.num_heads, kv_length, head_dim]
+        #  - value: [batch_size * self.num_heads, kv_length, head_dim]
+        key_layer = torch.cat((past_key, key_layer), dim=1)
+        value_layer = torch.cat((past_value, value_layer), dim=1)
+
+    # unused
+    # _, kv_length, _ = key_layer.shape
+    if use_cache:
+        present = (key_layer, value_layer)
+    else:
+        present = None
+    # unused
+    # attention_mask_float = (attention_mask * 1.0).masked_fill(attention_mask, float("-1e9")).to(query_layer.dtype)
+    query_layer_ = (
+        query_layer.reshape(batch_size, self.num_heads, -1, self.head_dim)
+        .transpose(1, 2)
+        .to(torch.bfloat16)
+    )
+    key_layer_ = (
+        key_layer.reshape(batch_size, num_kv_heads, -1, self.head_dim)
+        .transpose(1, 2)
+        .to(torch.bfloat16)
+    )
+    value_layer_ = (
+        value_layer.reshape(batch_size, num_kv_heads, -1, self.head_dim)
+        .transpose(1, 2)
+        .to(torch.bfloat16)
+    )
+
+    if alibi is not None:
+        raise ValueError("`alibi` is not supported when `use_flash_attn` is True")
+
+    # below output will have shape (batch_size, seqlen, nheads, headdim)
+    attn_output = flash_attn_func(query_layer_, key_layer_, value_layer_, causal=True)
+    attn_output = attn_output.reshape(
+        batch_size, query_length, self.num_heads * self.head_dim
+    )
+    output_tensor = self.dense(attn_output)
+    return output_tensor, present
+
+
+def replace_falcon_attn_with_flash_attn():
+    transformers.models.falcon.modeling_falcon.FalconAttention.forward = forward
--- a/src/axolotl/monkeypatch/fastchat_conversation_turns.py
+++ b/src/axolotl/monkeypatch/fastchat_conversation_turns.py
@@ -1,212 +0,0 @@
-"""
-monkeypatch to add a get_turns method
-"""
-
-import logging
-from typing import Generator, Tuple
-
-from fastchat.conversation import SeparatorStyle
-
-LOG = logging.getLogger("axolotl.monkeypatch.fastchat_conversation_turns")
-
-
-def get_prompt(self) -> str:
-    ret = ""
-    for role, msg in self.get_turns():
-        ret += role + msg
-    return ret
-
-
-def get_turns(  # pylint: disable=too-many-return-statements
-    self,
-) -> Generator[Tuple[str, str], None, None]:
-    """Get the prompt for generation."""
-    system_prompt = self.system_template.format(system_message=self.system_message)
-    if self.sep_style == SeparatorStyle.ADD_COLON_SINGLE:
-        yield "", system_prompt + self.sep
-        for role, message in self.messages:
-            if message:
-                yield role + ": ", message + self.sep
-            else:
-                yield role + ":", ""
-        return
-    if self.sep_style == SeparatorStyle.ADD_COLON_TWO:
-        seps = [self.sep, self.sep2]
-        yield "", system_prompt + seps[0]
-        for i, (role, message) in enumerate(self.messages):
-            if message:
-                yield role + ": ", message + seps[i % 2]
-            else:
-                yield role + ":", ""
-        return
-    if self.sep_style == SeparatorStyle.ADD_COLON_SPACE_SINGLE:
-        yield "", system_prompt + self.sep
-        for role, message in self.messages:
-            if message:
-                yield role + ": ", message + self.sep
-            else:
-                yield role + ": ", ""  # must be end with a space
-        return
-    if self.sep_style == SeparatorStyle.ADD_NEW_LINE_SINGLE:
-        yield "", "" if system_prompt == "" else system_prompt + self.sep
-        for role, message in self.messages:
-            if message:
-                yield role + "\n", message + self.sep
-            else:
-                yield role + "\n", ""
-        return
-    if self.sep_style == SeparatorStyle.NO_COLON_SINGLE:
-        yield "", system_prompt
-        for role, message in self.messages:
-            if message:
-                yield role, message + self.sep
-            else:
-                yield role, ""
-        return
-    if self.sep_style == SeparatorStyle.NO_COLON_TWO:
-        seps = [self.sep, self.sep2]
-        yield "", system_prompt
-        for i, (role, message) in enumerate(self.messages):
-            if message:
-                yield role, message + seps[i % 2]
-            else:
-                yield role, ""
-        return
-    if self.sep_style == SeparatorStyle.RWKV:
-        yield "", system_prompt
-        for i, (role, message) in enumerate(self.messages):
-            if message:
-                yield role + ": ", message.replace("\r\n", "\n").replace(
-                    "\n\n", "\n"
-                ) + "\n\n"
-            else:
-                yield role + ":", ""
-        return
-    if self.sep_style == SeparatorStyle.LLAMA2 and self.name != "mistral":
-        if self.system_message:
-            if self.messages:
-                # For llama, the system message is incorporated into the first human instruction
-                first_role, first_msg = self.messages[0]
-                if first_role == self.roles[0]:
-                    system_prompt += first_msg
-                    self.messages.pop(0)
-            yield "", system_prompt
-        for i, (role, message) in enumerate(self.messages):
-            if message:
-                if (i % 2 == 0 and not self.system_message) or (
-                    i % 2 != 0 and self.system_message
-                ):
-                    role = "<s> " + role
-                yield role + " ", message
-            else:
-                yield role, ""
-        return
-    if self.sep_style == SeparatorStyle.LLAMA2 and self.name == "mistral":
-        contains_sys_msg = False
-        if self.system_message:
-            contains_sys_msg = True
-            if self.messages:
-                # There is no clear guidance on how to handle system messages in Mistral so we just prepend it to the first human instruction seperated by a newline
-                first_role, first_msg = self.messages[0]
-                if first_role == self.roles[0]:
-                    system_prompt = self.system_template.format(
-                        system_message=" " + self.system_message
-                    )
-                    system_prompt += first_msg
-                    self.messages.pop(0)
-            yield "", system_prompt
-        for i, (role, message) in enumerate(self.messages):
-            if message and i == 0 and not contains_sys_msg:
-                yield "", system_prompt.strip() + " " + message  # if there is no system message, we need to make sure there is the a `<s> [INST]` at the beginning of the first instruction.
-            elif message:
-                yield role + " ", message
-            else:
-                yield role, ""
-        return
-    if self.sep_style == SeparatorStyle.CHATGLM:
-        # source: https://huggingface.co/THUDM/chatglm-6b/blob/1d240ba371910e9282298d4592532d7f0f3e9f3e/modeling_chatglm.py#L1302-L1308
-        # source2: https://huggingface.co/THUDM/chatglm2-6b/blob/e186c891cf64310ac66ef10a87e6635fa6c2a579/modeling_chatglm.py#L926
-        round_add_n = 1 if self.name == "chatglm2" else 0
-        if system_prompt:
-            yield "", system_prompt + self.sep
-
-        for i, (role, message) in enumerate(self.messages):
-            if i % 2 == 0:
-                yield "", f"[Round {i//2 + round_add_n}]{self.sep}"
-
-            if message:
-                yield f"{role}：", f"{message}{self.sep}"
-            else:
-                yield f"{role}：", ""
-        return
-    if self.sep_style == SeparatorStyle.CHATML:
-        yield "", "" if system_prompt == "" else system_prompt + self.sep + "\n"
-        for role, message in self.messages:
-            if message:
-                yield role + "\n", message + self.sep + "\n"
-            else:
-                yield role + "\n", ""
-        return
-    if self.sep_style == SeparatorStyle.CHATGLM3:
-        if self.system_message:
-            yield "", system_prompt
-        for role, message in self.messages:
-            if message:
-                yield role + "\n", " " + message
-            else:
-                yield role
-        return
-    if self.sep_style == SeparatorStyle.CHATINTERN:
-        # source: https://huggingface.co/internlm/internlm-chat-7b-8k/blob/bd546fa984b4b0b86958f56bf37f94aa75ab8831/modeling_internlm.py#L771
-        seps = [self.sep, self.sep2]
-        yield "", system_prompt
-        for i, (role, message) in enumerate(self.messages):
-            prefix = "<s>" if i % 2 == 0 else ""
-            if message:
-                yield prefix + role + ":", message + seps[i % 2] + "\n"
-            else:
-                yield role + ":", ""
-        return
-    if self.sep_style == SeparatorStyle.DOLLY:
-        seps = [self.sep, self.sep2]
-        yield "", system_prompt
-        for i, (role, message) in enumerate(self.messages):
-            if message:
-                suffix = "\n\n" if i % 2 == 1 else ""
-                yield role + ":\n", message + seps[i % 2] + suffix
-            else:
-                yield role + ":\n", ""
-        return
-    if self.sep_style == SeparatorStyle.PHOENIX:
-        yield "", system_prompt
-        for role, message in self.messages:
-            if message:
-                yield role + ": ", "<s>" + message + "</s>"
-            else:
-                yield role + ": " + "<s>", ""
-        return
-    if self.sep_style == SeparatorStyle.ROBIN:
-        yield "", system_prompt + self.sep
-        for role, message in self.messages:
-            if message:
-                yield role + ":\n", message + self.sep
-            else:
-                yield role + ":\n", ""
-        return
-    if self.sep_style == SeparatorStyle.FALCON_CHAT:
-        if self.system_message:
-            yield "", system_prompt + self.sep
-        for role, message in self.messages:
-            if message:
-                yield role + ": ", message + self.sep
-            else:
-                yield role + ":", ""
-    else:
-        raise ValueError(f"Invalid style: {self.sep_style}")
-
-
-def add_get_turns_to_conversation():
-    import fastchat.conversation
-
-    fastchat.conversation.Conversation.get_turns = get_turns
-    fastchat.conversation.Conversation.get_prompt = get_prompt
--- a/src/axolotl/monkeypatch/llama_attn_hijack_flash.py
+++ b/src/axolotl/monkeypatch/llama_attn_hijack_flash.py
@@ -12,19 +12,15 @@ import torch.nn.functional as F
 import transformers
 from einops import rearrange
 from flash_attn.bert_padding import pad_input, unpad_input
+from torch import nn
+from transformers import LlamaConfig
 from transformers.modeling_outputs import BaseModelOutputWithPast
-from transformers.models.llama.modeling_llama import LlamaAttention
 from transformers.models.llama.modeling_llama import (
    LlamaDecoderLayer as OriginalLlamaDecoderLayer,
 )
-from transformers.models.llama.modeling_llama import (
-    LlamaMLP,
-    apply_rotary_pos_emb,
-    repeat_kv,
-)
-from xformers.ops import SwiGLU
+from transformers.models.llama.modeling_llama import apply_rotary_pos_emb, repeat_kv

-from axolotl.monkeypatch.utils import get_cu_seqlens_from_pos_ids, set_module_name
+from axolotl.monkeypatch.utils import get_cu_seqlens_from_pos_ids

 try:
    from flash_attn.flash_attn_interface import (  # pylint: disable=ungrouped-imports
@@ -44,33 +40,7 @@ except ImportError:
 LOG = logging.getLogger("axolotl")


-def replace_llama_mlp_with_swiglu(model):
-    for name, module in model.named_modules():
-        if isinstance(module, LlamaMLP):
-            mlp = FusedMLP(
-                module.config, module.gate_proj, module.up_proj, module.down_proj
-            )
-            set_module_name(model, name, mlp)
-
-
-def replace_llama_qkv_with_fused(model):
-    for name, module in model.named_modules():
-        if isinstance(module, LlamaAttention):
-            qkv = FusedAttention(
-                module.config,
-                module.q_proj,
-                module.k_proj,
-                module.v_proj,
-                module.o_proj,
-            )
-            set_module_name(model, name, qkv)
-
-
-def replace_llama_attn_with_flash_attn(
-    packed: Optional[bool] = False,
-    cross_entropy: Optional[bool] = False,
-    rms_norm: Optional[bool] = False,
-):
+def replace_llama_attn_with_flash_attn(packed: Optional[bool] = False):
    transformers.models.llama.modeling_llama.LlamaModel._prepare_decoder_attention_mask = (  # pylint: disable=protected-access
        _prepare_decoder_attention_mask
    )
@@ -81,123 +51,46 @@ def replace_llama_attn_with_flash_attn(
            llama_model_forward
        )

-    # skip only if explicitly disabled
-    if cross_entropy:
-        try:
-            from flash_attn.losses.cross_entropy import CrossEntropyLoss
+    try:
+        from flash_attn.losses.cross_entropy import CrossEntropyLoss

-            LOG.info("patching with flash_attn.losses.cross_entropy")
-            transformers.models.llama.modeling_llama.CrossEntropyLoss = partial(
-                CrossEntropyLoss, inplace_backward=True
-            )
-        except ImportError:
-            LOG.info(
-                "optimized flash-attention CrossEntropyLoss not found (run `pip install 'git+https://github.com/Dao-AILab/flash-attention.git#egg=xentropy_cuda_lib&subdirectory=csrc/xentropy'`)"
-            )
-
-    # skip only if explicitly disabled
-    if rms_norm:
-        try:
-            from flash_attn.ops.rms_norm import RMSNorm
-
-            class LlamaRMSNorm(RMSNorm):
-                """Patched LLamaRMSNorm"""
-
-                def __init__(self, hidden_size, eps=1e-6):
-                    super().__init__(hidden_size, eps=eps)
-
-            LOG.info("patching with flash_attn.ops.rms_norm")
-            transformers.models.llama.modeling_llama.LlamaRMSNorm = LlamaRMSNorm
-        except ImportError:
-            LOG.info(
-                "optimized flash-attention RMSNorm not found (run `pip install 'git+https://github.com/Dao-AILab/flash-attention.git#egg=dropout_layer_norm&subdirectory=csrc/layer_norm'`)"
-            )
-
-
-class FusedAttention(LlamaAttention):
-    """
-    Fused QKV Attention layer for incrementally improved training efficiency
-    """
-
-    def __init__(
-        self,
-        config,
-        q: torch.nn.Linear,  # pylint: disable=invalid-name
-        k: torch.nn.Linear,  # pylint: disable=invalid-name
-        v: torch.nn.Linear,  # pylint: disable=invalid-name
-        o: torch.nn.Linear,  # pylint: disable=invalid-name
-    ):
-        super().__init__(config)
-        self.config = config
-        self.init_device = next(iter(q.state_dict().values())).device
-
-        # define equivalent fused qkv projection
-        self.out_features: List[int] = [q.out_features, k.out_features, v.out_features]
-        self.qkv_proj = torch.nn.Linear(
-            q.in_features, sum(self.out_features), device=self.init_device, bias=False
+        LOG.info("patching with flash_attn.losses.cross_entropy")
+        transformers.models.llama.modeling_llama.CrossEntropyLoss = partial(
+            CrossEntropyLoss, inplace_backward=True
        )
-        self.o_proj = o
-
-        # overwrite initialized weights with pretrained weights
-        self.qkv_proj.weight.data = torch.cat(
-            (q.weight.data, k.weight.data, v.weight.data), dim=0
+    except ImportError:
+        LOG.info(
+            "optimized flash-attention CrossEntropyLoss not found (run `pip install 'git+https://github.com/Dao-AILab/flash-attention.git#egg=xentropy_cuda_lib&subdirectory=csrc/xentropy'`)"
        )

-    def _post_training(self, model, name):
-        q_proj, k_proj, v_proj = torch.split(
-            self.qkv_proj.weight.data, self.out_features, dim=0
+    try:
+        from flash_attn.ops.rms_norm import RMSNorm
+
+        class LlamaRMSNorm(RMSNorm):
+            """Patched LLamaRMSNorm"""
+
+            def __init__(self, hidden_size, eps=1e-6):
+                super().__init__(hidden_size, eps=eps)
+
+        LOG.info("patching with flash_attn.ops.rms_norm")
+        transformers.models.llama.modeling_llama.LlamaRMSNorm = LlamaRMSNorm
+    except ImportError:
+        LOG.info(
+            "optimized flash-attention RMSNorm not found (run `pip install 'git+https://github.com/Dao-AILab/flash-attention.git#egg=dropout_layer_norm&subdirectory=csrc/layer_norm'`)"
        )

-        new_attn = LlamaAttention(self.config)
-        new_attn.q_proj.weight.data = q_proj
-        new_attn.k_proj.weight.data = k_proj
-        new_attn.v_proj.weight.data = v_proj
-        new_attn.o_proj.weight.data = self.o_proj.weight.data

-        set_module_name(model, name, new_attn)
+class GaussianDropout(nn.Module):
+    def __init__(self, p=0.5):
+        super(GaussianDropout, self).__init__()
+        if p <= 0 or p >= 1:
+            raise Exception("p value should accomplish 0 < p < 1")
+        self.p = p

-
-class FusedMLP(torch.nn.Module):
-    """
-    Fused MLP layer for incrementally improved training efficiency
-    """
-
-    def __init__(
-        self,
-        config,
-        gate_proj: torch.nn.Linear,
-        up_proj: torch.nn.Linear,
-        down_proj: torch.nn.Linear,
-    ):
-        super().__init__()
-        self.config = config
-        self.swiglu = SwiGLU(
-            in_features=config.hidden_size,
-            hidden_features=config.intermediate_size,
-            bias=False,
-            _pack_weights=True,
-        )
-        # overwrite initialized weights with pretrained weights
-        self.swiglu.w12.weight.data = torch.cat(
-            (gate_proj.weight.data, up_proj.weight.data), dim=0
-        )
-        self.swiglu.w3.weight.data = down_proj.weight.data
-
-    def _post_training(self, model, name):
-        w1, w2 = torch.split(  # pylint: disable=invalid-name
-            self.swiglu.w12.weight.data, self.config.intermediate_size, dim=0
-        )
-
-        # Assign the split weights back to the original layers
-        new_mlp = LlamaMLP(self.config)
-        new_mlp.gate_proj.weight.data = w1
-        new_mlp.up_proj.weight.data = w2
-        new_mlp.down_proj.weight.data = self.swiglu.w3.weight.data
-
-        set_module_name(model, name, new_mlp)
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:  # pylint: disable=invalid-name
-        return self.swiglu(x)
+    def forward(self, x):
+        stddev = (self.p / (1.0 - self.p)) ** 0.5
+        epsilon = torch.randn_like(x) * stddev
+        return x * epsilon


 # Disable the transformation of the attention mask in LlamaModel as the flash attention
@@ -221,7 +114,6 @@ def flashattn_forward(
    past_key_value: Optional[Tuple[torch.Tensor]] = None,
    output_attentions: bool = False,
    use_cache: bool = False,
-    padding_mask: Optional[torch.LongTensor] = None,  # pylint: disable=unused-argument
    cu_seqlens: Optional[torch.Tensor] = None,
    max_seqlen: Optional[torch.Tensor] = None,
 ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
@@ -261,14 +153,9 @@ def flashattn_forward(
        value_states = torch.cat(value_states, dim=-1)

    else:
-        if isinstance(self, FusedAttention):
-            query_states, key_states, value_states = self.qkv_proj(hidden_states).split(
-                self.out_features, dim=-1
-            )
-        else:
-            query_states = self.q_proj(hidden_states)
-            key_states = self.k_proj(hidden_states)
-            value_states = self.v_proj(hidden_states)
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)

    query_states = query_states.view(
        bsz, q_len, self.num_heads, self.head_dim
@@ -321,8 +208,6 @@ def flashattn_forward(
        # only on first autoregressive step q,k,v have same seqlen
        is_causal = key_states.shape == query_states.shape

-    dropout_rate = 0.0 if not self.training else getattr(self, "attention_dropout", 0.0)
-
    if cu_seqlens is not None and max_seqlen is not None and cu_seqlens.dim() == 1:
        # special handling using sample packing
        qkv = torch.stack(
@@ -332,12 +217,7 @@ def flashattn_forward(
        qkv = rearrange(qkv, "b s ... -> (b s) ...")

        output = flash_attn_varlen_qkvpacked_func(
-            qkv,
-            cu_seqlens,
-            max_seqlen,
-            dropout_p=dropout_rate,
-            softmax_scale=None,
-            causal=True,
+            qkv, cu_seqlens, max_seqlen, dropout_p=0.0, softmax_scale=None, causal=True
        )
        output = rearrange(output, "(b s) ... -> b s ...", b=bsz)
    elif query_states.shape == key_states.shape:
@@ -360,7 +240,7 @@ def flashattn_forward(
            qkv_unpad,
            cu_seqlens_q,
            max_seqlen_q,
-            dropout_p=dropout_rate,
+            0.0,
            softmax_scale=None,
            causal=is_causal,
        )
@@ -373,7 +253,6 @@ def flashattn_forward(
            output = flash_attn_kvpacked_func(
                query_states,
                torch.stack([key_states, value_states], 2),
-                dropout_p=dropout_rate,
                causal=is_causal,
            )
        else:
@@ -406,7 +285,7 @@ def flashattn_forward(
                cu_seqlens_k,
                max_seqlen_q,
                max_seqlen_k,
-                dropout_p=dropout_rate,
+                0.0,
                softmax_scale=None,
                causal=is_causal,
            )
@@ -612,13 +491,6 @@ def llama_model_forward(
            dtype=torch.bool,
            device=inputs_embeds.device,
        )
-        padding_mask = None
-    else:
-        if 0 in attention_mask:
-            padding_mask = attention_mask
-        else:
-            padding_mask = None
-
    attention_mask = (
        self._prepare_decoder_attention_mask(  # pylint: disable=protected-access
            attention_mask,
@@ -653,9 +525,7 @@ def llama_model_forward(
            def create_custom_forward(module):
                def custom_forward(*inputs):
                    # None for past_key_value
-                    return module(
-                        *inputs,
-                    )
+                    return module(*inputs)

                return custom_forward

@@ -664,10 +534,9 @@ def llama_model_forward(
                hidden_states,
                attention_mask,
                position_ids,
-                past_key_value,
+                None,
                output_attentions,
                None,
-                padding_mask,
                cu_seqlens,
                max_seqlen,
            )
@@ -679,7 +548,6 @@ def llama_model_forward(
                past_key_value=past_key_value,
                output_attentions=output_attentions,
                use_cache=use_cache,
-                padding_mask=padding_mask,
                cu_seqlens=cu_seqlens,
                max_seqlen=max_seqlen,
            )
@@ -718,6 +586,15 @@ class LlamaDecoderLayer(OriginalLlamaDecoderLayer):
    patched version of LlamaDecoderLayer to pass through the precalculated cu_seqlens
    """

+    def __init__(self, config: LlamaConfig):
+        super(LlamaDecoderLayer, self).__init__(config)
+        self.attn_dropout = None
+        self.mlp_dropout = None
+        if config.dropout_attn:
+            self.attn_dropout = GaussianDropout(p=config.dropout_attn)
+        if config.dropout_mlp:
+            self.mlp_dropout = GaussianDropout(p=config.dropout_mlp)
+
    def forward(
        self,
        hidden_states: torch.Tensor,
@@ -726,7 +603,6 @@ class LlamaDecoderLayer(OriginalLlamaDecoderLayer):
        past_key_value: Optional[Tuple[torch.Tensor]] = None,
        output_attentions: Optional[bool] = False,
        use_cache: Optional[bool] = False,
-        padding_mask: Optional[torch.LongTensor] = None,
        cu_seqlens: Optional[torch.Tensor] = None,
        max_seqlen: Optional[torch.Tensor] = None,
    ) -> Tuple[
@@ -759,16 +635,19 @@ class LlamaDecoderLayer(OriginalLlamaDecoderLayer):
            past_key_value=past_key_value,
            output_attentions=output_attentions,
            use_cache=use_cache,
-            padding_mask=padding_mask,
            cu_seqlens=cu_seqlens,
            max_seqlen=max_seqlen,
        )
+        if self.training and self.attn_dropout:
+            hidden_states = self.attn_dropout(hidden_states)
        hidden_states = residual + hidden_states

        # Fully Connected
        residual = hidden_states
        hidden_states = self.post_attention_layernorm(hidden_states)
        hidden_states = self.mlp(hidden_states)
+        if self.training and self.mlp_dropout:
+            hidden_states = self.mlp_dropout(hidden_states)
        hidden_states = residual + hidden_states

        outputs = (hidden_states,)
--- a/src/axolotl/monkeypatch/llama_attn_hijack_sdp.py
+++ b/src/axolotl/monkeypatch/llama_attn_hijack_sdp.py
@@ -25,8 +25,6 @@ def sdp_attention_forward(
    past_key_value: Optional[Tuple[torch.Tensor]] = None,
    output_attentions: bool = False,
    use_cache: bool = False,
-    padding_mask: Optional[torch.LongTensor] = None,  # pylint: disable=unused-argument
-    **kwargs,  # pylint: disable=unused-argument
 ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
    # pylint: disable=duplicate-code
    bsz, q_len, _ = hidden_states.size()
--- a/Show More
+++ b/Show More
				`@@ -1 +0,0 @@`
				`See [docs/debugging.md](../docs/debugging.md) for guidance on how to modify these files to debug axolotl with VSCode.`
				`@@ -1 +0,0 @@`
				`This directory contains example config files that might be useful for debugging. Please see [docs/debugging.md](../docs/debugging.md) for more information.`