use low_cpu_mem_usage with ds zero 1 or 2

use low_cpu_mem_usage when using deepspeed
2024-01-16 19:33:44 -05:00 · 2024-01-16 07:44:35 -05:00
186 changed files with 4402 additions and 10858 deletions
--- a/.github/FUNDING.yml
+++ b/.github/FUNDING.yml
@@ -1,6 +1,6 @@
 # These are supported funding model platforms

-github: [winglian, OpenAccess-AI-Collective] # Replace with up to 4 GitHub Sponsors-enabled usernames e.g., [user1, user2]
+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
--- a/.github/ISSUE_TEMPLATE/bug-report.yaml
+++ b/.github/ISSUE_TEMPLATE/bug-report.yaml
@@ -59,7 +59,6 @@ body:
      label: Config yaml
      description: |
        Please attach the config yaml!
-      render: yaml

  - type: textarea
    id: possible-solution
--- a/.github/workflows/base.yml
+++ b/.github/workflows/base.yml
@@ -1,31 +1,39 @@
 name: ci-cd-base

 on:
-  workflow_dispatch:
+  push:
+    branches:
+      - "main-base"
+      - "dev-base"

 jobs:
  build-base:
    if: github.repository_owner == 'OpenAccess-AI-Collective'
    # this job needs to be run on self-hosted GPU runners...
-    runs-on: axolotl-gpu-runner
+    runs-on: self-hosted
    strategy:
      fail-fast: false
      matrix:
        include:
+          - cuda: "118"
+            cuda_version: 11.8.0
+            python_version: "3.9"
+            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: 2.1.2
+            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: 2.1.2
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 9.0+PTX"
-          - cuda: "121"
-            cuda_version: 12.1.0
-            python_version: "3.11"
-            pytorch: 2.1.2
+            pytorch: 2.1.1
            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 9.0+PTX"
    steps:
      - name: Checkout
@@ -48,7 +56,7 @@ jobs:
          context: .
          file: ./docker/Dockerfile-base
          push: ${{ github.event_name != 'pull_request' }}
-          tags: ${{ steps.metadata.outputs.tags }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+          tags: ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
          labels: ${{ steps.metadata.outputs.labels }}
          build-args: |
            CUDA_VERSION=${{ matrix.cuda_version }}
--- a/.github/workflows/lint.yml
+++ b/.github/workflows/lint.yml
@@ -17,6 +17,6 @@ jobs:
      - uses: actions/checkout@v3
      - uses: actions/setup-python@v4
        with:
-          python-version: "3.10"
+          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
@@ -4,33 +4,37 @@ on:
  push:
    branches:
      - "main"
-  workflow_dispatch:

 jobs:
  build-axolotl:
-    if: ${{ ! contains(github.event.commits[0].message, '[skip docker]]') && github.repository_owner == 'OpenAccess-AI-Collective' }}
+    if: github.repository_owner == 'OpenAccess-AI-Collective'
+    # this job needs to be run on self-hosted GPU runners...
    strategy:
      fail-fast: false
      matrix:
        include:
          - cuda: 118
            cuda_version: 11.8.0
-            python_version: "3.10"
-            pytorch: 2.1.2
+            python_version: "3.9"
+            pytorch: 2.0.1
+            axolotl_extras:
+          - cuda: 118
+            cuda_version: 11.8.0
+            python_version: "3.10"
+            pytorch: 2.0.1
            axolotl_extras:
-            axolotl_args: "--extra-index-url https://download.pytorch.org/whl/cu118"
            is_latest: true
+          - cuda: 118
+            cuda_version: 11.8.0
+            python_version: "3.10"
+            pytorch: 2.1.1
+            axolotl_extras:
          - cuda: 121
            cuda_version: 12.1.0
            python_version: "3.10"
-            pytorch: 2.1.2
+            pytorch: 2.1.1
            axolotl_extras:
-          - cuda: 121
-            cuda_version: 12.1.0
-            python_version: "3.11"
-            pytorch: 2.1.2
-            axolotl_extras:
-    runs-on: axolotl-gpu-runner
+    runs-on: [self-hosted, gpu, docker]
    steps:
      - name: Checkout
        uses: actions/checkout@v4
@@ -51,42 +55,57 @@ jobs:
        uses: docker/build-push-action@v5
        with:
          context: .
+          load: true
          build-args: |
            BASE_TAG=${{ github.ref_name }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}
            CUDA=${{ matrix.cuda }}
            PYTORCH_VERSION=${{ matrix.pytorch }}
-            AXOLOTL_ARGS=${{ matrix.axolotl_args }}
          file: ./docker/Dockerfile
-          push: ${{ github.event_name != 'pull_request' }}
          tags: |
            ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
            ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
          labels: ${{ steps.metadata.outputs.labels }}
+      - name: Unit Tests
+        run: |
+          docker run --rm ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }} pytest --ignore=tests/e2e/ /workspace/axolotl/tests/
+      - name: Push to Docker Hub
+        if: github.event_name != 'pull_request'
+        run: |
+          docker push ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+          latest_tag=${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
+          if [ -n "$latest_tag" ]; then
+            docker push "$latest_tag"
+          fi

  build-axolotl-runpod:
    needs: build-axolotl
-    if: ${{ ! contains(github.event.commits[0].message, '[skip docker]]') && github.repository_owner == 'OpenAccess-AI-Collective' }}
+    if: github.repository_owner == 'OpenAccess-AI-Collective'
    # this job needs to be run on self-hosted GPU runners...
    strategy:
      matrix:
        include:
+          - cuda: 118
+            cuda_version: 11.8.0
+            python_version: "3.9"
+            pytorch: 2.0.1
+            axolotl_extras:
          - cuda: 118
            cuda_version: 11.8.0
            python_version: "3.10"
-            pytorch: 2.1.2
+            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: 2.1.2
+            pytorch: 2.1.1
            axolotl_extras:
-          - cuda: 121
-            cuda_version: 12.1.0
-            python_version: "3.11"
-            pytorch: 2.1.2
-            axolotl_extras:
-    runs-on: axolotl-gpu-runner
+    runs-on: [self-hosted, gpu, docker]
    steps:
      - name: Checkout
        uses: actions/checkout@v4
--- a/.github/workflows/tests.yml
+++ b/.github/workflows/tests.yml
@@ -23,7 +23,7 @@ jobs:
      - uses: actions/checkout@v3
      - uses: actions/setup-python@v4
        with:
-          python-version: "3.10"
+          python-version: "3.9"
          cache: 'pip' # caching pip dependencies
      - uses: pre-commit/action@v3.0.0

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

    steps:
@@ -58,8 +58,8 @@ jobs:
  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, modal]
-    timeout-minutes: 60
+    runs-on: [self-hosted, gpu, docker]
+    timeout-minutes: 30
    needs: [pre-commit, pytest]

    strategy:
@@ -69,32 +69,40 @@ jobs:
          - cuda: 118
            cuda_version: 11.8.0
            python_version: "3.10"
-            pytorch: 2.1.2
-            axolotl_args: "--extra-index-url https://download.pytorch.org/whl/cu118"
-            num_gpus: 1
+            pytorch: 2.0.1
          - cuda: 121
            cuda_version: 12.1.0
            python_version: "3.10"
-            pytorch: 2.1.2
-            num_gpus: 1
+            pytorch: 2.1.1
    steps:
      - name: Checkout
        uses: actions/checkout@v4
-      - name: Install Python
-        uses: actions/setup-python@v5
+      - name: Docker metadata
+        id: metadata
+        uses: docker/metadata-action@v5
        with:
-          python-version: "3.10"
-      - name: Install Modal
+          images: winglian/axolotl-tests
+      - name: Build Docker image
        run: |
-          python -m pip install --upgrade pip
-          pip install modal jinja2
-      - name: Update env vars
+          # 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
        run: |
-          echo "BASE_TAG=main-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}" >> $GITHUB_ENV
-          echo "PYTORCH_VERSION=${{ matrix.pytorch}}" >> $GITHUB_ENV
-          echo "AXOLOTL_ARGS=${{ matrix.axolotl_args}}" >> $GITHUB_ENV
-          echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
-          echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
-      - name: Run tests job on Modal
+          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: |
-          modal run cicd.tests
+          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/
--- a/.gitignore
+++ b/.gitignore
@@ -167,8 +167,3 @@ cython_debug/
 # WandB
 # wandb creates a folder to store logs for training runs
 wandb
-
-# Runs
-lora-out/*
-qlora-out/*
-mlruns/*
--- a/.mypy.ini
+++ b/.mypy.ini
@@ -1,5 +1,5 @@
 [mypy]
-plugins = pydantic.mypy
+
 exclude = venv

 [mypy-alpaca_lora_4bit.*]
@@ -32,9 +32,6 @@ ignore_missing_imports = True
 [mypy-bitsandbytes]
 ignore_missing_imports = True

-[mypy-requests]
-ignore_missing_imports = True
-
 [mypy-datasets]
 ignore_missing_imports = True

--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -31,7 +31,6 @@ repos:
      additional_dependencies:
        [
            'types-PyYAML',
-            'pydantic>=2.5.3',
        ]
 -   repo: https://github.com/PyCQA/bandit
    rev: 1.7.5
--- a/README.md
+++ b/README.md
@@ -22,11 +22,11 @@ Features:
 - [Introduction](#axolotl)
 - [Supported Features](#axolotl-supports)
 - [Quickstart](#quickstart-)
- [Environment](#environment)
+- [Installation](#installation)
  - [Docker](#docker)
  - [Conda/Pip venv](#condapip-venv)
-  - [Cloud GPU](#cloud-gpu) - Latitude.sh, JarvisLabs, RunPod
-  - [Bare Metal Cloud GPU](#bare-metal-cloud-gpu)
+  - [Cloud GPU](#cloud-gpu) - Runpod, Latitude
+  - [LambdaLabs](#lambdalabs)
  - [Windows](#windows)
  - [Launching on public clouds via SkyPilot](#launching-on-public-clouds-via-skypilot)
 - [Dataset](#dataset)
@@ -34,12 +34,9 @@ Features:
  - [How to Use Custom Pretokenized Dataset](#how-to-use-your-custom-pretokenized-dataset)
 - [Config](#config)
  - [Train](#train)
-  - [Inference](#inference-playground)
+  - [Inference](#inference)
  - [Merge LORA to Base](#merge-lora-to-base)
  - [Special Tokens](#special-tokens)
- Advanced Topics
-  - [Multipack](./docs/multipack.md)<svg width="24" height="24" viewBox="0 0 24 24" xmlns="http://www.w3.org/2000/svg"><path d="M17 13.5v6H5v-12h6m3-3h6v6m0-6-9 9" class="icon_svg-stroke" stroke="#666" stroke-width="1.5" fill="none" fill-rule="evenodd" stroke-linecap="round" stroke-linejoin="round"></path></svg>
-  - [RLHF & DPO](./docs/rlhf.md)<svg width="24" height="24" viewBox="0 0 24 24" xmlns="http://www.w3.org/2000/svg"><path d="M17 13.5v6H5v-12h6m3-3h6v6m0-6-9 9" class="icon_svg-stroke" stroke="#666" stroke-width="1.5" fill="none" fill-rule="evenodd" stroke-linecap="round" stroke-linejoin="round"></path></svg>
 - [Common Errors](#common-errors-)
  - [Tokenization Mismatch b/w Training & Inference](#tokenization-mismatch-bw-inference--training)
 - [Debugging Axolotl](#debugging-axolotl)
@@ -87,17 +84,15 @@ Features:
 | phi         | ✅         | ✅    | ✅     | ❓             | ❓                 | ❓          | ❓            |
 | RWKV        | ✅         | ❓    | ❓     | ❓             | ❓                 | ❓          | ❓            |
 | Qwen        | ✅         | ✅    | ✅     | ❓             | ❓                 | ❓          | ❓            |
-| Gemma       | ✅         | ✅    | ✅     | ❓             | ❓                 | ✅          | ❓            |

-✅: supported
-❌: not supported
-❓: untested

 ## Quickstart ⚡

 Get started with Axolotl in just a few steps! This quickstart guide will walk you through setting up and running a basic fine-tuning task.

-**Requirements**: Python >=3.10 and Pytorch >=2.1.1.
+**Requirements**: Python >=3.9 and Pytorch >=2.0.
+
+`pip3 install "axolotl[flash-attn,deepspeed] @ git+https://github.com/OpenAccess-AI-Collective/axolotl"`

 ### For developers
 ```bash
@@ -105,23 +100,11 @@ git clone https://github.com/OpenAccess-AI-Collective/axolotl
 cd axolotl

 pip3 install packaging
-```
-
-General case:
-```
 pip3 install -e '.[flash-attn,deepspeed]'
 ```

-Mac: see https://github.com/OpenAccess-AI-Collective/axolotl/blob/13199f678b9aab39e92961323bdbce3234ee4b2b/docs/mac.md
-```
-pip3 install -e '.'
-```
-
 ### Usage
 ```bash
-# preprocess datasets - optional but recommended
-CUDA_VISIBLE_DEVICES="" python -m axolotl.cli.preprocess examples/openllama-3b/lora.yml
-
 # finetune lora
 accelerate launch -m axolotl.cli.train examples/openllama-3b/lora.yml

@@ -132,20 +115,15 @@ accelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \
 # gradio
 accelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \
    --lora_model_dir="./lora-out" --gradio
-
-# remote yaml files - the yaml config can be hosted on a public URL
-# Note: the yaml config must directly link to the **raw** yaml
-accelerate launch -m axolotl.cli.train https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/examples/openllama-3b/lora.yml
 ```

-## Advanced Setup
+## Installation

 ### Environment

 #### Docker
-
  ```bash
-  docker run --gpus '"all"' --rm -it winglian/axolotl:main-latest
+  docker run --gpus '"all"' --rm -it winglian/axolotl:main-py3.10-cu118-2.0.1
  ```

  Or run on the current files for development:
@@ -164,7 +142,7 @@ accelerate launch -m axolotl.cli.train https://raw.githubusercontent.com/OpenAcc
  A more powerful Docker command to run would be this:

  ```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 ${HOME}/.cache/huggingface:/root/.cache/huggingface winglian/axolotl:main-latest
+docker run --privileged --gpus '"all"' --shm-size 10g --rm -it --name axolotl --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 --mount type=bind,src="${PWD}",target=/workspace/axolotl -v ${HOME}/.cache/huggingface:/root/.cache/huggingface winglian/axolotl:main-py3.10-cu118-2.0.1
  ```

  It additionally:
@@ -179,7 +157,7 @@ docker run --privileged --gpus '"all"' --shm-size 10g --rm -it --name axolotl --
  </details>

 #### Conda/Pip venv
-  1. Install python >=**3.10**
+  1. Install python >=**3.9**

  2. Install pytorch stable https://pytorch.org/get-started/locally/

@@ -198,14 +176,9 @@ docker run --privileged --gpus '"all"' --shm-size 10g --rm -it --name axolotl --

 For cloud GPU providers that support docker images, use [`winglian/axolotl-cloud:main-latest`](https://hub.docker.com/r/winglian/axolotl-cloud/tags)

- on Latitude.sh use this [direct link](https://latitude.sh/blueprint/989e0e79-3bf6-41ea-a46b-1f246e309d5c)
- on JarvisLabs.ai use this [direct link](https://jarvislabs.ai/templates/axolotl)
 - on RunPod use this [direct link](https://runpod.io/gsc?template=v2ickqhz9s&ref=6i7fkpdz)

-#### Bare Metal Cloud GPU
-
-##### LambdaLabs
-
+#### LambdaLabs
  <details>

  <summary>Click to Expand</summary>
@@ -213,11 +186,11 @@ For cloud GPU providers that support docker images, use [`winglian/axolotl-cloud
  1. Install python
  ```bash
  sudo apt update
-  sudo apt install -y python3.10
+  sudo apt install -y python3.9

-  sudo update-alternatives --install /usr/bin/python python /usr/bin/python3.10 1
-  sudo update-alternatives --config python # pick 3.10 if given option
-  python -V # should be 3.10
+  sudo update-alternatives --install /usr/bin/python python /usr/bin/python3.9 1
+  sudo update-alternatives --config python # pick 3.9 if given option
+  python -V # should be 3.9

  ```

@@ -255,18 +228,15 @@ Please use WSL or Docker!

 #### Launching on public clouds via SkyPilot
 To launch on GPU instances (both on-demand and spot instances) on 7+ clouds (GCP, AWS, Azure, OCI, and more), you can use [SkyPilot](https://skypilot.readthedocs.io/en/latest/index.html):
-
 ```bash
 pip install "skypilot-nightly[gcp,aws,azure,oci,lambda,kubernetes,ibm,scp]"  # choose your clouds
 sky check
 ```
-
 Get the [example YAMLs](https://github.com/skypilot-org/skypilot/tree/master/llm/axolotl) of using Axolotl to finetune `mistralai/Mistral-7B-v0.1`:
 ```
 git clone https://github.com/skypilot-org/skypilot.git
 cd skypilot/llm/axolotl
 ```
-
 Use one command to launch:
 ```bash
 # On-demand
@@ -276,32 +246,31 @@ HF_TOKEN=xx sky launch axolotl.yaml --env HF_TOKEN
 HF_TOKEN=xx BUCKET=<unique-name> sky spot launch axolotl-spot.yaml --env HF_TOKEN --env BUCKET
 ```

+
 ### Dataset

 Axolotl supports a variety of dataset formats. Below are some of the formats you can use.
 Have dataset(s) in one of the following format (JSONL recommended):

-#### Pretraining
-
- `completion`: raw corpus
-  ```json
-  {"text": "..."}
-  ```
-
-Note: Axolotl usually loads the entire dataset into memory. This will be challenging for large datasets. Use the following config to enable streaming:
-
-```yaml
-pretraining_dataset: # hf path only
-```
-
-#### Supervised finetuning
-
-##### Instruction
-
 - `alpaca`: instruction; input(optional)
  ```json
  {"instruction": "...", "input": "...", "output": "..."}
  ```
+- `sharegpt`: conversations where `from` is `human`/`gpt`. (optional: `system` to override default system prompt)
+  ```json
+  {"conversations": [{"from": "...", "value": "..."}]}
+  ```
+- `llama-2`: the json is the same format as `sharegpt` above, with the following config (see the [config section](#config) for more details)
+    ```yml
+    datasets:
+      - path: <your-path>
+        type: sharegpt
+        conversation: llama-2
+    ```
+- `completion`: raw corpus
+  ```json
+  {"text": "..."}
+  ```

 <details>

@@ -379,37 +348,14 @@ pretraining_dataset: # hf path only
  ```json
  {"scores": "...", "critiques": "...", "instruction": "...", "answer": "...", "revision": "..."}
  ```
- `metharme`: instruction, adds additional eos tokens
-  ```json
-  {"prompt": "...", "generation": "..."}
-  ```
-
-</details>
-
-##### Template-Free
-
- `input_output`: template-free prompt construction
-  ```json
-   {"segments": [{"label": true|false, "text": "..."}]}
-  ```
-
-This is a special format that allows you to construct prompts without using templates. This is for advanced users who want more freedom with prompt construction.  See [these docs](docs/input_output.md) for more details.
-
-##### Conversation
-
- `sharegpt`: conversations where `from` is `human`/`gpt`. (optional: first row with role `system` to override default system prompt)
-  ```json
-  {"conversations": [{"from": "...", "value": "..."}]}
-  ```
-
-<details>
-
-<summary>See other formats</summary>
-
 - `pygmalion`: pygmalion
  ```json
  {"conversations": [{"role": "...", "value": "..."}]}
  ```
+- `metharme`: instruction, adds additional eos tokens
+  ```json
+  {"prompt": "...", "generation": "..."}
+  ```
 - `sharegpt.load_role`: conversations where `role` is used instead of `from`
  ```json
  {"conversations": [{"role": "...", "value": "..."}]}
@@ -425,8 +371,6 @@ This is a special format that allows you to construct prompts without using temp

 </details>

-Note: `type: sharegpt` opens a special config `conversation:` that enables conversions to many Conversation types. See dataset section under [all yaml options](#all-yaml-options).
-
 #### How to add custom prompts

 For a dataset that is preprocessed for instruction purposes:
@@ -448,16 +392,12 @@ datasets:
      format: "[INST] {instruction} [/INST]"
      no_input_format: "[INST] {instruction} [/INST]"
 ```
-See full config options under [all yaml options](#all-yaml-options).

 #### How to use your custom pretokenized dataset

 - Do not pass a `type:`
 - Columns in Dataset must be exactly `input_ids`, `attention_mask`, `labels`

-```yaml
- path: ...
-```

 ### Config

@@ -471,18 +411,22 @@ See [examples](examples) for quick start. It is recommended to duplicate and mod

 - dataset
  ```yaml
-  datasets:
-      # huggingface repo
-    - path: vicgalle/alpaca-gpt4
-      type: alpaca
+  sequence_len: 2048 # max token length for prompt

-      # huggingface repo with specific configuration/subset
+  # huggingface repo
+  datasets:
+    - path: vicgalle/alpaca-gpt4
+      type: alpaca # format from earlier
+
+  # huggingface repo with specific configuration/subset
+  datasets:
    - path: EleutherAI/pile
      name: enron_emails
      type: completion # format from earlier
      field: text # Optional[str] default: text, field to use for completion data

-      # huggingface repo with multiple named configurations/subsets
+  # huggingface repo with multiple named configurations/subsets
+  datasets:
    - path: bigcode/commitpackft
      name:
        - ruby
@@ -490,42 +434,39 @@ See [examples](examples) for quick start. It is recommended to duplicate and mod
        - typescript
      type: ... # unimplemented custom format

-      # fastchat conversation
-      # See 'conversation' options: https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py
+  # fastchat conversation
+  # See 'conversation' options: https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py
+  datasets:
    - path: ...
      type: sharegpt
-      conversation: chatml # default: vicuna_v1.1
+      conversation: chatml

-      # local
+  # local
+  datasets:
    - path: data.jsonl # or json
      ds_type: json # see other options below
      type: alpaca

-      # dataset with splits, but no train split
+  # dataset with splits, but no train split
+  dataset:
    - path: knowrohit07/know_sql
      type: context_qa.load_v2
      train_on_split: validation

-      # loading from s3 or gcs
-      # s3 creds will be loaded from the system default and gcs only supports public access
+  # loading from s3 or gcs
+  # s3 creds will be loaded from the system default and gcs only supports public access
+  dataset:
    - path: s3://path_to_ds # Accepts folder with arrow/parquet or file path like above. Supports s3, gcs.
      ...
-
-      # Loading Data From a Public URL
-      # - The file format is `json` (which includes `jsonl`) by default. For different formats, adjust the `ds_type` option accordingly.
-    - path: https://some.url.com/yourdata.jsonl # The URL should be a direct link to the file you wish to load. URLs must use HTTPS protocol, not HTTP.
-      ds_type: json # this is the default, see other options below.
  ```

 - loading
  ```yaml
  load_in_4bit: true
  load_in_8bit: true
-
-  bf16: auto # require >=ampere, auto will detect if your GPU supports this and choose automatically.
-  fp16: # leave empty to use fp16 when bf16 is 'auto'. set to false if you want to fallback to fp32
+  bf16: true # require >=ampere
+  fp16: true
  tf32: true # require >=ampere
-
  bfloat16: true # require >=ampere, use instead of bf16 when you don't want AMP (automatic mixed precision)
  float16: true # use instead of fp16 when you don't want AMP
  ```
@@ -533,7 +474,7 @@ See [examples](examples) for quick start. It is recommended to duplicate and mod

 - lora
  ```yaml
-  adapter: lora # 'qlora' or leave blank for full finetune
+  adapter: lora # qlora or leave blank for full finetune
  lora_r: 8
  lora_alpha: 16
  lora_dropout: 0.05
@@ -542,9 +483,9 @@ See [examples](examples) for quick start. It is recommended to duplicate and mod
    - v_proj
  ```

-<details id="all-yaml-options">
+<details>

-<summary>All yaml options (click to expand)</summary>
+<summary>All yaml options (click me)</summary>

 ```yaml
 # This is the huggingface model that contains *.pt, *.safetensors, or *.bin files
@@ -556,8 +497,8 @@ base_model_ignore_patterns:
 # You can set that here, or leave this empty to default to base_model
 base_model_config: ./llama-7b-hf
 # You can specify to choose a specific model revision from huggingface hub
-revision_of_model:
-# Optional tokenizer configuration path in case you want to use a different tokenizer
+model_revision:
+# Optional tokenizer configuration override in case you want to use a different tokenizer
 # than the one defined in the base model
 tokenizer_config:
 # If you want to specify the type of model to load, AutoModelForCausalLM is a good choice too
@@ -574,16 +515,15 @@ tokenizer_legacy:
 # This is reported to improve training speed on some models
 resize_token_embeddings_to_32x:

-# (Internal use only)
 # Used to identify which the model is based on
 is_falcon_derived_model:
 is_llama_derived_model:
-is_qwen_derived_model:
 # Please note that if you set this to true, `padding_side` will be set to "left" by default
 is_mistral_derived_model:
+is_qwen_derived_model:

 # optional overrides to the base model configuration
-overrides_of_model_config:
+model_config:
  # RoPE Scaling https://github.com/huggingface/transformers/pull/24653
  rope_scaling:
    type: # linear | dynamic
@@ -600,6 +540,8 @@ bnb_config_kwargs:

 # Whether you are training a 4-bit GPTQ quantized model
 gptq: true
+gptq_groupsize: 128 # group size
+gptq_model_v1: false # v1 or v2

 # This will attempt to quantize the model down to 8 bits and use adam 8 bit optimizer
 load_in_8bit: true
@@ -662,25 +604,12 @@ datasets:
      # For `completion` datsets only, uses the provided field instead of `text` column
      field:

-# A list of one or more datasets to eval the model with.
-# You can use either test_datasets, or val_set_size, but not both.
-test_datasets:
-  - path: /workspace/data/eval.jsonl
-    ds_type: json
-    # You need to specify a split. For "json" datasets the default split is called "train".
-    split: train
-    type: completion
-    data_files:
-      - /workspace/data/eval.jsonl
-
-# use RL training: 'dpo', 'ipo', 'kto_pair'
+# use RL training: dpo, ipo, kto_pair
 rl:

 # Saves the desired chat template to the tokenizer_config.json for easier inferencing
 # Currently supports chatml and inst (mistral/mixtral)
 chat_template: chatml
-# Changes the default system message
-default_system_message: You are a helpful assistant. Please give a long and detailed answer. # Currently only supports chatml.
 # Axolotl attempts to save the dataset as an arrow after packing the data together so
 # subsequent training attempts load faster, relative path
 dataset_prepared_path: data/last_run_prepared
@@ -689,11 +618,8 @@ push_dataset_to_hub: # repo path
 # The maximum number of processes to use while preprocessing your input dataset. This defaults to `os.cpu_count()`
 # if not set.
 dataset_processes: # defaults to os.cpu_count() if not set
-# Keep dataset in memory while preprocessing
-# Only needed if cached dataset is taking too much storage
-dataset_keep_in_memory:
 # push checkpoints to hub
-hub_model_id: # private repo path to push finetuned model
+hub_model_id: # repo path to push finetuned model
 # how to push checkpoints to hub
 # https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy
 hub_strategy:
@@ -713,6 +639,10 @@ sequence_len: 2048
 # Pad inputs so each step uses constant sized buffers
 # This will reduce memory fragmentation and may prevent OOMs, by re-using memory more efficiently
 pad_to_sequence_len:
+# Max sequence length to concatenate training samples together up to
+# Inspired by StackLLaMA. see https://huggingface.co/blog/stackllama#supervised-fine-tuning
+# FutureWarning: This will soon be DEPRECATED
+max_packed_sequence_len: 1024
 # Use efficient multi-packing with block diagonal attention and per sequence position_ids. Recommend set to 'true'
 sample_packing:
 # Set to 'false' if getting errors during eval with sample_packing on.
@@ -762,18 +692,10 @@ lora_modules_to_save:

 lora_fan_in_fan_out: false

-peft:
-  # Configuration options for loftq initialization for LoRA
-  # https://huggingface.co/docs/peft/developer_guides/quantization#loftq-initialization
-  loftq_config:
-    loftq_bits:  # typically 4 bits
-
 # ReLoRA configuration
 # Must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed
 relora_steps: # Number of steps per ReLoRA restart
 relora_warmup_steps: # Number of per-restart warmup steps
-relora_anneal_steps: # Number of anneal steps for each relora cycle
-relora_prune_ratio: # threshold for optimizer magnitude when pruning
 relora_cpu_offload: # True to perform lora weight merges on cpu during restarts, for modest gpu memory savings

 # wandb configuration if you're using it
@@ -789,7 +711,6 @@ wandb_log_model: # "checkpoint" to log model to wandb Artifacts every `save_step
 # mlflow configuration if you're using it
 mlflow_tracking_uri: # URI to mlflow
 mlflow_experiment_name: # Your experiment name
-hf_mlflow_log_artifacts:  # set to true to copy each saved checkpoint on each save to mlflow artifact registry

 # Where to save the full-finetuned model to
 output_dir: ./completed-model
@@ -823,8 +744,7 @@ save_total_limit: # Checkpoints saved at a time
 max_steps:

 eval_table_size: # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0
-eval_max_new_tokens: # Total number of tokens generated for predictions sent to wandb. Default is 128
-eval_causal_lm_metrics: # HF evaluate metrics used during evaluation. Default is ["sacrebleu", "comet", "ter", chrf]
+eval_table_max_new_tokens: # Total number of tokens generated for predictions sent to wandb. Default is 128

 loss_watchdog_threshold: # High loss value, indicating the learning has broken down (a good estimate is ~2 times the loss at the start of training)
 loss_watchdog_patience: # Number of high-loss steps in a row before the trainer aborts (default: 3)
@@ -853,11 +773,14 @@ early_stopping_patience: 3
 lr_scheduler: # 'one_cycle' | 'log_sweep' | empty for cosine
 lr_scheduler_kwargs:
 cosine_min_lr_ratio: # decay lr to some percentage of the peak lr, e.g. cosine_min_lr_ratio=0.1 for 10% of peak lr
-cosine_constant_lr_ratio: # freeze lr at some percentage of the step, e.g. cosine_constant_lr_ratio=0.8 means start cosine_min_lr at 80% of training step (https://arxiv.org/pdf/2308.04014.pdf)

 # For one_cycle optim
 lr_div_factor: # Learning rate div factor

+# For log_sweep optim
+log_sweep_min_lr:
+log_sweep_max_lr:
+
 # Specify optimizer
 # Valid values are driven by the Transformers OptimizerNames class, see:
 # https://github.com/huggingface/transformers/blob/95b374952dc27d8511541d6f5a4e22c9ec11fb24/src/transformers/training_args.py#L134
@@ -911,8 +834,7 @@ flash_attn_fuse_mlp: # Whether to fuse part of the MLP into a single operation
 # Whether to use scaled-dot-product attention
 # https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html
 sdp_attention:
-# Shifted-sparse attention (only llama) - https://arxiv.org/pdf/2309.12307.pdf
-s2_attention:
+
 # Resume from a specific checkpoint dir
 resume_from_checkpoint:
 # If resume_from_checkpoint isn't set and you simply want it to start where it left off.
@@ -936,7 +858,7 @@ tokens:
 fsdp:
 fsdp_config:

-# Deepspeed config path. e.g., deepspeed_configs/zero3.json
+# Deepspeed config path. e.g., deepspeed/zero3.json
 deepspeed:

 # Advanced DDP Arguments
@@ -1029,9 +951,6 @@ Run
 accelerate launch -m axolotl.cli.train your_config.yml
 ```

-> [!TIP]
-> You can also reference a config file that is hosted on a public URL, for example `accelerate launch -m axolotl.cli.train https://yourdomain.com/your_config.yml`
-
 #### Preprocess dataset

 You can optionally pre-tokenize dataset with the following before finetuning.
@@ -1060,11 +979,11 @@ for deepspeed is available at https://huggingface.co/docs/accelerate/main/en/usa
 We provide several default deepspeed JSON configurations for ZeRO stage 1, 2, and 3.

 ```yaml
-deepspeed: deepspeed_configs/zero1.json
+deepspeed: deepspeed/zero1.json
 ```

 ```shell
-accelerate launch -m axolotl.cli.train examples/llama-2/config.py --deepspeed deepspeed_configs/zero1.json
+accelerate launch -m axolotl.cli.train examples/llama-2/config.py --deepspeed deepspeed/zero1.json
 ```

 ##### FSDP
@@ -1080,10 +999,6 @@ fsdp_config:
  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
 ```

-##### FSDP + QLoRA
-
-Axolotl supports training with FSDP and QLoRA, see [these docs](docs/fsdp_qlora.md) for more information.
-
 ##### Weights & Biases Logging

 Make sure your `WANDB_API_KEY` environment variable is set (recommended) or you login to wandb with `wandb login`.
@@ -1145,7 +1060,7 @@ Please use `--sample_packing False` if you have it on and receive the error simi

 ### Merge LORA to base

-The following command will merge your LORA adapater with your base model. You can optionally pass the argument `--lora_model_dir` to specify the directory where your LORA adapter was saved, otherwhise, this will be inferred from `output_dir` in your axolotl config file.  The merged model is saved in the sub-directory `{lora_model_dir}/merged`.
+The following command will merge your LORA adapater with your base model.  You can optionally pass the argument `--lora_model_dir` to specify the directory where your LORA adapter was saved, otherwhise, this will be inferred from `output_dir` in your axolotl config file.  The merged model is saved in the sub-directory `{lora_model_dir}/merged`.

 ```bash
 python3 -m axolotl.cli.merge_lora your_config.yml --lora_model_dir="./completed-model"
@@ -1206,8 +1121,8 @@ If you decode a prompt constructed by axolotl, you might see spaces between toke

 1. Materialize some data using `python -m axolotl.cli.preprocess your_config.yml --debug`, and then decode the first few rows with your model's tokenizer.
 2. During inference, right before you pass a tensor of token ids to your model, decode these tokens back into a string.
-3. Make sure the inference string from #2 looks **exactly** like the data you fine tuned on from #1, including spaces and new lines.  If they aren't the same, adjust your inference server accordingly.
-4. As an additional troubleshooting step, you can look at the token ids between 1 and 2 to make sure they are identical.
+3. Make sure the inference string from #2 looks **exactly** like the data you fine tuned on from #1, including spaces and new lines.  If they aren't the same adjust your inference server accordingly.
+4. As an additional troubleshooting step, you can look look at the token ids between 1 and 2 to make sure they are identical.

 Having misalignment between your prompts during training and inference can cause models to perform very poorly, so it is worth checking this.  See [this blog post](https://hamel.dev/notes/llm/05_tokenizer_gotchas.html) for a concrete example.

@@ -1215,11 +1130,9 @@ Having misalignment between your prompts during training and inference can cause

 See [this debugging guide](docs/debugging.md) for tips on debugging Axolotl, along with an example configuration for debugging with VSCode.

-## Need help? 🙋
+## Need help? 🙋♂️

-Join our [Discord server](https://discord.gg/HhrNrHJPRb) where we our community members can help you.
-
-Need dedicated support? Please contact us at [✉️wing@openaccessaicollective.org](mailto:wing@openaccessaicollective.org) for dedicated support options.
+Join our [Discord server](https://discord.gg/HhrNrHJPRb) where we can help you

 ## Badge ❤🏷️

@@ -1236,7 +1149,7 @@ Building something cool with Axolotl? Consider adding a badge to your model card
 Check out some of the projects and models that have been built using Axolotl! Have a model you'd like to add to our Community Showcase? Open a PR with your model.

 Open Access AI Collective
- [Minotaur 13b](https://huggingface.co/openaccess-ai-collective/minotaur-13b-fixed)
+- [Minotaur 13b](https://huggingface.co/openaccess-ai-collective/minotaur-13b)
 - [Manticore 13b](https://huggingface.co/openaccess-ai-collective/manticore-13b)
 - [Hippogriff 30b](https://huggingface.co/openaccess-ai-collective/hippogriff-30b-chat)

@@ -1253,28 +1166,13 @@ PRs are **greatly welcome**!

 Please run below to setup env
 ```bash
-git clone https://github.com/OpenAccess-AI-Collective/axolotl
-cd axolotl
-
-pip3 install packaging
-pip3 install -e '.[flash-attn,deepspeed]'
-
 pip3 install -r requirements-dev.txt -r requirements-tests.txt
 pre-commit install

 # test
 pytest tests/
-
-# optional: run against all files
-pre-commit run --all-files
 ```

-Thanks to all of our contributors to date. Help drive open source AI progress forward by contributing to Axolotl.
-
-<a href="https://github.com/openaccess-ai-collective/axolotl/graphs/contributors">
-  <img src="https://contrib.rocks/image?repo=openaccess-ai-collective/axolotl" alt="contributor chart by https://contrib.rocks"/>
-</a>
-
 ## Sponsors 🤝❤

 OpenAccess AI Collective is run by volunteer contributors such as [winglian](https://github.com/winglian),
@@ -1303,6 +1201,4 @@ consider sponsoring the project via [GitHub Sponsors](https://github.com/sponsor

 #### 🥉 Bronze Sponsors - $500/mo

- - [JarvisLabs.ai](https://jarvislabs.ai)
-
 ---
--- a/cicd/Dockerfile.jinja
+++ b/cicd/Dockerfile.jinja
@@ -1,39 +0,0 @@
-FROM winglian/axolotl-base:{{ BASE_TAG }}
-
-ENV TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
-ENV AXOLOTL_EXTRAS="{{ AXOLOTL_EXTRAS }}"
-ENV AXOLOTL_ARGS="{{ AXOLOTL_ARGS }}"
-ENV CUDA="{{ CUDA }}"
-ENV BNB_CUDA_VERSION="{{ CUDA }}"
-ENV PYTORCH_VERSION="{{ PYTORCH_VERSION }}"
-ENV GITHUB_REF="{{ GITHUB_REF }}"
-ENV GITHUB_SHA="{{ GITHUB_SHA }}"
-
-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] $AXOLOTL_ARGS; \
-    else \
-        pip install -e .[deepspeed,flash-attn,mamba-ssm] $AXOLOTL_ARGS; \
-    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/cicd/cicd.sh
+++ b/cicd/cicd.sh
@@ -1,5 +0,0 @@
-#!/bin/bash
-
-pytest --ignore=tests/e2e/ /workspace/axolotl/tests/
-pytest /workspace/axolotl/tests/e2e/patched/
-pytest --ignore=tests/e2e/patched/ /workspace/axolotl/tests/e2e/
--- a/cicd/tests.py
+++ b/cicd/tests.py
@@ -1,75 +0,0 @@
-"""
- modal application to run axolotl gpu tests in Modal
- """
-import os
-import pathlib
-import tempfile
-
-import jinja2
-import modal
-from jinja2 import select_autoescape
-from modal import Image, Stub
-
-cicd_path = pathlib.Path(__file__).parent.resolve()
-
-template_loader = jinja2.FileSystemLoader(searchpath=cicd_path)
-template_env = jinja2.Environment(
-    loader=template_loader, autoescape=select_autoescape()
-)
-df_template = template_env.get_template("Dockerfile.jinja")
-
-df_args = {
-    "AXOLOTL_EXTRAS": os.environ.get("AXOLOTL_EXTRAS", ""),
-    "AXOLOTL_ARGS": os.environ.get("AXOLOTL_ARGS", ""),
-    "PYTORCH_VERSION": os.environ.get("PYTORCH_VERSION", "2.0.1"),
-    "BASE_TAG": os.environ.get("BASE_TAG", "main-base-py3.10-cu118-2.0.1"),
-    "CUDA": os.environ.get("CUDA", "118"),
-    "GITHUB_REF": os.environ.get("GITHUB_REF", "refs/heads/main"),
-    "GITHUB_SHA": os.environ.get("GITHUB_SHA", ""),
-}
-
-dockerfile_contents = df_template.render(**df_args)
-
-temp_dir = tempfile.mkdtemp()
-with open(pathlib.Path(temp_dir) / "Dockerfile", "w", encoding="utf-8") as f:
-    f.write(dockerfile_contents)
-
-cicd_image = (
-    Image.from_dockerfile(
-        pathlib.Path(temp_dir) / "Dockerfile",
-        force_build=True,
-        gpu="A10G",
-    )
-    .env(df_args)
-    .pip_install("fastapi==0.110.0", "pydantic==2.6.3")
-)
-
-stub = Stub("Axolotl CI/CD", secrets=[])
-
-
-N_GPUS = int(os.environ.get("N_GPUS", 1))
-GPU_CONFIG = modal.gpu.A10G(count=N_GPUS)
-
-
-def run_cmd(cmd: str, run_folder: str):
-    import subprocess  # nosec
-
-    # Propagate errors from subprocess.
-    if exit_code := subprocess.call(cmd.split(), cwd=run_folder):  # nosec
-        exit(exit_code)  # pylint: disable=consider-using-sys-exit
-
-
-@stub.function(
-    image=cicd_image,
-    gpu=GPU_CONFIG,
-    timeout=45 * 60,
-    cpu=8.0,
-    memory=131072,
-)
-def cicd_pytest():
-    run_cmd("./cicd/cicd.sh", "/workspace/axolotl")
-
-
-@stub.local_entrypoint()
-def main():
-    cicd_pytest.remote()
--- a/deepspeed_configs/zero1.json
+++ b/deepspeed_configs/zero1.json
@@ -15,8 +15,16 @@
    "hysteresis": 2,
    "min_loss_scale": 1
  },
+  "optimizer": {
+    "type": "AdamW",
+    "params": {
+      "lr": "auto",
+      "betas": "auto",
+      "eps": "auto",
+      "weight_decay": "auto"
+    }
+  },
  "gradient_accumulation_steps": "auto",
-  "gradient_clipping": "auto",
  "train_batch_size": "auto",
  "train_micro_batch_size_per_gpu": "auto",
  "wall_clock_breakdown": false
--- a/deepspeed_configs/zero2.json
+++ b/deepspeed_configs/zero2.json
@@ -19,8 +19,16 @@
    "hysteresis": 2,
    "min_loss_scale": 1
  },
+  "optimizer": {
+    "type": "AdamW",
+    "params": {
+      "lr": "auto",
+      "betas": "auto",
+      "eps": "auto",
+      "weight_decay": "auto"
+    }
+  },
  "gradient_accumulation_steps": "auto",
-  "gradient_clipping": "auto",
  "train_batch_size": "auto",
  "train_micro_batch_size_per_gpu": "auto",
  "wall_clock_breakdown": false
--- a/deepspeed_configs/zero3.json
+++ b/deepspeed_configs/zero3.json
@@ -23,8 +23,16 @@
    "hysteresis": 2,
    "min_loss_scale": 1
  },
+  "optimizer": {
+    "type": "AdamW",
+    "params": {
+      "lr": "auto",
+      "betas": "auto",
+      "eps": "auto",
+      "weight_decay": "auto"
+    }
+  },
  "gradient_accumulation_steps": "auto",
-  "gradient_clipping": "auto",
  "train_batch_size": "auto",
  "train_micro_batch_size_per_gpu": "auto",
  "wall_clock_breakdown": false
--- a/deepspeed_configs/zero3_bf16.json
+++ b/deepspeed_configs/zero3_bf16.json
@@ -23,8 +23,16 @@
    "hysteresis": 2,
    "min_loss_scale": 1
  },
+  "optimizer": {
+    "type": "AdamW",
+    "params": {
+      "lr": "auto",
+      "betas": "auto",
+      "eps": "auto",
+      "weight_decay": "auto"
+    }
+  },
  "gradient_accumulation_steps": "auto",
-  "gradient_clipping": "auto",
  "train_batch_size": "auto",
  "train_micro_batch_size_per_gpu": "auto",
  "wall_clock_breakdown": false
--- a/devtools/dev_sharegpt.yml
+++ b/devtools/dev_sharegpt.yml
@@ -2,6 +2,7 @@
 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
--- a/docker/Dockerfile
+++ b/docker/Dockerfile
@@ -3,10 +3,9 @@ FROM winglian/axolotl-base:$BASE_TAG

 ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
 ARG AXOLOTL_EXTRAS=""
-ARG AXOLOTL_ARGS=""
 ARG CUDA="118"
 ENV BNB_CUDA_VERSION=$CUDA
-ARG PYTORCH_VERSION="2.1.2"
+ARG PYTORCH_VERSION="2.0.1"

 ENV PYTORCH_VERSION=$PYTORCH_VERSION

@@ -21,9 +20,9 @@ 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] $AXOLOTL_ARGS; \
+        pip install -e .[deepspeed,flash-attn,mamba-ssm,$AXOLOTL_EXTRAS]; \
    else \
-        pip install -e .[deepspeed,flash-attn,mamba-ssm] $AXOLOTL_ARGS; \
+        pip install -e .[deepspeed,flash-attn,mamba-ssm]; \
    fi

 # So we can test the Docker image
--- a/docker/Dockerfile-base
+++ b/docker/Dockerfile-base
@@ -7,8 +7,8 @@ FROM nvidia/cuda:$CUDA_VERSION-cudnn$CUDNN_VERSION-devel-ubuntu$UBUNTU_VERSION a

 ENV PATH="/root/miniconda3/bin:${PATH}"

-ARG PYTHON_VERSION="3.10"
-ARG PYTORCH_VERSION="2.1.2"
+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"

@@ -29,7 +29,7 @@ ENV PATH="/root/miniconda3/envs/py${PYTHON_VERSION}/bin:${PATH}"
 WORKDIR /workspace

 RUN python3 -m pip install --upgrade pip && pip3 install packaging && \
-    python3 -m pip install --no-cache-dir -U torch==${PYTORCH_VERSION}+cu${CUDA} --extra-index-url https://download.pytorch.org/whl/cu$CUDA
+    python3 -m pip install --no-cache-dir -U torch==${PYTORCH_VERSION}+cu${CUDA} deepspeed-kernels --extra-index-url https://download.pytorch.org/whl/cu$CUDA

 RUN git lfs install --skip-repo && \
    pip3 install awscli && \
--- a/docker/Dockerfile-cloud
+++ b/docker/Dockerfile-cloud
@@ -7,19 +7,14 @@ ENV TRANSFORMERS_CACHE="/workspace/data/huggingface-cache/hub"
 ENV HF_HOME="/workspace/data/huggingface-cache/hub"
 ENV HF_HUB_ENABLE_HF_TRANSFER="1"

-EXPOSE 8888
-EXPOSE 22
-
 COPY scripts/cloud-entrypoint.sh /root/cloud-entrypoint.sh
-COPY scripts/motd /etc/motd

-RUN pip install jupyterlab notebook ipywidgets && \
+RUN pip install jupyterlab notebook && \
    jupyter lab clean
 RUN apt install --yes --no-install-recommends openssh-server tmux && \
    mkdir -p ~/.ssh && \
    chmod 700 ~/.ssh && \
    printf "\n[[ -z \"\$TMUX\"  ]] && { tmux attach-session -t ssh_tmux || tmux new-session -s ssh_tmux; exit; }\n" >> ~/.bashrc && \
-    printf "[ ! -z \"\$TERM\" -a -r /etc/motd ] && cat /etc/motd\n" >> ~/.bashrc && \
    chmod +x /workspace/axolotl/scripts/cloud-entrypoint.sh && \
    chmod +x /root/cloud-entrypoint.sh

--- a/docker/Dockerfile-tests
+++ b/docker/Dockerfile-tests
@@ -3,10 +3,9 @@ FROM winglian/axolotl-base:$BASE_TAG

 ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
 ARG AXOLOTL_EXTRAS=""
-ARG AXOLOTL_ARGS=""
 ARG CUDA="118"
 ENV BNB_CUDA_VERSION=$CUDA
-ARG PYTORCH_VERSION="2.1.2"
+ARG PYTORCH_VERSION="2.0.1"
 ARG GITHUB_REF="main"

 ENV PYTORCH_VERSION=$PYTORCH_VERSION
@@ -25,9 +24,9 @@ RUN git fetch origin +$GITHUB_REF && \

 # If AXOLOTL_EXTRAS is set, append it in brackets
 RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
-        pip install -e .[deepspeed,flash-attn,mamba-ssm,$AXOLOTL_EXTRAS] $AXOLOTL_ARGS; \
+        pip install -e .[deepspeed,flash-attn,mamba-ssm,$AXOLOTL_EXTRAS]; \
    else \
-        pip install -e .[deepspeed,flash-attn,mamba-ssm] $AXOLOTL_ARGS; \
+        pip install -e .[deepspeed,flash-attn,mamba-ssm]; \
    fi

 # So we can test the Docker image
--- a/docs/debugging.md
+++ b/docs/debugging.md
@@ -74,6 +74,7 @@ pip3 install -e '.[flash-attn,deepspeed]'

 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

--- a/docs/fsdp_qlora.md
+++ b/docs/fsdp_qlora.md
@@ -1,37 +0,0 @@
-# FDSP + QLoRA
-
-## Background
-
-Using FSDP with QLoRA is essential for **fine-tuning larger (70b+ parameter) LLMs on consumer GPUs.**  For example, you can use FSDP + QLoRA to train a 70b model on two 24GB GPUs[^1].
-
-Below, we describe how to use this feature in Axolotl.
-
-## Usage
-
-To enable `QLoRA` with `FSDP`, you need to perform the following steps:
-
-> ![Tip]
-> See the [example config](#example-config) file in addition to reading these instructions.
-
-1. Set `adapter: qlora` in your axolotl config file.
-2. Enable FSDP in your axolotl config, as [described here](https://github.com/OpenAccess-AI-Collective/axolotl?tab=readme-ov-file#fsdp).
-3. Use one of the supported model types: `llama`, `mistral` or `mixtral`.
-
-## Example Config
-
-[examples/llama-2/qlora-fsdp.yml](../examples/llama-2/qlora-fsdp.yml) contains an example of how to enable QLoRA + FSDP in axolotl.
-
-## References
-
- [PR #1378](https://github.com/OpenAccess-AI-Collective/axolotl/pull/1378) enabling QLoRA in FSDP in Axolotl.
- [Blog Post](https://www.answer.ai/posts/2024-03-06-fsdp-qlora.html) from the [Answer.AI](https://www.answer.ai/) team describing the work that enabled QLoRA in FSDP.
- Related HuggingFace PRs Enabling FDSP + QLoRA:
-    - Accelerate [PR#2544](https://github.com/huggingface/accelerate/pull/2544 )
-    - Transformers [PR#29587](https://github.com/huggingface/transformers/pull/29587)
-    - TRL [PR#1416](https://github.com/huggingface/trl/pull/1416)
-    - PEFT [PR#1550](https://github.com/huggingface/peft/pull/1550)
-
-
-
-
-[^1]: This was enabled by [this work](https://www.answer.ai/posts/2024-03-06-fsdp-qlora.html) from the Answer.AI team.
--- a/docs/images/4d-mask.png
+++ b/docs/images/4d-mask.png
--- a/docs/input_output.md
+++ b/docs/input_output.md
@@ -1,260 +0,0 @@
-# Template-free prompt construction with the `input_output` format
-
-<!-- TOC -->
-
- [Background](#background)
-    - [Masking Inputs](#masking-inputs)
-    - [You may not want prompt templates](#you-may-not-want-prompt-templates)
-    - [The `input_output` format](#the-input_output-format)
- [Usage](#usage)
-    - [1. Prepare Data](#1-prepare-data)
-    - [2. Use `type: input_output`](#2-use-type-input_output)
-    - [3. Check the prompts](#3-check-the-prompts)
-
-<!-- /TOC -->
-
-<a id="markdown-background" name="background"></a>
-
-## Background
-
-<a id="markdown-masking-inputs" name="masking-inputs"></a>
-
-### Masking Inputs
-
-One of the most popular features of
-[axolotl](https://github.com/OpenAccess-AI-Collective/axolotl) is
-setting the following configuration value:
-
-
-```yaml
-train_on_inputs: false
-```
-
-If you declare a [dataset formats](https://github.com/OpenAccess-AI-Collective/axolotl?tab=readme-ov-file#dataset)
-such as `alpaca` or `chatml`, axolotl knows what is an input
-(i.e. human) vs. an output (i.e. the assistant) and masks the input
-labels so that your model can focus on predicting the outputs only.
-
-<a id="markdown-you-may-not-want-prompt-templates" name="you-may-not-want-prompt-templates"></a>
-
-### You may not want prompt templates
-
-However, there are many situations where you don't want to use one of
-these formats or templates (I usually don't!). This is because they can:
-
-   Add unnecessary boilerplate to your prompts.
-   Create artifacts like special delimiters `<|im_start|>` that can
-    quickly become footguns if you don't include them correctly at
-    inference time.
-   Enforce a *chat* interface when you do not want one. Sometimes you
-    just want to fine-tune a model to a very specific task and do NOT
-    want multi-turn conversations, roles, etc.
-   Limit you to only certain roles that the template allows.
-
-<a id="markdown-the-inputoutput-format" name="the-inputoutput-format"></a>
-
-### The `input_output` format
-
-You can construct your prompts without a template by using the
-`input_output` format, by setting `type: input_output` in your
-configuration file like this:
-
-**config.yml**
-
-```yaml
-train_on_inputs: false # Mask segments of your data
-datasets:
-  - path: output.jsonl
-    type: input_output  # use template free prompt construction
-```
-
-Unlike `type: completion`, which is also template-free,
-`type: input_output` allows you to mask segments of your text. More
-details on how this works are described below.
-
-<a id="markdown-usage" name="usage"></a>
-
-## Usage
-
-This is how you can use the `input_output` format:
-
-<a id="markdown-1-prepare-data" name="1-prepare-data"></a>
-
-### 1. Prepare Data
-
-To use the `input_output` format, collect your data in the following
-format into a jsonl file (below is the first row from the file
-`output`.jsonl` pretty printed):
-
-```bash
-$ head -n1 output.jsonl | python -m json.tool
-
-{.cell-output .cell-output-stdout}
-    {
-        "segments": [
-            {
-                "label": true,
-                "text": "<s>Hello\n"
-            },
-            {
-                "label": true,
-                "text": "hi there!. "
-            },
-            {
-                "label": false,
-                "text": "goodbye "
-            },
-            {
-                "label": true,
-                "text": "farewell</s>"
-            }
-        ]
-    }
-```
-
-Set `label:false` when you want to mask a segment of text so that the
-model isn't trained on it. Some things to keep in mind:
-
-> [!IMPORTANT]
-> 1.  **EOS, BOS, spaces, newlines etc. are entirely up to you. Axolotl
-    concatenates all the segments as-is.** The tokenizer doesn't add
-    anything additional. Notice how I added spaces, newlines, `<s>`
-    (BOS), and `</s>` (EOS) myself.
-> 2.  Make sure you check the materialized output to validate that the
-    prompt is getting assembled how you like.
-
-<a id="markdown-2-use-type-inputoutput" name="2-use-type-inputoutput"></a>
-
-### 2. Use `type: input_output`
-
-Let's materialize data with our `output.jsonl` file by setting
-`type: input_output` in our axolotl config:
-
-```yaml
-# training_config.yaml
-base_model: mistralai/Mistral-7B-v0.1
-data_seed: 49
-seed: 49
-
-datasets:
-  - path: output.jsonl
-    type: input_output
-val_set_size: 0.1
-
-sequence_len: 896
-sample_packing: false
-
-micro_batch_size: 2
-gradient_accumulation_steps: 3
-eval_batch_size: 2
-num_epochs: 1
-learning_rate: 0.0002
-
-train_on_inputs: false
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"
-```
-
-You can use the following command to materialize your data. The
-`--debug` flag will print the tokens, along with the labels so you can
-verify that the correct items are being ignored:
-
-```bash
-$ python -m axolotl.cli.preprocess training_config.yaml --debug
-
-...
-[2024-03-05 23:36:46,969] [INFO] [axolotl.check_example_labels:35] [PID:607731] [RANK:0] <s>(1, 1) Hello(22557, 22557)
-(13, 13) hi(12014, 12014) there(736, 736) !(28808, 28808) .(28723, 28723) (28705, 28705) good(-100, 1179) bye(-100, 17664) (-100, 28705) fare(19111, 19111) well(5458, 5458) </s>(2, 2)
-
-```
-
-The format is `decoded_token`(`label`, `token_id`), for example,
-`<s>(1, 1)` means that the token is `<s>`, the label is `1` and the
-token_id is `1`. When the label is `-100` then that token is ignored for
-training.
-
-<a id="markdown-3-check-the-prompts" name="3-check-the-prompts"></a>
-
-### 3. Check the prompts
-
-Here is another way to check the materialized output:
-
-```python
-from transformers import AutoTokenizer
-from datasets import load_from_disk
-import yaml
-
-directory = !ls last_run_prepared/
-with open('training_config.yaml', 'r') as f:
-    cfg = yaml.safe_load(f)
-model_id = cfg['base_model']
-tok = AutoTokenizer.from_pretrained(model_id)
-ds = load_from_disk(f'last_run_prepared/{directory[0]}/')
-```
-
-```python
->>> row = ds[0]
->>> print(tok.decode(row['input_ids']))
-<s> Hello
-    hi there!.  goodbye  farewell</s>
-```
-
-We can check that the right tokens are ingored by comparing the labels
-to each token:
-
-```python
-import pandas as pd
-pd.DataFrame([{'token': tok.decode(i), 'label': l, 'id':i} for i,l in
-              zip(row['input_ids'], row['labels'])])
-```
-
-| token | label | id    |
-|-------|-------|-------|
-| 0     | \<s\> | 1     |
-| 1     | Hello | 22557 |
-| 2     | \\n   | 13    |
-| 3     | hi    | 12014 |
-| 4     | there | 736   |
-| 5     | !     | 28808 |
-| 6     | .     | 28723 |
-| 7     |       | 28705 |
-| 8     | good  | -100  |
-| 9     | bye   | -100  |
-| 10    |       | -100  |
-| 11    | fare  | 19111 |
-| 12    | well  | 5458  |
-| 13    | \</s\>| 2     |
-
-
-
-If we look at the input data, the above table seems correct! (The jsonl
-version is repeated below for reference):
-
-
-```bash
-$ head -n1 output.jsonl | python -m json.tool
-
-{.cell-output .cell-output-stdout}
-    {
-        "segments": [
-            {
-                "label": true,
-                "text": "<s>Hello\n"
-            },
-            {
-                "label": true,
-                "text": "hi there!. "
-            },
-            {
-                "label": false,
-                "text": "goodbye "
-            },
-            {
-                "label": true,
-                "text": "farewell</s>"
-            }
-        ]
-    }
-```
--- a/docs/mac.md
+++ b/docs/mac.md
@@ -1,18 +0,0 @@
-# Mac M series support
-
-Currently Axolotl on Mac is partially usable, many of the dependencies of Axolotl including Pytorch do not support MPS or have incomplete support.
-
-Current support:
- [x] Support for all models
- [x] Full training of models
- [x] LoRA training
- [x] Sample packing
- [ ] FP16 and BF16 (awaiting AMP support for MPS in Pytorch)
- [ ] Tri-dao's flash-attn (until it is supported use spd_attention as an alternative)
- [ ] xformers
- [ ] bitsandbytes (meaning no 4/8 bits loading and bnb optimizers)
- [ ] qlora
- [ ] DeepSpeed
-
-Untested:
- FSDP
--- a/docs/multipack.md
+++ b/docs/multipack.md
@@ -1,11 +1,4 @@
-# Multipack (Sample Packing)
-
-## Visualization of Multipack with Flash Attention
-
-Because Flash Attention simply drops the attention mask, we do not need to
-construct a 4d attention mask. We only need to concatenate the sequences into
-a single batch and let flash attention know where each new sequence begins.
-
+# Multipack

 4k context, bsz =4,
 each character represents 256 tokens
@@ -56,18 +49,3 @@ w packing ( note it's the same effective number of tokens per step, but a true b
   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 ]]
 ```
-
-cu_seqlens:
-[[ 0, 11, 17, 24, 28, 36, 41 44, 48, 51, 55, 60, 64]]
-
-
-## Multipack without Flash Attention
-
-Multipack can still be achieved without Flash attention, but with lower packing
-efficiency as we are not able to join multiple batches into a single batch due to
-context length limits without flash attention. We can use either Pytorch's Scaled
-Dot Product Attention implementation or native Pytorch attention implementation
-along with [4d attention masks](https://github.com/huggingface/transformers/pull/27539)
-to pack sequences together and avoid cross attention.
-
-<img src="./images/4d-mask.png" alt="axolotl" width="800">
--- a/docs/rlhf.md
+++ b/docs/rlhf.md
@@ -12,21 +12,21 @@ feedback. Various methods include, but not limited to:

 ### 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.
+[!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: dpo
+rl: true
 datasets:
  - path: Intel/orca_dpo_pairs
    split: train
-    type: chatml.intel
+    type: intel_apply_chatml
  - path: argilla/ultrafeedback-binarized-preferences
    split: train
-    type: chatml.argilla
+    type: argilla_apply_chatml
 ```

 #### IPO
@@ -34,16 +34,6 @@ datasets:
 rl: ipo
 ```

-#### Using local dataset files
-```yaml
-datasets:
-  - ds_type: json
-    data_files:
-      - orca_rlhf.jsonl
-    split: train
-    type: chatml.intel
-```
-
 #### 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.
--- a/examples/cerebras/btlm-ft.yml
+++ b/examples/cerebras/btlm-ft.yml
@@ -53,8 +53,8 @@ lr_quadratic_warmup: true
 learning_rate: 0.000085
 train_on_inputs: true
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: true

 gradient_checkpointing: false
--- a/examples/cerebras/qlora.yml
+++ b/examples/cerebras/qlora.yml
@@ -11,6 +11,7 @@ val_set_size: 0.05
 adapter: qlora
 lora_model_dir:
 sequence_len: 2048
+max_packed_sequence_len: 2048
 lora_r: 16
 lora_alpha: 32
 lora_dropout: 0.05
@@ -35,8 +36,8 @@ lr_scheduler: cosine
 learning_rate: 0.0002
 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: true
 gradient_checkpointing: true
 early_stopping_patience:
--- a/examples/code-llama/13b/lora.yml
+++ b/examples/code-llama/13b/lora.yml
@@ -1,6 +1,7 @@
 base_model: codellama/CodeLlama-13b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
+is_llama_derived_model: true

 load_in_8bit: true
 load_in_4bit: false
@@ -40,8 +41,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
@@ -51,7 +52,6 @@ local_rank:
 logging_steps: 1
 xformers_attention:
 flash_attention: true
-s2_attention:

 warmup_steps: 10
 evals_per_epoch: 4
--- a/examples/code-llama/13b/qlora.yml
+++ b/examples/code-llama/13b/qlora.yml
@@ -1,6 +1,7 @@
 base_model: codellama/CodeLlama-13b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
+is_llama_derived_model: true

 load_in_8bit: false
 load_in_4bit: true
@@ -42,8 +43,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
--- a/examples/code-llama/34b/lora.yml
+++ b/examples/code-llama/34b/lora.yml
@@ -1,6 +1,7 @@
 base_model: codellama/CodeLlama-34b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
+is_llama_derived_model: true

 load_in_8bit: true
 load_in_4bit: false
@@ -40,8 +41,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
@@ -51,7 +52,6 @@ local_rank:
 logging_steps: 1
 xformers_attention:
 flash_attention: true
-s2_attention:

 warmup_steps: 10
 evals_per_epoch: 4
--- a/examples/code-llama/34b/qlora.yml
+++ b/examples/code-llama/34b/qlora.yml
@@ -1,6 +1,7 @@
 base_model: codellama/CodeLlama-34b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
+is_llama_derived_model: true

 load_in_8bit: false
 load_in_4bit: true
@@ -42,8 +43,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
--- a/examples/code-llama/7b/lora.yml
+++ b/examples/code-llama/7b/lora.yml
@@ -1,6 +1,7 @@
 base_model: codellama/CodeLlama-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
+is_llama_derived_model: true

 load_in_8bit: true
 load_in_4bit: false
@@ -40,8 +41,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
@@ -51,7 +52,6 @@ local_rank:
 logging_steps: 1
 xformers_attention:
 flash_attention: true
-s2_attention:

 warmup_steps: 10
 evals_per_epoch: 4
--- a/examples/code-llama/7b/qlora.yml
+++ b/examples/code-llama/7b/qlora.yml
@@ -1,6 +1,7 @@
 base_model: codellama/CodeLlama-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: CodeLlamaTokenizer
+is_llama_derived_model: true

 load_in_8bit: false
 load_in_4bit: true
@@ -42,8 +43,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
--- a/examples/colab-notebooks/colab-axolotl-example.ipynb
+++ b/examples/colab-notebooks/colab-axolotl-example.ipynb
@@ -1,216 +0,0 @@
-{
-  "cells": [
-    {
-      "cell_type": "markdown",
-      "metadata": {
-        "id": "AKjdG7tbTb-n"
-      },
-      "source": [
-        "# Example notebook for running Axolotl on google colab"
-      ]
-    },
-    {
-      "cell_type": "code",
-      "execution_count": null,
-      "metadata": {
-        "id": "RcbNpOgWRcii"
-      },
-      "outputs": [],
-      "source": [
-        "import torch\n",
-        "# Check so there is a gpu available, a T4(free tier) is enough to run this notebook\n",
-        "assert (torch.cuda.is_available()==True)"
-      ]
-    },
-    {
-      "cell_type": "markdown",
-      "metadata": {
-        "id": "h3nLav8oTRA5"
-      },
-      "source": [
-        "## Install Axolotl and dependencies"
-      ]
-    },
-    {
-      "cell_type": "code",
-      "execution_count": null,
-      "metadata": {
-        "colab": {
-          "base_uri": "https://localhost:8080/"
-        },
-        "id": "3c3yGAwnOIdi",
-        "outputId": "e3777b5a-40ef-424f-e181-62dfecd1dd01"
-      },
-      "outputs": [],
-      "source": [
-        "!pip install torch==\"2.1.2\"\n",
-        "!pip install -e git+https://github.com/OpenAccess-AI-Collective/axolotl#egg=axolotl\n",
-        "!pip install flash-attn==\"2.5.0\"\n",
-        "!pip install deepspeed==\"0.13.1\""
-      ]
-    },
-    {
-      "cell_type": "markdown",
-      "metadata": {
-        "id": "BW2MFr7HTjub"
-      },
-      "source": [
-        "## Create an yaml config file"
-      ]
-    },
-    {
-      "cell_type": "code",
-      "execution_count": null,
-      "metadata": {
-        "id": "9pkF2dSoQEUN"
-      },
-      "outputs": [],
-      "source": [
-        "import yaml\n",
-        "\n",
-        "# Your YAML string\n",
-        "yaml_string = \"\"\"\n",
-        "base_model: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T\n",
-        "model_type: LlamaForCausalLM\n",
-        "tokenizer_type: LlamaTokenizer\n",
-        "is_llama_derived_model: true\n",
-        "\n",
-        "load_in_8bit: false\n",
-        "load_in_4bit: true\n",
-        "strict: false\n",
-        "\n",
-        "datasets:\n",
-        "  - path: mhenrichsen/alpaca_2k_test\n",
-        "    type: alpaca\n",
-        "dataset_prepared_path:\n",
-        "val_set_size: 0.05\n",
-        "output_dir: ./qlora-out\n",
-        "\n",
-        "adapter: qlora\n",
-        "lora_model_dir:\n",
-        "\n",
-        "sequence_len: 1096\n",
-        "sample_packing: true\n",
-        "pad_to_sequence_len: true\n",
-        "\n",
-        "lora_r: 32\n",
-        "lora_alpha: 16\n",
-        "lora_dropout: 0.05\n",
-        "lora_target_modules:\n",
-        "lora_target_linear: true\n",
-        "lora_fan_in_fan_out:\n",
-        "\n",
-        "wandb_project:\n",
-        "wandb_entity:\n",
-        "wandb_watch:\n",
-        "wandb_name:\n",
-        "wandb_log_model:\n",
-        "\n",
-        "mlflow_experiment_name: colab-example\n",
-        "\n",
-        "gradient_accumulation_steps: 1\n",
-        "micro_batch_size: 1\n",
-        "num_epochs: 4\n",
-        "max_steps: 20\n",
-        "optimizer: paged_adamw_32bit\n",
-        "lr_scheduler: cosine\n",
-        "learning_rate: 0.0002\n",
-        "\n",
-        "train_on_inputs: false\n",
-        "group_by_length: false\n",
-        "bf16: false\n",
-        "fp16: true\n",
-        "tf32: false\n",
-        "\n",
-        "gradient_checkpointing: true\n",
-        "early_stopping_patience:\n",
-        "resume_from_checkpoint:\n",
-        "local_rank:\n",
-        "logging_steps: 1\n",
-        "xformers_attention:\n",
-        "flash_attention: false\n",
-        "\n",
-        "warmup_steps: 10\n",
-        "evals_per_epoch:\n",
-        "saves_per_epoch:\n",
-        "debug:\n",
-        "deepspeed:\n",
-        "weight_decay: 0.0\n",
-        "fsdp:\n",
-        "fsdp_config:\n",
-        "special_tokens:\n",
-        "\n",
-        "\"\"\"\n",
-        "\n",
-        "# Convert the YAML string to a Python dictionary\n",
-        "yaml_dict = yaml.safe_load(yaml_string)\n",
-        "\n",
-        "# Specify your file path\n",
-        "file_path = 'test_axolotl.yaml'\n",
-        "\n",
-        "# Write the YAML file\n",
-        "with open(file_path, 'w') as file:\n",
-        "    yaml.dump(yaml_dict, file)\n"
-      ]
-    },
-    {
-      "cell_type": "markdown",
-      "metadata": {
-        "id": "bidoj8YLTusD"
-      },
-      "source": [
-        "## Launch the training"
-      ]
-    },
-    {
-      "cell_type": "code",
-      "execution_count": null,
-      "metadata": {
-        "colab": {
-          "base_uri": "https://localhost:8080/"
-        },
-        "id": "ydTI2Jk2RStU",
-        "outputId": "d6d0df17-4b53-439c-c802-22c0456d301b"
-      },
-      "outputs": [],
-      "source": [
-        "# Buy using the ! the comand will be executed as a bash command\n",
-        "!accelerate launch -m axolotl.cli.train /content/test_axolotl.yaml"
-      ]
-    },
-    {
-      "cell_type": "markdown",
-      "metadata": {},
-      "source": [
-        "## Play with inference"
-      ]
-    },
-    {
-      "cell_type": "code",
-      "execution_count": null,
-      "metadata": {},
-      "outputs": [],
-      "source": [
-        "# Buy using the ! the comand will be executed as a bash command\n",
-        "!accelerate launch -m axolotl.cli.inference /content/test_axolotl.yaml \\\n",
-        "    --qlora_model_dir=\"./qlora-out\" --gradio"
-      ]
-    }
-  ],
-  "metadata": {
-    "accelerator": "GPU",
-    "colab": {
-      "gpuType": "T4",
-      "provenance": []
-    },
-    "kernelspec": {
-      "display_name": "Python 3",
-      "name": "python3"
-    },
-    "language_info": {
-      "name": "python"
-    }
-  },
-  "nbformat": 4,
-  "nbformat_minor": 0
-}
--- a/examples/falcon/config-7b-lora.yml
+++ b/examples/falcon/config-7b-lora.yml
@@ -2,7 +2,7 @@ base_model: 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
@@ -38,8 +38,8 @@ lr_scheduler: cosine
 learning_rate: 0.00003
 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: true
 gradient_checkpointing: true
 early_stopping_patience:
@@ -60,5 +60,5 @@ fsdp:
 fsdp_config:
 special_tokens:
  pad_token: "<|endoftext|>"
-  bos_token: "<|endoftext|>"
+  bos_token: ">>ABSTRACT<<"
  eos_token: "<|endoftext|>"
--- a/examples/falcon/config-7b-qlora.yml
+++ b/examples/falcon/config-7b-qlora.yml
@@ -5,7 +5,7 @@ base_model: tiiuae/falcon-7b
 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
@@ -64,8 +64,8 @@ lr_scheduler: cosine
 learning_rate: 0.0002
 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: true
 gradient_checkpointing: true
 # stop training after this many evaluation losses have increased in a row
@@ -89,5 +89,5 @@ fsdp:
 fsdp_config:
 special_tokens:
  pad_token: "<|endoftext|>"
-  bos_token: "<|endoftext|>"
+  bos_token: ">>ABSTRACT<<"
  eos_token: "<|endoftext|>"
--- a/examples/falcon/config-7b.yml
+++ b/examples/falcon/config-7b.yml
@@ -2,7 +2,7 @@ base_model: 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
@@ -38,8 +38,8 @@ lr_scheduler: cosine
 learning_rate: 0.00003
 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: true
 gradient_checkpointing: true
 early_stopping_patience:
@@ -60,5 +60,5 @@ fsdp:
 fsdp_config:
 special_tokens:
  pad_token: "<|endoftext|>"
-  bos_token: "<|endoftext|>"
+  bos_token: ">>ABSTRACT<<"
  eos_token: "<|endoftext|>"
--- a/examples/gemma/qlora.yml
+++ b/examples/gemma/qlora.yml
@@ -1,65 +0,0 @@
-# use google/gemma-7b if you have access
-base_model: mhenrichsen/gemma-7b
-model_type: AutoModelForCausalLM
-tokenizer_type: AutoTokenizer
-
-load_in_8bit: false
-load_in_4bit: true
-strict: false
-
-# huggingface repo
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-val_set_size: 0.1
-output_dir: ./out
-
-adapter: qlora
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-
-sequence_len: 4096
-sample_packing: false
-pad_to_sequence_len: true
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-
-gradient_accumulation_steps: 3
-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: auto
-fp16:
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_ratio: 0.1
-evals_per_epoch: 4
-eval_table_size:
-eval_max_new_tokens: 128
-saves_per_epoch: 1
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
--- a/examples/gptj/qlora.yml
+++ b/examples/gptj/qlora.yml
@@ -33,8 +33,8 @@ lr_scheduler: cosine
 learning_rate: 0.0001
 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: true
 gradient_checkpointing: true
 early_stopping_patience:
--- a/examples/jeopardy-bot/config.yml
+++ b/examples/jeopardy-bot/config.yml
@@ -31,7 +31,7 @@ lr_scheduler: cosine
 learning_rate: 0.00003
 train_on_inputs: false
 group_by_length: false
-bf16: auto
+bf16: true
 tf32: true
 early_stopping_patience:
 resume_from_checkpoint:
--- a/examples/llama-2/fft_optimized.yml
+++ b/examples/llama-2/fft_optimized.yml
@@ -1,6 +1,7 @@
 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
@@ -40,8 +41,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
@@ -61,8 +62,11 @@ evals_per_epoch: 4
 eval_table_size:
 saves_per_epoch: 1
 debug:
-deepspeed: #deepspeed_configs/zero2.json # multi-gpu only
+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
+is_llama_derived_model: false
 gptq: true
 gptq_disable_exllama: true
 model_type: AutoModelForCausalLM
--- a/examples/llama-2/loftq.yml
+++ b/examples/llama-2/loftq.yml
@@ -1,69 +0,0 @@
-base_model: NousResearch/Llama-2-7b-hf
-model_type: LlamaForCausalLM
-tokenizer_type: LlamaTokenizer
-
-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: ./lora-out
-
-sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
-
-adapter: lora
-lora_model_dir:
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-lora_fan_in_fan_out:
-peft:
-  loftq_config:
-    loftq_bits: 4
-
-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: auto
-fp16:
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-s2_attention:
-
-warmup_steps: 10
-evals_per_epoch: 4
-eval_table_size:
-eval_max_new_tokens: 128
-saves_per_epoch: 1
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
--- a/examples/llama-2/lora.yml
+++ b/examples/llama-2/lora.yml
@@ -1,6 +1,7 @@
 base_model: NousResearch/Llama-2-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
+is_llama_derived_model: true

 load_in_8bit: true
 load_in_4bit: false
@@ -40,8 +41,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
@@ -51,12 +52,11 @@ local_rank:
 logging_steps: 1
 xformers_attention:
 flash_attention: true
-s2_attention:

 warmup_steps: 10
 evals_per_epoch: 4
 eval_table_size:
-eval_max_new_tokens: 128
+eval_table_max_new_tokens: 128
 saves_per_epoch: 1
 debug:
 deepspeed:
@@ -64,3 +64,6 @@ weight_decay: 0.0
 fsdp:
 fsdp_config:
 special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"
--- a/examples/llama-2/qlora-fsdp.yml
+++ b/examples/llama-2/qlora-fsdp.yml
@@ -1,70 +0,0 @@
-base_model: NousResearch/Llama-2-7b-hf
-model_type: LlamaForCausalLM
-tokenizer_type: LlamaTokenizer
-
-load_in_8bit: false
-load_in_4bit: true
-strict: false
-
-datasets:
-  - path: yahma/alpaca-cleaned
-    type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.05
-output_dir: ./qlora-out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 512
-sample_packing: false
-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: 4
-num_epochs: 4
-optimizer: paged_adamw_8bit
-lr_scheduler: cosine
-learning_rate: 0.00001
-
-train_on_inputs: false
-group_by_length: false
-bf16: auto
-fp16:
-tf32: false
-
-gradient_checkpointing: true
-gradient_checkpointing_kwargs:
-  use_reentrant: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_steps: 10
-evals_per_epoch: 4
-eval_table_size:
-saves_per_epoch: 1
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-  - full_shard
-fsdp_config:
-  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
-special_tokens:
--- a/examples/llama-2/qlora.yml
+++ b/examples/llama-2/qlora.yml
@@ -1,6 +1,7 @@
 base_model: NousResearch/Llama-2-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
+is_llama_derived_model: true

 load_in_8bit: false
 load_in_4bit: true
@@ -42,8 +43,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
@@ -64,3 +65,6 @@ weight_decay: 0.0
 fsdp:
 fsdp_config:
 special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"
--- a/examples/llama-2/relora.yml
+++ b/examples/llama-2/relora.yml
@@ -1,7 +1,7 @@
 base_model: NousResearch/Llama-2-7b-hf
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
-
+is_llama_derived_model: true

 load_in_8bit: false
 load_in_4bit: true
@@ -47,8 +47,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
--- a/examples/mamba/config.yml
+++ b/examples/mamba/config.yml
@@ -34,8 +34,8 @@ learning_rate: 5e-5
 train_on_inputs: false
 group_by_length: true

-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: true

 gradient_checkpointing: false
@@ -49,7 +49,7 @@ flash_attention:
 warmup_steps: 10
 evals_per_epoch: 4
 eval_table_size:
-eval_max_new_tokens: 128
+eval_table_max_new_tokens: 128
 saves_per_epoch: 1
 debug:
 deepspeed:
--- a/examples/mistral/Mistral-7b-example/README.md
+++ b/examples/mistral/Mistral-7b-example/README.md
@@ -1,12 +0,0 @@
-# Description
-This repository presents an in-depth guide for fine-tuning Mistral-7b or any other compatible model using Axolotl, tailored specifically for chatbot development. It streamlines the process of fine-tuning and uploading the enhanced model to HuggingFace 🤗, thereby serving as an invaluable tool for developers in the AI and chatbot domain.
-
-**What’s Inside:**
-
-Beginner-Friendly Instructions: Comprehensive steps to guide you through fine-tuning your chosen model, including details on the data structure (jsonl), configuration, and the code itself.
-
-Hardware Utilized: For reference, the fine-tuning in this guide was performed using 4x NVIDIA GeForce RTX 3090 (rented 2.1.2-cuda12.1-cudnn8-devel).
-
-**Uploading to HuggingFace 🤗:**
-To upload your fine-tuned model to Hugging Face, include the following files:
-![Screenshot 2024-01-19 213932](https://github.com/OpenAccess-AI-Collective/axolotl/assets/138583191/d660eb84-2d76-46a1-9846-cf0aeb3006d9)
--- a/examples/mistral/Mistral-7b-example/code.ipynb
+++ b/examples/mistral/Mistral-7b-example/code.ipynb
@@ -1,970 +0,0 @@
-{
- "cells": [
-  {
-   "cell_type": "code",
-   "execution_count": 1,
-   "id": "3fe31229-8f6b-48bc-a86d-af8e5466d11c",
-   "metadata": {
-    "scrolled": true
-   },
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "GPU available? True\n",
-      "BF16 is supported? True\n"
-     ]
-    }
-   ],
-   "source": [
-    "# Check if GPU is available I used 4x NVIDIA GeForce RTX 3090 (rented 2.1.2-cuda12.1-cudnn8-devel)\n",
-    "import torch\n",
-    "print('GPU available?', torch.cuda.is_available())\n",
-    "print('BF16 is supported?', torch.cuda.is_bf16_supported())"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 2,
-   "id": "1dee845b-f3cb-4b1e-bdd9-1a918eac140b",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Collecting huggingface_hub\n",
-      "  Downloading huggingface_hub-0.20.1-py3-none-any.whl.metadata (12 kB)\n",
-      "Requirement already satisfied: filelock in /opt/conda/lib/python3.10/site-packages (from huggingface_hub) (3.9.0)\n",
-      "Requirement already satisfied: fsspec>=2023.5.0 in /opt/conda/lib/python3.10/site-packages (from huggingface_hub) (2023.10.0)\n",
-      "Requirement already satisfied: requests in /opt/conda/lib/python3.10/site-packages (from huggingface_hub) (2.31.0)\n",
-      "Requirement already satisfied: tqdm>=4.42.1 in /opt/conda/lib/python3.10/site-packages (from huggingface_hub) (4.65.0)\n",
-      "Requirement already satisfied: pyyaml>=5.1 in /opt/conda/lib/python3.10/site-packages (from huggingface_hub) (6.0.1)\n",
-      "Requirement already satisfied: typing-extensions>=3.7.4.3 in /opt/conda/lib/python3.10/site-packages (from huggingface_hub) (4.7.1)\n",
-      "Requirement already satisfied: packaging>=20.9 in /opt/conda/lib/python3.10/site-packages (from huggingface_hub) (23.1)\n",
-      "Requirement already satisfied: charset-normalizer<4,>=2 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface_hub) (2.0.4)\n",
-      "Requirement already satisfied: idna<4,>=2.5 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface_hub) (3.4)\n",
-      "Requirement already satisfied: urllib3<3,>=1.21.1 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface_hub) (1.26.18)\n",
-      "Requirement already satisfied: certifi>=2017.4.17 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface_hub) (2023.7.22)\n",
-      "Downloading huggingface_hub-0.20.1-py3-none-any.whl (330 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m330.1/330.1 kB\u001b[0m \u001b[31m8.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m:00:01\u001b[0m\n",
-      "\u001b[?25hInstalling collected packages: huggingface_hub\n",
-      "Successfully installed huggingface_hub-0.20.1\n",
-      "\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\u001b[33m\n",
-      "\u001b[0m"
-     ]
-    }
-   ],
-   "source": [
-    "!pip install huggingface_hub"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 4,
-   "id": "88731672-9050-4034-8266-11aaace2a44e",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from huggingface_hub import notebook_login"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 5,
-   "id": "6b5aa7d7-3b18-4c14-afd4-043c2c545259",
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "60df98d7b0294289aad8b6c8cd023c3b",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "VBox(children=(HTML(value='<center> <img\\nsrc=https://huggingface.co/front/assets/huggingface_logo-noborder.sv…"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    }
-   ],
-   "source": [
-    "#Login to huggingface so you can push the model to hub later\n",
-    "import sys\n",
-    "stdout = sys.stdout\n",
-    "notebook_login()"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 6,
-   "id": "b74d0635-5033-4494-b7bd-ff6822103d93",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "#I noticed that when you use notebook_login() nothing gets printed after so we use sys \n",
-    "sys.stdout = stdout"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 7,
-   "id": "e3c3b088-45e7-484b-ae39-66beabc48da8",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Cloning into 'axolotl'...\n",
-      "remote: Enumerating objects: 235, done.\u001b[K\n",
-      "remote: Counting objects: 100% (235/235), done.\u001b[K\n",
-      "remote: Compressing objects: 100% (207/207), done.\u001b[K\n",
-      "remote: Total 235 (delta 48), reused 123 (delta 13), pack-reused 0\u001b[K\n",
-      "Receiving objects: 100% (235/235), 1.46 MiB | 11.65 MiB/s, done.\n",
-      "Resolving deltas: 100% (48/48), done.\n"
-     ]
-    }
-   ],
-   "source": [
-    "#axolotl\n",
-    "!git clone -b main --depth 1 https://github.com/OpenAccess-AI-Collective/axolotl"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 8,
-   "id": "66927751-4fd6-4477-97fc-6ab08c9d9a74",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "/axolotl\n"
-     ]
-    }
-   ],
-   "source": [
-    "cd axolotl"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 9,
-   "id": "fcccf8da-353b-4d70-8f55-5cfe08c7e6b9",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Requirement already satisfied: packaging in /opt/conda/lib/python3.10/site-packages (23.1)\n",
-      "\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\u001b[33m\n",
-      "\u001b[0mObtaining file:///axolotl\n",
-      "  Preparing metadata (setup.py) ... \u001b[?25ldone\n",
-      "\u001b[?25hCollecting auto-gptq==0.5.1\n",
-      "  Downloading auto_gptq-0.5.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (20 kB)\n",
-      "Requirement already satisfied: packaging in /opt/conda/lib/python3.10/site-packages (23.1)\n",
-      "Collecting peft==0.6.0\n",
-      "  Downloading peft-0.6.0-py3-none-any.whl.metadata (23 kB)\n",
-      "Collecting transformers==4.36.2\n",
-      "  Downloading transformers-4.36.2-py3-none-any.whl.metadata (126 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m126.8/126.8 kB\u001b[0m \u001b[31m9.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting tokenizers==0.15.0\n",
-      "  Downloading tokenizers-0.15.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (6.7 kB)\n",
-      "Collecting bitsandbytes>=0.41.1\n",
-      "  Downloading bitsandbytes-0.41.3.post2-py3-none-any.whl.metadata (9.8 kB)\n",
-      "Collecting accelerate==0.24.1\n",
-      "  Downloading accelerate-0.24.1-py3-none-any.whl.metadata (18 kB)\n",
-      "Collecting addict\n",
-      "  Downloading addict-2.4.0-py3-none-any.whl (3.8 kB)\n",
-      "Collecting fire\n",
-      "  Downloading fire-0.5.0.tar.gz (88 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m88.3/88.3 kB\u001b[0m \u001b[31m28.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25h  Preparing metadata (setup.py) ... \u001b[?25ldone\n",
-      "\u001b[?25hRequirement already satisfied: PyYAML>=6.0 in /opt/conda/lib/python3.10/site-packages (6.0.1)\n",
-      "Collecting datasets>=2.15.0\n",
-      "  Downloading datasets-2.16.0-py3-none-any.whl.metadata (20 kB)\n",
-      "Collecting sentencepiece\n",
-      "  Downloading sentencepiece-0.1.99-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (1.3 MB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m1.3/1.3 MB\u001b[0m \u001b[31m47.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting wandb\n",
-      "  Downloading wandb-0.16.1-py3-none-any.whl.metadata (9.8 kB)\n",
-      "Collecting einops\n",
-      "  Downloading einops-0.7.0-py3-none-any.whl.metadata (13 kB)\n",
-      "Collecting optimum==1.13.2\n",
-      "  Downloading optimum-1.13.2.tar.gz (300 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m301.0/301.0 kB\u001b[0m \u001b[31m72.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25h  Installing build dependencies ... \u001b[?25ldone\n",
-      "\u001b[?25h  Getting requirements to build wheel ... \u001b[?25ldone\n",
-      "\u001b[?25h  Preparing metadata (pyproject.toml) ... \u001b[?25ldone\n",
-      "\u001b[?25hCollecting hf_transfer\n",
-      "  Downloading hf_transfer-0.1.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (1.5 kB)\n",
-      "Collecting colorama\n",
-      "  Downloading colorama-0.4.6-py2.py3-none-any.whl (25 kB)\n",
-      "Collecting numba\n",
-      "  Downloading numba-0.58.1-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl.metadata (2.7 kB)\n",
-      "Requirement already satisfied: numpy>=1.24.4 in /opt/conda/lib/python3.10/site-packages (1.26.0)\n",
-      "Collecting bert-score==0.3.13\n",
-      "  Downloading bert_score-0.3.13-py3-none-any.whl (61 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m61.1/61.1 kB\u001b[0m \u001b[31m20.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting evaluate==0.4.0\n",
-      "  Downloading evaluate-0.4.0-py3-none-any.whl (81 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m81.4/81.4 kB\u001b[0m \u001b[31m26.0 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting rouge-score==0.1.2\n",
-      "  Downloading rouge_score-0.1.2.tar.gz (17 kB)\n",
-      "  Preparing metadata (setup.py) ... \u001b[?25ldone\n",
-      "\u001b[?25hCollecting scipy\n",
-      "  Downloading scipy-1.11.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (60 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m60.4/60.4 kB\u001b[0m \u001b[31m17.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting scikit-learn==1.2.2\n",
-      "  Downloading scikit_learn-1.2.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (9.6 MB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m9.6/9.6 MB\u001b[0m \u001b[31m83.2 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m00:01\u001b[0m00:01\u001b[0mm\n",
-      "\u001b[?25hCollecting pynvml\n",
-      "  Downloading pynvml-11.5.0-py3-none-any.whl (53 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m53.1/53.1 kB\u001b[0m \u001b[31m13.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting art\n",
-      "  Downloading art-6.1-py3-none-any.whl.metadata (69 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m69.9/69.9 kB\u001b[0m \u001b[31m21.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting fschat==0.2.34\n",
-      "  Downloading fschat-0.2.34-py3-none-any.whl.metadata (20 kB)\n",
-      "Collecting gradio==3.50.2\n",
-      "  Downloading gradio-3.50.2-py3-none-any.whl.metadata (17 kB)\n",
-      "Collecting tensorboard\n",
-      "  Downloading tensorboard-2.15.1-py3-none-any.whl.metadata (1.7 kB)\n",
-      "Collecting s3fs\n",
-      "  Downloading s3fs-2023.12.2-py3-none-any.whl.metadata (1.6 kB)\n",
-      "Collecting gcsfs\n",
-      "  Downloading gcsfs-2023.12.2.post1-py2.py3-none-any.whl.metadata (1.6 kB)\n",
-      "Collecting xformers==0.0.23\n",
-      "  Downloading xformers-0.0.23-cp310-cp310-manylinux2014_x86_64.whl.metadata (1.0 kB)\n",
-      "Collecting deepspeed\n",
-      "  Downloading deepspeed-0.12.6.tar.gz (1.2 MB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m1.2/1.2 MB\u001b[0m \u001b[31m109.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25h  Preparing metadata (setup.py) ... \u001b[?25ldone\n",
-      "\u001b[?25hCollecting flash-attn==2.3.3\n",
-      "  Downloading flash_attn-2.3.3.tar.gz (2.3 MB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m2.3/2.3 MB\u001b[0m \u001b[31m111.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25h  Preparing metadata (setup.py) ... \u001b[?25ldone\n",
-      "\u001b[?25hRequirement already satisfied: psutil in /opt/conda/lib/python3.10/site-packages (from accelerate==0.24.1) (5.9.0)\n",
-      "Requirement already satisfied: torch>=1.10.0 in /opt/conda/lib/python3.10/site-packages (from accelerate==0.24.1) (2.1.1)\n",
-      "Requirement already satisfied: huggingface-hub in /opt/conda/lib/python3.10/site-packages (from accelerate==0.24.1) (0.20.1)\n",
-      "Collecting rouge (from auto-gptq==0.5.1)\n",
-      "  Downloading rouge-1.0.1-py3-none-any.whl (13 kB)\n",
-      "Collecting gekko (from auto-gptq==0.5.1)\n",
-      "  Downloading gekko-1.0.6-py3-none-any.whl (12.2 MB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m12.2/12.2 MB\u001b[0m \u001b[31m77.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m00:01\u001b[0m0:01\u001b[0m\n",
-      "\u001b[?25hCollecting safetensors (from auto-gptq==0.5.1)\n",
-      "  Downloading safetensors-0.4.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (3.8 kB)\n",
-      "Requirement already satisfied: tqdm in /opt/conda/lib/python3.10/site-packages (from auto-gptq==0.5.1) (4.65.0)\n",
-      "Collecting pandas>=1.0.1 (from bert-score==0.3.13)\n",
-      "  Downloading pandas-2.1.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (18 kB)\n",
-      "Requirement already satisfied: requests in /opt/conda/lib/python3.10/site-packages (from bert-score==0.3.13) (2.31.0)\n",
-      "Collecting matplotlib (from bert-score==0.3.13)\n",
-      "  Downloading matplotlib-3.8.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (5.8 kB)\n",
-      "Collecting dill (from evaluate==0.4.0)\n",
-      "  Downloading dill-0.3.7-py3-none-any.whl.metadata (9.9 kB)\n",
-      "Collecting xxhash (from evaluate==0.4.0)\n",
-      "  Downloading xxhash-3.4.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (12 kB)\n",
-      "Collecting multiprocess (from evaluate==0.4.0)\n",
-      "  Downloading multiprocess-0.70.15-py310-none-any.whl.metadata (7.2 kB)\n",
-      "Requirement already satisfied: fsspec>=2021.05.0 in /opt/conda/lib/python3.10/site-packages (from fsspec[http]>=2021.05.0->evaluate==0.4.0) (2023.10.0)\n",
-      "Collecting responses<0.19 (from evaluate==0.4.0)\n",
-      "  Downloading responses-0.18.0-py3-none-any.whl (38 kB)\n",
-      "Collecting ninja (from flash-attn==2.3.3)\n",
-      "  Downloading ninja-1.11.1.1-py2.py3-none-manylinux1_x86_64.manylinux_2_5_x86_64.whl.metadata (5.3 kB)\n",
-      "Collecting aiohttp (from fschat==0.2.34)\n",
-      "  Downloading aiohttp-3.9.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (7.4 kB)\n",
-      "Collecting fastapi (from fschat==0.2.34)\n",
-      "  Downloading fastapi-0.108.0-py3-none-any.whl.metadata (24 kB)\n",
-      "Collecting httpx (from fschat==0.2.34)\n",
-      "  Downloading httpx-0.26.0-py3-none-any.whl.metadata (7.6 kB)\n",
-      "Collecting markdown2[all] (from fschat==0.2.34)\n",
-      "  Downloading markdown2-2.4.12-py2.py3-none-any.whl.metadata (2.0 kB)\n",
-      "Collecting nh3 (from fschat==0.2.34)\n",
-      "  Downloading nh3-0.2.15-cp37-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (1.7 kB)\n",
-      "Requirement already satisfied: prompt-toolkit>=3.0.0 in /opt/conda/lib/python3.10/site-packages (from fschat==0.2.34) (3.0.36)\n",
-      "Collecting pydantic<2,>=1 (from fschat==0.2.34)\n",
-      "  Downloading pydantic-1.10.13-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (149 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m149.6/149.6 kB\u001b[0m \u001b[31m42.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting rich>=10.0.0 (from fschat==0.2.34)\n",
-      "  Downloading rich-13.7.0-py3-none-any.whl.metadata (18 kB)\n",
-      "Collecting shortuuid (from fschat==0.2.34)\n",
-      "  Downloading shortuuid-1.0.11-py3-none-any.whl (10 kB)\n",
-      "Collecting tiktoken (from fschat==0.2.34)\n",
-      "  Downloading tiktoken-0.5.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (6.6 kB)\n",
-      "Collecting uvicorn (from fschat==0.2.34)\n",
-      "  Downloading uvicorn-0.25.0-py3-none-any.whl.metadata (6.4 kB)\n",
-      "Collecting aiofiles<24.0,>=22.0 (from gradio==3.50.2)\n",
-      "  Downloading aiofiles-23.2.1-py3-none-any.whl.metadata (9.7 kB)\n",
-      "Collecting altair<6.0,>=4.2.0 (from gradio==3.50.2)\n",
-      "  Downloading altair-5.2.0-py3-none-any.whl.metadata (8.7 kB)\n",
-      "Collecting ffmpy (from gradio==3.50.2)\n",
-      "  Downloading ffmpy-0.3.1.tar.gz (5.5 kB)\n",
-      "  Preparing metadata (setup.py) ... \u001b[?25ldone\n",
-      "\u001b[?25hCollecting gradio-client==0.6.1 (from gradio==3.50.2)\n",
-      "  Downloading gradio_client-0.6.1-py3-none-any.whl.metadata (7.1 kB)\n",
-      "Collecting importlib-resources<7.0,>=1.3 (from gradio==3.50.2)\n",
-      "  Downloading importlib_resources-6.1.1-py3-none-any.whl.metadata (4.1 kB)\n",
-      "Requirement already satisfied: jinja2<4.0 in /opt/conda/lib/python3.10/site-packages (from gradio==3.50.2) (3.1.2)\n",
-      "Requirement already satisfied: markupsafe~=2.0 in /opt/conda/lib/python3.10/site-packages (from gradio==3.50.2) (2.1.1)\n",
-      "Collecting orjson~=3.0 (from gradio==3.50.2)\n",
-      "  Downloading orjson-3.9.10-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (49 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m49.3/49.3 kB\u001b[0m \u001b[31m14.0 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hRequirement already satisfied: pillow<11.0,>=8.0 in /opt/conda/lib/python3.10/site-packages (from gradio==3.50.2) (10.0.1)\n",
-      "Collecting pydub (from gradio==3.50.2)\n",
-      "  Downloading pydub-0.25.1-py2.py3-none-any.whl (32 kB)\n",
-      "Collecting python-multipart (from gradio==3.50.2)\n",
-      "  Downloading python_multipart-0.0.6-py3-none-any.whl (45 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m45.7/45.7 kB\u001b[0m \u001b[31m13.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting semantic-version~=2.0 (from gradio==3.50.2)\n",
-      "  Downloading semantic_version-2.10.0-py2.py3-none-any.whl (15 kB)\n",
-      "Requirement already satisfied: typing-extensions~=4.0 in /opt/conda/lib/python3.10/site-packages (from gradio==3.50.2) (4.7.1)\n",
-      "Collecting websockets<12.0,>=10.0 (from gradio==3.50.2)\n",
-      "  Downloading websockets-11.0.3-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl (129 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m129.9/129.9 kB\u001b[0m \u001b[31m30.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting coloredlogs (from optimum==1.13.2)\n",
-      "  Downloading coloredlogs-15.0.1-py2.py3-none-any.whl (46 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m46.0/46.0 kB\u001b[0m \u001b[31m11.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hRequirement already satisfied: sympy in /opt/conda/lib/python3.10/site-packages (from optimum==1.13.2) (1.11.1)\n",
-      "Collecting absl-py (from rouge-score==0.1.2)\n",
-      "  Downloading absl_py-2.0.0-py3-none-any.whl.metadata (2.3 kB)\n",
-      "Collecting nltk (from rouge-score==0.1.2)\n",
-      "  Downloading nltk-3.8.1-py3-none-any.whl (1.5 MB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m1.5/1.5 MB\u001b[0m \u001b[31m90.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hRequirement already satisfied: six>=1.14.0 in /opt/conda/lib/python3.10/site-packages (from rouge-score==0.1.2) (1.16.0)\n",
-      "Collecting joblib>=1.1.1 (from scikit-learn==1.2.2)\n",
-      "  Downloading joblib-1.3.2-py3-none-any.whl.metadata (5.4 kB)\n",
-      "Collecting threadpoolctl>=2.0.0 (from scikit-learn==1.2.2)\n",
-      "  Downloading threadpoolctl-3.2.0-py3-none-any.whl.metadata (10.0 kB)\n",
-      "Requirement already satisfied: filelock in /opt/conda/lib/python3.10/site-packages (from transformers==4.36.2) (3.9.0)\n",
-      "Collecting regex!=2019.12.17 (from transformers==4.36.2)\n",
-      "  Downloading regex-2023.12.25-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (40 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m40.9/40.9 kB\u001b[0m \u001b[31m12.0 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hRequirement already satisfied: networkx in /opt/conda/lib/python3.10/site-packages (from torch>=1.10.0->accelerate==0.24.1) (3.1)\n",
-      "Collecting pyarrow>=8.0.0 (from datasets>=2.15.0)\n",
-      "  Downloading pyarrow-14.0.2-cp310-cp310-manylinux_2_28_x86_64.whl.metadata (3.0 kB)\n",
-      "Collecting pyarrow-hotfix (from datasets>=2.15.0)\n",
-      "  Downloading pyarrow_hotfix-0.6-py3-none-any.whl.metadata (3.6 kB)\n",
-      "Collecting hjson (from deepspeed)\n",
-      "  Downloading hjson-3.1.0-py3-none-any.whl (54 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m54.0/54.0 kB\u001b[0m \u001b[31m19.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting py-cpuinfo (from deepspeed)\n",
-      "  Downloading py_cpuinfo-9.0.0-py3-none-any.whl (22 kB)\n",
-      "Collecting termcolor (from fire)\n",
-      "  Downloading termcolor-2.4.0-py3-none-any.whl.metadata (6.1 kB)\n",
-      "Requirement already satisfied: decorator>4.1.2 in /opt/conda/lib/python3.10/site-packages (from gcsfs) (5.1.1)\n",
-      "INFO: pip is looking at multiple versions of gcsfs to determine which version is compatible with other requirements. This could take a while.\n",
-      "Collecting gcsfs\n",
-      "  Downloading gcsfs-2023.12.1-py2.py3-none-any.whl.metadata (1.6 kB)\n",
-      "  Downloading gcsfs-2023.12.0-py2.py3-none-any.whl.metadata (1.6 kB)\n",
-      "  Downloading gcsfs-2023.10.0-py2.py3-none-any.whl.metadata (1.6 kB)\n",
-      "Collecting google-auth>=1.2 (from gcsfs)\n",
-      "  Downloading google_auth-2.25.2-py2.py3-none-any.whl.metadata (4.7 kB)\n",
-      "Collecting google-auth-oauthlib (from gcsfs)\n",
-      "  Downloading google_auth_oauthlib-1.2.0-py2.py3-none-any.whl.metadata (2.7 kB)\n",
-      "Collecting google-cloud-storage (from gcsfs)\n",
-      "  Downloading google_cloud_storage-2.14.0-py2.py3-none-any.whl.metadata (6.1 kB)\n",
-      "Collecting llvmlite<0.42,>=0.41.0dev0 (from numba)\n",
-      "  Downloading llvmlite-0.41.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (4.8 kB)\n",
-      "Collecting aiobotocore<3.0.0,>=2.5.4 (from s3fs)\n",
-      "  Downloading aiobotocore-2.9.0-py3-none-any.whl.metadata (20 kB)\n",
-      "INFO: pip is looking at multiple versions of s3fs to determine which version is compatible with other requirements. This could take a while.\n",
-      "Collecting s3fs\n",
-      "  Downloading s3fs-2023.12.1-py3-none-any.whl.metadata (1.6 kB)\n",
-      "  Downloading s3fs-2023.10.0-py3-none-any.whl.metadata (1.6 kB)\n",
-      "Collecting aiobotocore~=2.7.0 (from s3fs)\n",
-      "  Downloading aiobotocore-2.7.0-py3-none-any.whl.metadata (20 kB)\n",
-      "Collecting grpcio>=1.48.2 (from tensorboard)\n",
-      "  Downloading grpcio-1.60.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (4.0 kB)\n",
-      "Collecting markdown>=2.6.8 (from tensorboard)\n",
-      "  Downloading Markdown-3.5.1-py3-none-any.whl.metadata (7.1 kB)\n",
-      "Collecting protobuf<4.24,>=3.19.6 (from tensorboard)\n",
-      "  Downloading protobuf-4.23.4-cp37-abi3-manylinux2014_x86_64.whl.metadata (540 bytes)\n",
-      "Requirement already satisfied: setuptools>=41.0.0 in /opt/conda/lib/python3.10/site-packages (from tensorboard) (68.0.0)\n",
-      "Collecting tensorboard-data-server<0.8.0,>=0.7.0 (from tensorboard)\n",
-      "  Downloading tensorboard_data_server-0.7.2-py3-none-manylinux_2_31_x86_64.whl.metadata (1.1 kB)\n",
-      "Collecting werkzeug>=1.0.1 (from tensorboard)\n",
-      "  Downloading werkzeug-3.0.1-py3-none-any.whl.metadata (4.1 kB)\n",
-      "Requirement already satisfied: Click!=8.0.0,>=7.1 in /opt/conda/lib/python3.10/site-packages (from wandb) (8.1.7)\n",
-      "Collecting GitPython!=3.1.29,>=1.0.0 (from wandb)\n",
-      "  Downloading GitPython-3.1.40-py3-none-any.whl.metadata (12 kB)\n",
-      "Collecting sentry-sdk>=1.0.0 (from wandb)\n",
-      "  Downloading sentry_sdk-1.39.1-py2.py3-none-any.whl.metadata (9.7 kB)\n",
-      "Collecting docker-pycreds>=0.4.0 (from wandb)\n",
-      "  Downloading docker_pycreds-0.4.0-py2.py3-none-any.whl (9.0 kB)\n",
-      "Collecting setproctitle (from wandb)\n",
-      "  Downloading setproctitle-1.3.3-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (9.9 kB)\n",
-      "Collecting appdirs>=1.4.3 (from wandb)\n",
-      "  Downloading appdirs-1.4.4-py2.py3-none-any.whl (9.6 kB)\n",
-      "Collecting botocore<1.31.65,>=1.31.16 (from aiobotocore~=2.7.0->s3fs)\n",
-      "  Downloading botocore-1.31.64-py3-none-any.whl.metadata (6.1 kB)\n",
-      "Collecting wrapt<2.0.0,>=1.10.10 (from aiobotocore~=2.7.0->s3fs)\n",
-      "  Downloading wrapt-1.16.0-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (6.6 kB)\n",
-      "Collecting aioitertools<1.0.0,>=0.5.1 (from aiobotocore~=2.7.0->s3fs)\n",
-      "  Downloading aioitertools-0.11.0-py3-none-any.whl (23 kB)\n",
-      "Requirement already satisfied: attrs>=17.3.0 in /opt/conda/lib/python3.10/site-packages (from aiohttp->fschat==0.2.34) (23.1.0)\n",
-      "Collecting multidict<7.0,>=4.5 (from aiohttp->fschat==0.2.34)\n",
-      "  Downloading multidict-6.0.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (114 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m114.5/114.5 kB\u001b[0m \u001b[31m37.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting yarl<2.0,>=1.0 (from aiohttp->fschat==0.2.34)\n",
-      "  Downloading yarl-1.9.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (31 kB)\n",
-      "Collecting frozenlist>=1.1.1 (from aiohttp->fschat==0.2.34)\n",
-      "  Downloading frozenlist-1.4.1-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (12 kB)\n",
-      "Collecting aiosignal>=1.1.2 (from aiohttp->fschat==0.2.34)\n",
-      "  Downloading aiosignal-1.3.1-py3-none-any.whl (7.6 kB)\n",
-      "Collecting async-timeout<5.0,>=4.0 (from aiohttp->fschat==0.2.34)\n",
-      "  Downloading async_timeout-4.0.3-py3-none-any.whl.metadata (4.2 kB)\n",
-      "Requirement already satisfied: jsonschema>=3.0 in /opt/conda/lib/python3.10/site-packages (from altair<6.0,>=4.2.0->gradio==3.50.2) (4.20.0)\n",
-      "Requirement already satisfied: toolz in /opt/conda/lib/python3.10/site-packages (from altair<6.0,>=4.2.0->gradio==3.50.2) (0.12.0)\n",
-      "Collecting gitdb<5,>=4.0.1 (from GitPython!=3.1.29,>=1.0.0->wandb)\n",
-      "  Downloading gitdb-4.0.11-py3-none-any.whl.metadata (1.2 kB)\n",
-      "Collecting cachetools<6.0,>=2.0.0 (from google-auth>=1.2->gcsfs)\n",
-      "  Downloading cachetools-5.3.2-py3-none-any.whl.metadata (5.2 kB)\n",
-      "Collecting pyasn1-modules>=0.2.1 (from google-auth>=1.2->gcsfs)\n",
-      "  Downloading pyasn1_modules-0.3.0-py2.py3-none-any.whl (181 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m181.3/181.3 kB\u001b[0m \u001b[31m59.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting rsa<5,>=3.1.4 (from google-auth>=1.2->gcsfs)\n",
-      "  Downloading rsa-4.9-py3-none-any.whl (34 kB)\n",
-      "Collecting requests-oauthlib>=0.7.0 (from google-auth-oauthlib->gcsfs)\n",
-      "  Downloading requests_oauthlib-1.3.1-py2.py3-none-any.whl (23 kB)\n",
-      "Collecting contourpy>=1.0.1 (from matplotlib->bert-score==0.3.13)\n",
-      "  Downloading contourpy-1.2.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (5.8 kB)\n",
-      "Collecting cycler>=0.10 (from matplotlib->bert-score==0.3.13)\n",
-      "  Downloading cycler-0.12.1-py3-none-any.whl.metadata (3.8 kB)\n",
-      "Collecting fonttools>=4.22.0 (from matplotlib->bert-score==0.3.13)\n",
-      "  Downloading fonttools-4.47.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (157 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m157.2/157.2 kB\u001b[0m \u001b[31m41.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting kiwisolver>=1.3.1 (from matplotlib->bert-score==0.3.13)\n",
-      "  Downloading kiwisolver-1.4.5-cp310-cp310-manylinux_2_12_x86_64.manylinux2010_x86_64.whl.metadata (6.4 kB)\n",
-      "Collecting pyparsing>=2.3.1 (from matplotlib->bert-score==0.3.13)\n",
-      "  Downloading pyparsing-3.1.1-py3-none-any.whl.metadata (5.1 kB)\n",
-      "Requirement already satisfied: python-dateutil>=2.7 in /opt/conda/lib/python3.10/site-packages (from matplotlib->bert-score==0.3.13) (2.8.2)\n",
-      "Requirement already satisfied: pytz>=2020.1 in /opt/conda/lib/python3.10/site-packages (from pandas>=1.0.1->bert-score==0.3.13) (2023.3.post1)\n",
-      "Collecting tzdata>=2022.1 (from pandas>=1.0.1->bert-score==0.3.13)\n",
-      "  Downloading tzdata-2023.3-py2.py3-none-any.whl (341 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m341.8/341.8 kB\u001b[0m \u001b[31m72.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hRequirement already satisfied: wcwidth in /opt/conda/lib/python3.10/site-packages (from prompt-toolkit>=3.0.0->fschat==0.2.34) (0.2.5)\n",
-      "Requirement already satisfied: charset-normalizer<4,>=2 in /opt/conda/lib/python3.10/site-packages (from requests->bert-score==0.3.13) (2.0.4)\n",
-      "Requirement already satisfied: idna<4,>=2.5 in /opt/conda/lib/python3.10/site-packages (from requests->bert-score==0.3.13) (3.4)\n",
-      "Requirement already satisfied: urllib3<3,>=1.21.1 in /opt/conda/lib/python3.10/site-packages (from requests->bert-score==0.3.13) (1.26.18)\n",
-      "Requirement already satisfied: certifi>=2017.4.17 in /opt/conda/lib/python3.10/site-packages (from requests->bert-score==0.3.13) (2023.7.22)\n",
-      "Collecting markdown-it-py>=2.2.0 (from rich>=10.0.0->fschat==0.2.34)\n",
-      "  Downloading markdown_it_py-3.0.0-py3-none-any.whl.metadata (6.9 kB)\n",
-      "Requirement already satisfied: pygments<3.0.0,>=2.13.0 in /opt/conda/lib/python3.10/site-packages (from rich>=10.0.0->fschat==0.2.34) (2.15.1)\n",
-      "Collecting h11>=0.8 (from uvicorn->fschat==0.2.34)\n",
-      "  Downloading h11-0.14.0-py3-none-any.whl (58 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m58.3/58.3 kB\u001b[0m \u001b[31m21.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting humanfriendly>=9.1 (from coloredlogs->optimum==1.13.2)\n",
-      "  Downloading humanfriendly-10.0-py2.py3-none-any.whl (86 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m86.8/86.8 kB\u001b[0m \u001b[31m27.2 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hCollecting starlette<0.33.0,>=0.29.0 (from fastapi->fschat==0.2.34)\n",
-      "  Downloading starlette-0.32.0.post1-py3-none-any.whl.metadata (5.8 kB)\n",
-      "Collecting typing-extensions~=4.0 (from gradio==3.50.2)\n",
-      "  Downloading typing_extensions-4.9.0-py3-none-any.whl.metadata (3.0 kB)\n",
-      "Collecting google-api-core!=2.0.*,!=2.1.*,!=2.2.*,!=2.3.0,<3.0.0dev,>=1.31.5 (from google-cloud-storage->gcsfs)\n",
-      "  Downloading google_api_core-2.15.0-py3-none-any.whl.metadata (2.7 kB)\n",
-      "Collecting google-cloud-core<3.0dev,>=2.3.0 (from google-cloud-storage->gcsfs)\n",
-      "  Downloading google_cloud_core-2.4.1-py2.py3-none-any.whl.metadata (2.7 kB)\n",
-      "Collecting google-resumable-media>=2.6.0 (from google-cloud-storage->gcsfs)\n",
-      "  Downloading google_resumable_media-2.7.0-py2.py3-none-any.whl.metadata (2.2 kB)\n",
-      "Collecting google-crc32c<2.0dev,>=1.0 (from google-cloud-storage->gcsfs)\n",
-      "  Downloading google_crc32c-1.5.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (32 kB)\n",
-      "Requirement already satisfied: anyio in /opt/conda/lib/python3.10/site-packages (from httpx->fschat==0.2.34) (4.2.0)\n",
-      "Collecting httpcore==1.* (from httpx->fschat==0.2.34)\n",
-      "  Downloading httpcore-1.0.2-py3-none-any.whl.metadata (20 kB)\n",
-      "Requirement already satisfied: sniffio in /opt/conda/lib/python3.10/site-packages (from httpx->fschat==0.2.34) (1.3.0)\n",
-      "Collecting wavedrom (from markdown2[all]->fschat==0.2.34)\n",
-      "  Downloading wavedrom-2.0.3.post3.tar.gz (137 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m137.7/137.7 kB\u001b[0m \u001b[31m47.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25h  Preparing metadata (setup.py) ... \u001b[?25ldone\n",
-      "\u001b[?25hRequirement already satisfied: mpmath>=0.19 in /opt/conda/lib/python3.10/site-packages (from sympy->optimum==1.13.2) (1.3.0)\n",
-      "Collecting jmespath<2.0.0,>=0.7.1 (from botocore<1.31.65,>=1.31.16->aiobotocore~=2.7.0->s3fs)\n",
-      "  Downloading jmespath-1.0.1-py3-none-any.whl (20 kB)\n",
-      "Collecting smmap<6,>=3.0.1 (from gitdb<5,>=4.0.1->GitPython!=3.1.29,>=1.0.0->wandb)\n",
-      "  Downloading smmap-5.0.1-py3-none-any.whl.metadata (4.3 kB)\n",
-      "Collecting googleapis-common-protos<2.0.dev0,>=1.56.2 (from google-api-core!=2.0.*,!=2.1.*,!=2.2.*,!=2.3.0,<3.0.0dev,>=1.31.5->google-cloud-storage->gcsfs)\n",
-      "  Downloading googleapis_common_protos-1.62.0-py2.py3-none-any.whl.metadata (1.5 kB)\n",
-      "Requirement already satisfied: jsonschema-specifications>=2023.03.6 in /opt/conda/lib/python3.10/site-packages (from jsonschema>=3.0->altair<6.0,>=4.2.0->gradio==3.50.2) (2023.12.1)\n",
-      "Requirement already satisfied: referencing>=0.28.4 in /opt/conda/lib/python3.10/site-packages (from jsonschema>=3.0->altair<6.0,>=4.2.0->gradio==3.50.2) (0.32.0)\n",
-      "Requirement already satisfied: rpds-py>=0.7.1 in /opt/conda/lib/python3.10/site-packages (from jsonschema>=3.0->altair<6.0,>=4.2.0->gradio==3.50.2) (0.15.2)\n",
-      "Collecting mdurl~=0.1 (from markdown-it-py>=2.2.0->rich>=10.0.0->fschat==0.2.34)\n",
-      "  Downloading mdurl-0.1.2-py3-none-any.whl (10.0 kB)\n",
-      "Collecting pyasn1<0.6.0,>=0.4.6 (from pyasn1-modules>=0.2.1->google-auth>=1.2->gcsfs)\n",
-      "  Downloading pyasn1-0.5.1-py2.py3-none-any.whl.metadata (8.6 kB)\n",
-      "Collecting oauthlib>=3.0.0 (from requests-oauthlib>=0.7.0->google-auth-oauthlib->gcsfs)\n",
-      "  Downloading oauthlib-3.2.2-py3-none-any.whl (151 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m151.7/151.7 kB\u001b[0m \u001b[31m50.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hRequirement already satisfied: exceptiongroup>=1.0.2 in /opt/conda/lib/python3.10/site-packages (from anyio->httpx->fschat==0.2.34) (1.0.4)\n",
-      "Collecting svgwrite (from wavedrom->markdown2[all]->fschat==0.2.34)\n",
-      "  Downloading svgwrite-1.4.3-py3-none-any.whl (67 kB)\n",
-      "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m67.1/67.1 kB\u001b[0m \u001b[31m21.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading accelerate-0.24.1-py3-none-any.whl (261 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m261.4/261.4 kB\u001b[0m \u001b[31m53.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading auto_gptq-0.5.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (4.8 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m4.8/4.8 MB\u001b[0m \u001b[31m89.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0mta \u001b[36m0:00:01\u001b[0m\n",
-      "\u001b[?25hDownloading fschat-0.2.34-py3-none-any.whl (220 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m220.1/220.1 kB\u001b[0m \u001b[31m63.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading gradio-3.50.2-py3-none-any.whl (20.3 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m20.3/20.3 MB\u001b[0m \u001b[31m82.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m:00:01\u001b[0m00:01\u001b[0m\n",
-      "\u001b[?25hDownloading peft-0.6.0-py3-none-any.whl (134 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m134.9/134.9 kB\u001b[0m \u001b[31m40.0 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading tokenizers-0.15.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (3.8 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m3.8/3.8 MB\u001b[0m \u001b[31m87.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0mta \u001b[36m0:00:01\u001b[0m\n",
-      "\u001b[?25hDownloading transformers-4.36.2-py3-none-any.whl (8.2 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m8.2/8.2 MB\u001b[0m \u001b[31m90.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0mta \u001b[36m0:00:01\u001b[0m\n",
-      "\u001b[?25hDownloading xformers-0.0.23-cp310-cp310-manylinux2014_x86_64.whl (213.0 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m213.0/213.0 MB\u001b[0m \u001b[31m36.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m00:01\u001b[0m00:01\u001b[0m\n",
-      "\u001b[?25hDownloading gradio_client-0.6.1-py3-none-any.whl (299 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m299.2/299.2 kB\u001b[0m \u001b[31m64.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading bitsandbytes-0.41.3.post2-py3-none-any.whl (92.6 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m92.6/92.6 MB\u001b[0m \u001b[31m56.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m:00:01\u001b[0m00:01\u001b[0m\n",
-      "\u001b[?25hDownloading datasets-2.16.0-py3-none-any.whl (507 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m507.1/507.1 kB\u001b[0m \u001b[31m87.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading scipy-1.11.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (36.4 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m36.4/36.4 MB\u001b[0m \u001b[31m77.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m:00:01\u001b[0m00:01\u001b[0m\n",
-      "\u001b[?25hDownloading art-6.1-py3-none-any.whl (599 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m599.8/599.8 kB\u001b[0m \u001b[31m96.0 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading einops-0.7.0-py3-none-any.whl (44 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m44.6/44.6 kB\u001b[0m \u001b[31m13.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading gcsfs-2023.10.0-py2.py3-none-any.whl (33 kB)\n",
-      "Downloading hf_transfer-0.1.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (3.9 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m3.9/3.9 MB\u001b[0m \u001b[31m99.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m:00:01\u001b[0m\n",
-      "\u001b[?25hDownloading numba-0.58.1-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl (3.6 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m3.6/3.6 MB\u001b[0m \u001b[31m100.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m00:01\u001b[0m\n",
-      "\u001b[?25hDownloading s3fs-2023.10.0-py3-none-any.whl (28 kB)\n",
-      "Downloading tensorboard-2.15.1-py3-none-any.whl (5.5 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m5.5/5.5 MB\u001b[0m \u001b[31m96.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0mta \u001b[36m0:00:01\u001b[0m\n",
-      "\u001b[?25hDownloading wandb-0.16.1-py3-none-any.whl (2.1 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m2.1/2.1 MB\u001b[0m \u001b[31m99.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading absl_py-2.0.0-py3-none-any.whl (130 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m130.2/130.2 kB\u001b[0m \u001b[31m36.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading aiobotocore-2.7.0-py3-none-any.whl (73 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m73.5/73.5 kB\u001b[0m \u001b[31m25.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading aiofiles-23.2.1-py3-none-any.whl (15 kB)\n",
-      "Downloading aiohttp-3.9.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (1.2 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m1.2/1.2 MB\u001b[0m \u001b[31m99.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading altair-5.2.0-py3-none-any.whl (996 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m996.9/996.9 kB\u001b[0m \u001b[31m110.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading dill-0.3.7-py3-none-any.whl (115 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m115.3/115.3 kB\u001b[0m \u001b[31m34.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading GitPython-3.1.40-py3-none-any.whl (190 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m190.6/190.6 kB\u001b[0m \u001b[31m47.0 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading google_auth-2.25.2-py2.py3-none-any.whl (184 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m184.2/184.2 kB\u001b[0m \u001b[31m44.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading google_auth_oauthlib-1.2.0-py2.py3-none-any.whl (24 kB)\n",
-      "Downloading grpcio-1.60.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (5.4 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m5.4/5.4 MB\u001b[0m \u001b[31m102.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0ma \u001b[36m0:00:01\u001b[0m\n",
-      "\u001b[?25hDownloading importlib_resources-6.1.1-py3-none-any.whl (33 kB)\n",
-      "Downloading joblib-1.3.2-py3-none-any.whl (302 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m302.2/302.2 kB\u001b[0m \u001b[31m64.0 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading llvmlite-0.41.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (43.6 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m43.6/43.6 MB\u001b[0m \u001b[31m74.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m:00:01\u001b[0m00:01\u001b[0m\n",
-      "\u001b[?25hDownloading Markdown-3.5.1-py3-none-any.whl (102 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m102.2/102.2 kB\u001b[0m \u001b[31m34.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading matplotlib-3.8.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (11.6 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m11.6/11.6 MB\u001b[0m \u001b[31m99.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m:00:01\u001b[0m0:01\u001b[0m\n",
-      "\u001b[?25hDownloading orjson-3.9.10-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (138 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m138.7/138.7 kB\u001b[0m \u001b[31m38.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading pandas-2.1.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (12.3 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m12.3/12.3 MB\u001b[0m \u001b[31m96.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m:00:01\u001b[0m0:01\u001b[0m\n",
-      "\u001b[?25hDownloading protobuf-4.23.4-cp37-abi3-manylinux2014_x86_64.whl (304 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m304.5/304.5 kB\u001b[0m \u001b[31m68.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading pyarrow-14.0.2-cp310-cp310-manylinux_2_28_x86_64.whl (38.0 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m38.0/38.0 MB\u001b[0m \u001b[31m78.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m:00:01\u001b[0m00:01\u001b[0m\n",
-      "\u001b[?25hDownloading pydantic-1.10.13-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (3.1 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m3.1/3.1 MB\u001b[0m \u001b[31m95.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading regex-2023.12.25-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (773 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m774.0/774.0 kB\u001b[0m \u001b[31m116.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading rich-13.7.0-py3-none-any.whl (240 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m240.6/240.6 kB\u001b[0m \u001b[31m59.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading safetensors-0.4.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (1.3 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m1.3/1.3 MB\u001b[0m \u001b[31m102.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading sentry_sdk-1.39.1-py2.py3-none-any.whl (254 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m254.1/254.1 kB\u001b[0m \u001b[31m71.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading tensorboard_data_server-0.7.2-py3-none-manylinux_2_31_x86_64.whl (6.6 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m6.6/6.6 MB\u001b[0m \u001b[31m104.2 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0ma \u001b[36m0:00:01\u001b[0m\n",
-      "\u001b[?25hDownloading threadpoolctl-3.2.0-py3-none-any.whl (15 kB)\n",
-      "Downloading uvicorn-0.25.0-py3-none-any.whl (60 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m60.3/60.3 kB\u001b[0m \u001b[31m19.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading werkzeug-3.0.1-py3-none-any.whl (226 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m226.7/226.7 kB\u001b[0m \u001b[31m67.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading fastapi-0.108.0-py3-none-any.whl (92 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m92.0/92.0 kB\u001b[0m \u001b[31m33.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading typing_extensions-4.9.0-py3-none-any.whl (32 kB)\n",
-      "Downloading google_cloud_storage-2.14.0-py2.py3-none-any.whl (121 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m121.6/121.6 kB\u001b[0m \u001b[31m36.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading httpx-0.26.0-py3-none-any.whl (75 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m75.9/75.9 kB\u001b[0m \u001b[31m24.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading httpcore-1.0.2-py3-none-any.whl (76 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m76.9/76.9 kB\u001b[0m \u001b[31m28.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading multiprocess-0.70.15-py310-none-any.whl (134 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m134.8/134.8 kB\u001b[0m \u001b[31m48.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading nh3-0.2.15-cp37-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (1.7 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m1.7/1.7 MB\u001b[0m \u001b[31m108.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading ninja-1.11.1.1-py2.py3-none-manylinux1_x86_64.manylinux_2_5_x86_64.whl (307 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m307.2/307.2 kB\u001b[0m \u001b[31m66.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading pyarrow_hotfix-0.6-py3-none-any.whl (7.9 kB)\n",
-      "Downloading setproctitle-1.3.3-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl (30 kB)\n",
-      "Downloading termcolor-2.4.0-py3-none-any.whl (7.7 kB)\n",
-      "Downloading tiktoken-0.5.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (2.0 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m2.0/2.0 MB\u001b[0m \u001b[31m101.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading xxhash-3.4.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (194 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m194.1/194.1 kB\u001b[0m \u001b[31m44.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading async_timeout-4.0.3-py3-none-any.whl (5.7 kB)\n",
-      "Downloading botocore-1.31.64-py3-none-any.whl (11.3 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m11.3/11.3 MB\u001b[0m \u001b[31m98.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m:00:01\u001b[0m0:01\u001b[0m\n",
-      "\u001b[?25hDownloading cachetools-5.3.2-py3-none-any.whl (9.3 kB)\n",
-      "Downloading contourpy-1.2.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (310 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m310.7/310.7 kB\u001b[0m \u001b[31m69.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading cycler-0.12.1-py3-none-any.whl (8.3 kB)\n",
-      "Downloading fonttools-4.47.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (4.6 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m4.6/4.6 MB\u001b[0m \u001b[31m102.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0ma \u001b[36m0:00:01\u001b[0m\n",
-      "\u001b[?25hDownloading frozenlist-1.4.1-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl (239 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m239.5/239.5 kB\u001b[0m \u001b[31m71.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading gitdb-4.0.11-py3-none-any.whl (62 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m62.7/62.7 kB\u001b[0m \u001b[31m23.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading google_api_core-2.15.0-py3-none-any.whl (121 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m122.0/122.0 kB\u001b[0m \u001b[31m32.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading google_cloud_core-2.4.1-py2.py3-none-any.whl (29 kB)\n",
-      "Downloading google_resumable_media-2.7.0-py2.py3-none-any.whl (80 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m80.6/80.6 kB\u001b[0m \u001b[31m22.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading kiwisolver-1.4.5-cp310-cp310-manylinux_2_12_x86_64.manylinux2010_x86_64.whl (1.6 MB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m1.6/1.6 MB\u001b[0m \u001b[31m102.0 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading markdown_it_py-3.0.0-py3-none-any.whl (87 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m87.5/87.5 kB\u001b[0m \u001b[31m25.0 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading pyparsing-3.1.1-py3-none-any.whl (103 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m103.1/103.1 kB\u001b[0m \u001b[31m32.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading starlette-0.32.0.post1-py3-none-any.whl (70 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m70.0/70.0 kB\u001b[0m \u001b[31m19.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading wrapt-1.16.0-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl (80 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m80.3/80.3 kB\u001b[0m \u001b[31m30.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading yarl-1.9.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (301 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m301.6/301.6 kB\u001b[0m \u001b[31m80.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading markdown2-2.4.12-py2.py3-none-any.whl (41 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m41.2/41.2 kB\u001b[0m \u001b[31m12.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading googleapis_common_protos-1.62.0-py2.py3-none-any.whl (228 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m228.7/228.7 kB\u001b[0m \u001b[31m57.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading pyasn1-0.5.1-py2.py3-none-any.whl (84 kB)\n",
-      "\u001b[2K   \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m84.9/84.9 kB\u001b[0m \u001b[31m30.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
-      "\u001b[?25hDownloading smmap-5.0.1-py3-none-any.whl (24 kB)\n",
-      "Building wheels for collected packages: flash-attn, optimum, rouge-score, deepspeed, fire, ffmpy, wavedrom\n",
-      "  Building wheel for flash-attn (setup.py) ... \u001b[?25ldone\n",
-      "\u001b[?25h  Created wheel for flash-attn: filename=flash_attn-2.3.3-cp310-cp310-linux_x86_64.whl size=57042553 sha256=b1df92cb5bd7657d38b789dd48e907aa3e0bd2715c817eb85f3c4320bb11fb3f\n",
-      "  Stored in directory: /root/.cache/pip/wheels/e5/e6/fa/941802ec61d1afd320d27160ab1db98e6dba65381f84b76d4a\n",
-      "  Building wheel for optimum (pyproject.toml) ... \u001b[?25ldone\n",
-      "\u001b[?25h  Created wheel for optimum: filename=optimum-1.13.2-py3-none-any.whl size=395599 sha256=ff3a73120e1b6eeeda28f76e3fc8cd4cd826e5d66c869b7848ba150e7af79c62\n",
-      "  Stored in directory: /root/.cache/pip/wheels/6e/b7/2c/79405d98f0943373d8546daeae25a3d377f7659ca0cbe48699\n",
-      "  Building wheel for rouge-score (setup.py) ... \u001b[?25ldone\n",
-      "\u001b[?25h  Created wheel for rouge-score: filename=rouge_score-0.1.2-py3-none-any.whl size=24932 sha256=8118ecbbcd3529085e794c803f0ddb182fc6c6d3e8a494103b49a94abf1bec37\n",
-      "  Stored in directory: /root/.cache/pip/wheels/5f/dd/89/461065a73be61a532ff8599a28e9beef17985c9e9c31e541b4\n",
-      "  Building wheel for deepspeed (setup.py) ... \u001b[?25ldone\n",
-      "\u001b[?25h  Created wheel for deepspeed: filename=deepspeed-0.12.6-py3-none-any.whl size=1306729 sha256=35c46b6f0275b0d3063522e0af4f3cbd9ec1c310114d8917d87cbe2bf43346e2\n",
-      "  Stored in directory: /root/.cache/pip/wheels/a3/dc/a2/f585faaed4dec84108916dcc8e8a7c129a216df8202ca32984\n",
-      "  Building wheel for fire (setup.py) ... \u001b[?25ldone\n",
-      "\u001b[?25h  Created wheel for fire: filename=fire-0.5.0-py2.py3-none-any.whl size=116934 sha256=e76d5185f237f34ec69bb8aa657497bef07408978e4f7efdaef48663bb8cd4ef\n",
-      "  Stored in directory: /root/.cache/pip/wheels/90/d4/f7/9404e5db0116bd4d43e5666eaa3e70ab53723e1e3ea40c9a95\n",
-      "  Building wheel for ffmpy (setup.py) ... \u001b[?25ldone\n",
-      "\u001b[?25h  Created wheel for ffmpy: filename=ffmpy-0.3.1-py3-none-any.whl size=5579 sha256=da3b54dc0ac1a825a1a233315970ac80b8b4c53ebd9cb2a2cfdeab118f453a64\n",
-      "  Stored in directory: /root/.cache/pip/wheels/01/a6/d1/1c0828c304a4283b2c1639a09ad86f83d7c487ef34c6b4a1bf\n",
-      "  Building wheel for wavedrom (setup.py) ... \u001b[?25ldone\n",
-      "\u001b[?25h  Created wheel for wavedrom: filename=wavedrom-2.0.3.post3-py2.py3-none-any.whl size=30052 sha256=7f0cbd15d63ee9c120190bac122ab51bbbfc91ee374bc3c046fadb320816c17e\n",
-      "  Stored in directory: /root/.cache/pip/wheels/9c/52/8c/38b454b42f712f325e26f633287484c7dc1ad469e1580c5954\n",
-      "Successfully built flash-attn optimum rouge-score deepspeed fire ffmpy wavedrom\n",
-      "Installing collected packages: sentencepiece, pydub, py-cpuinfo, ninja, nh3, hjson, ffmpy, bitsandbytes, appdirs, addict, xxhash, wrapt, werkzeug, websockets, tzdata, typing-extensions, threadpoolctl, termcolor, tensorboard-data-server, svgwrite, smmap, shortuuid, setproctitle, sentry-sdk, semantic-version, scipy, safetensors, rouge, regex, python-multipart, pyparsing, pynvml, pyasn1, pyarrow-hotfix, pyarrow, protobuf, orjson, oauthlib, multidict, mdurl, markdown2, markdown, llvmlite, kiwisolver, joblib, jmespath, importlib-resources, humanfriendly, hf_transfer, h11, grpcio, google-crc32c, gekko, frozenlist, fonttools, einops, docker-pycreds, dill, cycler, contourpy, colorama, cachetools, async-timeout, art, aioitertools, aiofiles, absl-py, yarl, wavedrom, uvicorn, tiktoken, scikit-learn, rsa, responses, requests-oauthlib, pydantic, pyasn1-modules, pandas, numba, nltk, multiprocess, matplotlib, markdown-it-py, httpcore, googleapis-common-protos, google-resumable-media, gitdb, fire, coloredlogs, botocore, aiosignal, xformers, tokenizers, starlette, rouge-score, rich, httpx, google-auth, GitPython, flash-attn, deepspeed, aiohttp, accelerate, wandb, transformers, gradio-client, google-auth-oauthlib, google-api-core, fastapi, altair, aiobotocore, tensorboard, s3fs, peft, gradio, google-cloud-core, fschat, datasets, bert-score, optimum, google-cloud-storage, evaluate, auto-gptq, gcsfs, axolotl\n",
-      "  Attempting uninstall: typing-extensions\n",
-      "    Found existing installation: typing_extensions 4.7.1\n",
-      "    Uninstalling typing_extensions-4.7.1:\n",
-      "      Successfully uninstalled typing_extensions-4.7.1\n",
-      "  Running setup.py develop for axolotl\n",
-      "Successfully installed GitPython-3.1.40 absl-py-2.0.0 accelerate-0.24.1 addict-2.4.0 aiobotocore-2.7.0 aiofiles-23.2.1 aiohttp-3.9.1 aioitertools-0.11.0 aiosignal-1.3.1 altair-5.2.0 appdirs-1.4.4 art-6.1 async-timeout-4.0.3 auto-gptq-0.5.1 axolotl-0.3.0 bert-score-0.3.13 bitsandbytes-0.41.3.post2 botocore-1.31.64 cachetools-5.3.2 colorama-0.4.6 coloredlogs-15.0.1 contourpy-1.2.0 cycler-0.12.1 datasets-2.16.0 deepspeed-0.12.6 dill-0.3.7 docker-pycreds-0.4.0 einops-0.7.0 evaluate-0.4.0 fastapi-0.108.0 ffmpy-0.3.1 fire-0.5.0 flash-attn-2.3.3 fonttools-4.47.0 frozenlist-1.4.1 fschat-0.2.34 gcsfs-2023.10.0 gekko-1.0.6 gitdb-4.0.11 google-api-core-2.15.0 google-auth-2.25.2 google-auth-oauthlib-1.2.0 google-cloud-core-2.4.1 google-cloud-storage-2.14.0 google-crc32c-1.5.0 google-resumable-media-2.7.0 googleapis-common-protos-1.62.0 gradio-3.50.2 gradio-client-0.6.1 grpcio-1.60.0 h11-0.14.0 hf_transfer-0.1.4 hjson-3.1.0 httpcore-1.0.2 httpx-0.26.0 humanfriendly-10.0 importlib-resources-6.1.1 jmespath-1.0.1 joblib-1.3.2 kiwisolver-1.4.5 llvmlite-0.41.1 markdown-3.5.1 markdown-it-py-3.0.0 markdown2-2.4.12 matplotlib-3.8.2 mdurl-0.1.2 multidict-6.0.4 multiprocess-0.70.15 nh3-0.2.15 ninja-1.11.1.1 nltk-3.8.1 numba-0.58.1 oauthlib-3.2.2 optimum-1.13.2 orjson-3.9.10 pandas-2.1.4 peft-0.6.0 protobuf-4.23.4 py-cpuinfo-9.0.0 pyarrow-14.0.2 pyarrow-hotfix-0.6 pyasn1-0.5.1 pyasn1-modules-0.3.0 pydantic-1.10.13 pydub-0.25.1 pynvml-11.5.0 pyparsing-3.1.1 python-multipart-0.0.6 regex-2023.12.25 requests-oauthlib-1.3.1 responses-0.18.0 rich-13.7.0 rouge-1.0.1 rouge-score-0.1.2 rsa-4.9 s3fs-2023.10.0 safetensors-0.4.1 scikit-learn-1.2.2 scipy-1.11.4 semantic-version-2.10.0 sentencepiece-0.1.99 sentry-sdk-1.39.1 setproctitle-1.3.3 shortuuid-1.0.11 smmap-5.0.1 starlette-0.32.0.post1 svgwrite-1.4.3 tensorboard-2.15.1 tensorboard-data-server-0.7.2 termcolor-2.4.0 threadpoolctl-3.2.0 tiktoken-0.5.2 tokenizers-0.15.0 transformers-4.36.2 typing-extensions-4.8.0 tzdata-2023.3 uvicorn-0.25.0 wandb-0.16.1 wavedrom-2.0.3.post3 websockets-11.0.3 werkzeug-3.0.1 wrapt-1.16.0 xformers-0.0.23 xxhash-3.4.1 yarl-1.9.4\n",
-      "\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\u001b[33m\n",
-      "\u001b[0mCollecting git+https://github.com/huggingface/peft.git\n",
-      "  Cloning https://github.com/huggingface/peft.git to /tmp/pip-req-build-hka8xgk2\n",
-      "  Running command git clone --filter=blob:none --quiet https://github.com/huggingface/peft.git /tmp/pip-req-build-hka8xgk2\n",
-      "  Resolved https://github.com/huggingface/peft.git to commit cf04d0353f0343cbf66627228c4495f51669af34\n",
-      "  Installing build dependencies ... \u001b[?25ldone\n",
-      "\u001b[?25h  Getting requirements to build wheel ... \u001b[?25ldone\n",
-      "\u001b[?25h  Preparing metadata (pyproject.toml) ... \u001b[?25ldone\n",
-      "\u001b[?25hRequirement already satisfied: numpy>=1.17 in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (1.26.0)\n",
-      "Requirement already satisfied: packaging>=20.0 in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (23.1)\n",
-      "Requirement already satisfied: psutil in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (5.9.0)\n",
-      "Requirement already satisfied: pyyaml in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (6.0.1)\n",
-      "Requirement already satisfied: torch>=1.13.0 in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (2.1.1)\n",
-      "Requirement already satisfied: transformers in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (4.36.2)\n",
-      "Requirement already satisfied: tqdm in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (4.65.0)\n",
-      "Requirement already satisfied: accelerate>=0.21.0 in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (0.24.1)\n",
-      "Requirement already satisfied: safetensors in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (0.4.1)\n",
-      "Requirement already satisfied: huggingface-hub>=0.17.0 in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (0.20.1)\n",
-      "Requirement already satisfied: filelock in /opt/conda/lib/python3.10/site-packages (from huggingface-hub>=0.17.0->peft==0.7.2.dev0) (3.9.0)\n",
-      "Requirement already satisfied: fsspec>=2023.5.0 in /opt/conda/lib/python3.10/site-packages (from huggingface-hub>=0.17.0->peft==0.7.2.dev0) (2023.10.0)\n",
-      "Requirement already satisfied: requests in /opt/conda/lib/python3.10/site-packages (from huggingface-hub>=0.17.0->peft==0.7.2.dev0) (2.31.0)\n",
-      "Requirement already satisfied: typing-extensions>=3.7.4.3 in /opt/conda/lib/python3.10/site-packages (from huggingface-hub>=0.17.0->peft==0.7.2.dev0) (4.8.0)\n",
-      "Requirement already satisfied: sympy in /opt/conda/lib/python3.10/site-packages (from torch>=1.13.0->peft==0.7.2.dev0) (1.11.1)\n",
-      "Requirement already satisfied: networkx in /opt/conda/lib/python3.10/site-packages (from torch>=1.13.0->peft==0.7.2.dev0) (3.1)\n",
-      "Requirement already satisfied: jinja2 in /opt/conda/lib/python3.10/site-packages (from torch>=1.13.0->peft==0.7.2.dev0) (3.1.2)\n",
-      "Requirement already satisfied: regex!=2019.12.17 in /opt/conda/lib/python3.10/site-packages (from transformers->peft==0.7.2.dev0) (2023.12.25)\n",
-      "Requirement already satisfied: tokenizers<0.19,>=0.14 in /opt/conda/lib/python3.10/site-packages (from transformers->peft==0.7.2.dev0) (0.15.0)\n",
-      "Requirement already satisfied: MarkupSafe>=2.0 in /opt/conda/lib/python3.10/site-packages (from jinja2->torch>=1.13.0->peft==0.7.2.dev0) (2.1.1)\n",
-      "Requirement already satisfied: charset-normalizer<4,>=2 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface-hub>=0.17.0->peft==0.7.2.dev0) (2.0.4)\n",
-      "Requirement already satisfied: idna<4,>=2.5 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface-hub>=0.17.0->peft==0.7.2.dev0) (3.4)\n",
-      "Requirement already satisfied: urllib3<3,>=1.21.1 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface-hub>=0.17.0->peft==0.7.2.dev0) (1.26.18)\n",
-      "Requirement already satisfied: certifi>=2017.4.17 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface-hub>=0.17.0->peft==0.7.2.dev0) (2023.7.22)\n",
-      "Requirement already satisfied: mpmath>=0.19 in /opt/conda/lib/python3.10/site-packages (from sympy->torch>=1.13.0->peft==0.7.2.dev0) (1.3.0)\n",
-      "Building wheels for collected packages: peft\n",
-      "  Building wheel for peft (pyproject.toml) ... \u001b[?25ldone\n",
-      "\u001b[?25h  Created wheel for peft: filename=peft-0.7.2.dev0-py3-none-any.whl size=169456 sha256=4c70d23e759fa6abb3827fb2f3a8683be3b24d78777d0f403bbc2c0548e5dd4b\n",
-      "  Stored in directory: /tmp/pip-ephem-wheel-cache-my5ncou6/wheels/d7/c7/de/1368fac8590e1b103ddc2ec2a28ad51d83aded1a3830e8a087\n",
-      "Successfully built peft\n",
-      "Installing collected packages: peft\n",
-      "  Attempting uninstall: peft\n",
-      "    Found existing installation: peft 0.6.0\n",
-      "    Uninstalling peft-0.6.0:\n",
-      "      Successfully uninstalled peft-0.6.0\n",
-      "\u001b[31mERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.\n",
-      "axolotl 0.3.0 requires peft==0.6.0, but you have peft 0.7.2.dev0 which is incompatible.\u001b[0m\u001b[31m\n",
-      "\u001b[0mSuccessfully installed peft-0.7.2.dev0\n",
-      "\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\u001b[33m\n",
-      "\u001b[0m"
-     ]
-    }
-   ],
-   "source": [
-    "#instaling what is needed inside axolotl file\n",
-    "!pip install packaging\n",
-    "!pip install -e '.[flash-attn,deepspeed]'\n",
-    "!pip install -U git+https://github.com/huggingface/peft.git"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 16,
-   "id": "82d1a380-1e87-48fe-89fe-25331326014d",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "The following values were not passed to `accelerate launch` and had defaults used instead:\n",
-      "\t`--num_processes` was set to a value of `3`\n",
-      "\t\tMore than one GPU was found, enabling multi-GPU training.\n",
-      "\t\tIf this was unintended please pass in `--num_processes=1`.\n",
-      "\t`--num_machines` was set to a value of `1`\n",
-      "\t`--mixed_precision` was set to a value of `'no'`\n",
-      "\t`--dynamo_backend` was set to a value of `'no'`\n",
-      "To avoid this warning pass in values for each of the problematic parameters or run `accelerate config`.\n",
-      "/opt/conda/lib/python3.10/site-packages/transformers/deepspeed.py:23: FutureWarning: transformers.deepspeed module is deprecated and will be removed in a future version. Please import deepspeed modules directly from transformers.integrations\n",
-      "  warnings.warn(\n",
-      "[2023-12-28 15:44:09,979] [INFO] [datasets.<module>:58] [PID:2814] PyTorch version 2.1.1 available.\n",
-      "/opt/conda/lib/python3.10/site-packages/transformers/deepspeed.py:23: FutureWarning: transformers.deepspeed module is deprecated and will be removed in a future version. Please import deepspeed modules directly from transformers.integrations\n",
-      "  warnings.warn(\n",
-      "/opt/conda/lib/python3.10/site-packages/transformers/deepspeed.py:23: FutureWarning: transformers.deepspeed module is deprecated and will be removed in a future version. Please import deepspeed modules directly from transformers.integrations\n",
-      "  warnings.warn(\n",
-      "[2023-12-28 15:44:10,011] [INFO] [datasets.<module>:58] [PID:2812] PyTorch version 2.1.1 available.\n",
-      "[2023-12-28 15:44:10,013] [INFO] [datasets.<module>:58] [PID:2813] PyTorch version 2.1.1 available.\n",
-      "[2023-12-28 15:44:10,805] [INFO] [axolotl.normalize_config:150] [PID:2814] [RANK:2] GPU memory usage baseline: 0.000GB (+0.317GB misc)\u001b[39m\n",
-      "[2023-12-28 15:44:10,830] [INFO] [real_accelerator.py:161:get_accelerator] Setting ds_accelerator to cuda (auto detect)\n",
-      "[2023-12-28 15:44:10,842] [INFO] [axolotl.normalize_config:150] [PID:2813] [RANK:1] GPU memory usage baseline: 0.000GB (+0.317GB misc)\u001b[39m\n",
-      "[2023-12-28 15:44:10,865] [INFO] [real_accelerator.py:161:get_accelerator] Setting ds_accelerator to cuda (auto detect)\n",
-      "[2023-12-28 15:44:10,869] [INFO] [axolotl.normalize_config:150] [PID:2812] [RANK:0] GPU memory usage baseline: 0.000GB (+0.351GB misc)\u001b[39m\n",
-      "[2023-12-28 15:44:10,887] [INFO] [real_accelerator.py:161:get_accelerator] Setting ds_accelerator to cuda (auto detect)\n",
-      "[2023-12-28 15:44:10,961] [INFO] [comm.py:637:init_distributed] cdb=None\n",
-      "[2023-12-28 15:44:10,994] [INFO] [comm.py:637:init_distributed] cdb=None\n",
-      "[2023-12-28 15:44:11,015] [INFO] [comm.py:637:init_distributed] cdb=None\n",
-      "[2023-12-28 15:44:11,015] [INFO] [comm.py:668:init_distributed] Initializing TorchBackend in DeepSpeed with backend nccl\n",
-      "                                 dP            dP   dP \n",
-      "                                 88            88   88 \n",
-      "      .d8888b. dP.  .dP .d8888b. 88 .d8888b. d8888P 88 \n",
-      "      88'  `88  `8bd8'  88'  `88 88 88'  `88   88   88 \n",
-      "      88.  .88  .d88b.  88.  .88 88 88.  .88   88   88 \n",
-      "      `88888P8 dP'  `dP `88888P' dP `88888P'   dP   dP \n",
-      "                                                       \n",
-      "                                                       \n",
-      "\n",
-      "[2023-12-28 15:44:11,412] [DEBUG] [axolotl.load_tokenizer:184] [PID:2812] [RANK:0] EOS: 2 / </s>\u001b[39m\n",
-      "[2023-12-28 15:44:11,412] [DEBUG] [axolotl.load_tokenizer:185] [PID:2812] [RANK:0] BOS: 1 / <s>\u001b[39m\n",
-      "[2023-12-28 15:44:11,412] [DEBUG] [axolotl.load_tokenizer:186] [PID:2812] [RANK:0] PAD: 2 / </s>\u001b[39m\n",
-      "[2023-12-28 15:44:11,412] [DEBUG] [axolotl.load_tokenizer:187] [PID:2812] [RANK:0] UNK: 0 / <unk>\u001b[39m\n",
-      "[2023-12-28 15:44:11,413] [INFO] [axolotl.load_tokenized_prepared_datasets:143] [PID:2812] [RANK:0] Loading prepared dataset from disk at tilemachos/GF_new.json/1adc45d2edc1e98ce657814412c6593c...\u001b[39m\n",
-      "[2023-12-28 15:44:11,415] [INFO] [axolotl.load_tokenized_prepared_datasets:145] [PID:2812] [RANK:0] Prepared dataset loaded from disk...\u001b[39m\n",
-      "[2023-12-28 15:44:11,432] [DEBUG] [axolotl.load_tokenizer:184] [PID:2814] [RANK:2] EOS: 2 / </s>\u001b[39m\n",
-      "[2023-12-28 15:44:11,432] [DEBUG] [axolotl.load_tokenizer:185] [PID:2814] [RANK:2] BOS: 1 / <s>\u001b[39m\n",
-      "[2023-12-28 15:44:11,432] [DEBUG] [axolotl.load_tokenizer:186] [PID:2814] [RANK:2] PAD: 2 / </s>\u001b[39m\n",
-      "[2023-12-28 15:44:11,432] [DEBUG] [axolotl.load_tokenizer:187] [PID:2814] [RANK:2] UNK: 0 / <unk>\u001b[39m\n",
-      "[2023-12-28 15:44:11,530] [DEBUG] [axolotl.load_tokenizer:184] [PID:2813] [RANK:1] EOS: 2 / </s>\u001b[39m\n",
-      "[2023-12-28 15:44:11,531] [DEBUG] [axolotl.load_tokenizer:185] [PID:2813] [RANK:1] BOS: 1 / <s>\u001b[39m\n",
-      "[2023-12-28 15:44:11,531] [DEBUG] [axolotl.load_tokenizer:186] [PID:2813] [RANK:1] PAD: 2 / </s>\u001b[39m\n",
-      "[2023-12-28 15:44:11,531] [DEBUG] [axolotl.load_tokenizer:187] [PID:2813] [RANK:1] UNK: 0 / <unk>\u001b[39m\n",
-      "[2023-12-28 15:44:12,158] [INFO] [axolotl.load_tokenized_prepared_datasets:143] [PID:2813] [RANK:1] Loading prepared dataset from disk at tilemachos/GF_new.json/1adc45d2edc1e98ce657814412c6593c...\u001b[39m\n",
-      "[2023-12-28 15:44:12,158] [INFO] [axolotl.load_tokenized_prepared_datasets:143] [PID:2814] [RANK:2] Loading prepared dataset from disk at tilemachos/GF_new.json/1adc45d2edc1e98ce657814412c6593c...\u001b[39m\n",
-      "[2023-12-28 15:44:12,160] [INFO] [axolotl.load_tokenized_prepared_datasets:145] [PID:2813] [RANK:1] Prepared dataset loaded from disk...\u001b[39m\n",
-      "[2023-12-28 15:44:12,161] [INFO] [axolotl.load_tokenized_prepared_datasets:145] [PID:2814] [RANK:2] Prepared dataset loaded from disk...\u001b[39m\n",
-      "[2023-12-28 15:44:12,236] [DEBUG] [axolotl.log:60] [PID:2812] [RANK:0] total_num_tokens: 28120\u001b[39m\n",
-      "[2023-12-28 15:44:12,238] [DEBUG] [axolotl.log:60] [PID:2812] [RANK:0] `total_supervised_tokens: 7990`\u001b[39m\n",
-      "[2023-12-28 15:44:12,238] [DEBUG] [axolotl.log:60] [PID:2812] [RANK:0] total_num_steps: 6\u001b[39m\n",
-      "[2023-12-28 15:44:12,242] [DEBUG] [axolotl.train.log:60] [PID:2812] [RANK:0] loading tokenizer... mistralai/Mistral-7B-v0.1\u001b[39m\n",
-      "[2023-12-28 15:44:12,518] [DEBUG] [axolotl.load_tokenizer:184] [PID:2812] [RANK:0] EOS: 2 / </s>\u001b[39m\n",
-      "[2023-12-28 15:44:12,518] [DEBUG] [axolotl.load_tokenizer:185] [PID:2812] [RANK:0] BOS: 1 / <s>\u001b[39m\n",
-      "[2023-12-28 15:44:12,518] [DEBUG] [axolotl.load_tokenizer:186] [PID:2812] [RANK:0] PAD: 2 / </s>\u001b[39m\n",
-      "[2023-12-28 15:44:12,518] [DEBUG] [axolotl.load_tokenizer:187] [PID:2812] [RANK:0] UNK: 0 / <unk>\u001b[39m\n",
-      "[2023-12-28 15:44:12,518] [DEBUG] [axolotl.train.log:60] [PID:2812] [RANK:0] loading model and peft_config...\u001b[39m\n",
-      "[2023-12-28 15:44:12,589] [DEBUG] [axolotl.load_tokenizer:184] [PID:2814] [RANK:2] EOS: 2 / </s>\u001b[39m\n",
-      "[2023-12-28 15:44:12,589] [DEBUG] [axolotl.load_tokenizer:185] [PID:2814] [RANK:2] BOS: 1 / <s>\u001b[39m\n",
-      "[2023-12-28 15:44:12,589] [DEBUG] [axolotl.load_tokenizer:186] [PID:2814] [RANK:2] PAD: 2 / </s>\u001b[39m\n",
-      "[2023-12-28 15:44:12,589] [DEBUG] [axolotl.load_tokenizer:187] [PID:2814] [RANK:2] UNK: 0 / <unk>\u001b[39m\n",
-      "[2023-12-28 15:44:12,599] [DEBUG] [axolotl.load_tokenizer:184] [PID:2813] [RANK:1] EOS: 2 / </s>\u001b[39m\n",
-      "[2023-12-28 15:44:12,599] [DEBUG] [axolotl.load_tokenizer:185] [PID:2813] [RANK:1] BOS: 1 / <s>\u001b[39m\n",
-      "[2023-12-28 15:44:12,599] [DEBUG] [axolotl.load_tokenizer:186] [PID:2813] [RANK:1] PAD: 2 / </s>\u001b[39m\n",
-      "[2023-12-28 15:44:12,599] [DEBUG] [axolotl.load_tokenizer:187] [PID:2813] [RANK:1] UNK: 0 / <unk>\u001b[39m\n",
-      "[2023-12-28 15:44:13,049] [INFO] [partition_parameters.py:348:__exit__] finished initializing model - num_params = 291, num_elems = 7.24B\n",
-      "Loading checkpoint shards: 100%|██████████████████| 2/2 [00:11<00:00,  5.81s/it]\n",
-      "Loading checkpoint shards: 100%|██████████████████| 2/2 [00:11<00:00,  5.98s/it]\n",
-      "[2023-12-28 15:44:25,395] [INFO] [axolotl.load_model:503] [PID:2813] [RANK:1] GPU memory usage after model load: 7.576GB (+0.524GB cache, +0.708GB misc)\u001b[39m\n",
-      "[2023-12-28 15:44:25,399] [INFO] [axolotl.load_model:526] [PID:2813] [RANK:1] converting PEFT model w/ prepare_model_for_kbit_training\u001b[39m\n",
-      "[2023-12-28 15:44:25,403] [INFO] [axolotl.load_model:538] [PID:2813] [RANK:1] converting modules to torch.bfloat16 for flash attention\u001b[39m\n",
-      "trainable params: 3,407,872 || all params: 7,245,139,968 || trainable%: 0.04703666202518836\n",
-      "[2023-12-28 15:44:25,480] [INFO] [axolotl.load_model:568] [PID:2813] [RANK:1] GPU memory usage after adapters: 7.589GB (+1.501GB cache, +0.708GB misc)\u001b[39m\n",
-      "[2023-12-28 15:44:25,572] [INFO] [axolotl.load_model:503] [PID:2814] [RANK:2] GPU memory usage after model load: 7.576GB (+0.410GB cache, +0.708GB misc)\u001b[39m\n",
-      "[2023-12-28 15:44:25,576] [INFO] [axolotl.load_model:526] [PID:2814] [RANK:2] converting PEFT model w/ prepare_model_for_kbit_training\u001b[39m\n",
-      "[2023-12-28 15:44:25,580] [INFO] [axolotl.load_model:538] [PID:2814] [RANK:2] converting modules to torch.bfloat16 for flash attention\u001b[39m\n",
-      "trainable params: 3,407,872 || all params: 7,245,139,968 || trainable%: 0.04703666202518836\n",
-      "[2023-12-28 15:44:25,660] [INFO] [axolotl.load_model:568] [PID:2814] [RANK:2] GPU memory usage after adapters: 7.589GB (+1.388GB cache, +0.708GB misc)\u001b[39m\n",
-      "Loading checkpoint shards: 100%|██████████████████| 2/2 [00:12<00:00,  6.30s/it]\n",
-      "[2023-12-28 15:44:26,170] [INFO] [axolotl.load_model:503] [PID:2812] [RANK:0] GPU memory usage after model load: 7.576GB (+0.776GB cache, +0.741GB misc)\u001b[39m\n",
-      "[2023-12-28 15:44:26,177] [INFO] [axolotl.load_model:526] [PID:2812] [RANK:0] converting PEFT model w/ prepare_model_for_kbit_training\u001b[39m\n",
-      "[2023-12-28 15:44:26,181] [INFO] [axolotl.load_model:538] [PID:2812] [RANK:0] converting modules to torch.bfloat16 for flash attention\u001b[39m\n",
-      "trainable params: 3,407,872 || all params: 7,245,139,968 || trainable%: 0.04703666202518836\n",
-      "[2023-12-28 15:44:26,259] [INFO] [axolotl.load_model:568] [PID:2812] [RANK:0] GPU memory usage after adapters: 7.589GB (+1.753GB cache, +0.741GB misc)\u001b[39m\n",
-      "[2023-12-28 15:44:26,293] [INFO] [axolotl.train.log:60] [PID:2812] [RANK:0] Pre-saving adapter config to ./out\u001b[39m\n",
-      "[2023-12-28 15:44:26,296] [INFO] [axolotl.train.log:60] [PID:2812] [RANK:0] Starting trainer...\u001b[39m\n",
-      "Using /root/.cache/torch_extensions/py310_cu121 as PyTorch extensions root...\n",
-      "Using /root/.cache/torch_extensions/py310_cu121 as PyTorch extensions root...\n",
-      "Using /root/.cache/torch_extensions/py310_cu121 as PyTorch extensions root...\n",
-      "Detected CUDA files, patching ldflags\n",
-      "Emitting ninja build file /root/.cache/torch_extensions/py310_cu121/fused_adam/build.ninja...\n",
-      "Building extension module fused_adam...\n",
-      "Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)\n",
-      "ninja: no work to do.\n",
-      "Loading extension module fused_adam...\n",
-      "Time to load fused_adam op: 0.05891108512878418 seconds\n",
-      "Loading extension module fused_adam...\n",
-      "Time to load fused_adam op: 0.10173463821411133 seconds\n",
-      "Loading extension module fused_adam...\n",
-      "Time to load fused_adam op: 0.10152459144592285 seconds\n",
-      "/opt/conda/lib/python3.10/site-packages/deepspeed/ops/adam/fused_adam.py:96: UserWarning: The torch.cuda.*DtypeTensor constructors are no longer recommended. It's best to use methods such as torch.tensor(data, dtype=*, device='cuda') to create tensors. (Triggered internally at /opt/conda/conda-bld/pytorch_1699449201336/work/torch/csrc/tensor/python_tensor.cpp:83.)\n",
-      "  self._dummy_overflow_buf = get_accelerator().IntTensor([0])\n",
-      "/opt/conda/lib/python3.10/site-packages/deepspeed/ops/adam/fused_adam.py:96: UserWarning: The torch.cuda.*DtypeTensor constructors are no longer recommended. It's best to use methods such as torch.tensor(data, dtype=*, device='cuda') to create tensors. (Triggered internally at /opt/conda/conda-bld/pytorch_1699449201336/work/torch/csrc/tensor/python_tensor.cpp:83.)\n",
-      "  self._dummy_overflow_buf = get_accelerator().IntTensor([0])\n",
-      "/opt/conda/lib/python3.10/site-packages/deepspeed/ops/adam/fused_adam.py:96: UserWarning: The torch.cuda.*DtypeTensor constructors are no longer recommended. It's best to use methods such as torch.tensor(data, dtype=*, device='cuda') to create tensors. (Triggered internally at /opt/conda/conda-bld/pytorch_1699449201336/work/torch/csrc/tensor/python_tensor.cpp:83.)\n",
-      "  self._dummy_overflow_buf = get_accelerator().IntTensor([0])\n",
-      "Parameter Offload: Total persistent parameters: 3674112 in 193 params\n",
-      "  0%|                                                    | 0/17 [00:00<?, ?it/s]/opt/conda/lib/python3.10/site-packages/torch/utils/checkpoint.py:429: UserWarning: torch.utils.checkpoint: please pass in use_reentrant=True or use_reentrant=False explicitly. The default value of use_reentrant will be updated to be False in the future. To maintain current behavior, pass use_reentrant=True. It is recommended that you use use_reentrant=False. Refer to docs for more details on the differences between the two variants.\n",
-      "  warnings.warn(\n",
-      "/opt/conda/lib/python3.10/site-packages/torch/utils/checkpoint.py:429: UserWarning: torch.utils.checkpoint: please pass in use_reentrant=True or use_reentrant=False explicitly. The default value of use_reentrant will be updated to be False in the future. To maintain current behavior, pass use_reentrant=True. It is recommended that you use use_reentrant=False. Refer to docs for more details on the differences between the two variants.\n",
-      "  warnings.warn(\n",
-      "/opt/conda/lib/python3.10/site-packages/torch/utils/checkpoint.py:429: UserWarning: torch.utils.checkpoint: please pass in use_reentrant=True or use_reentrant=False explicitly. The default value of use_reentrant will be updated to be False in the future. To maintain current behavior, pass use_reentrant=True. It is recommended that you use use_reentrant=False. Refer to docs for more details on the differences between the two variants.\n",
-      "  warnings.warn(\n",
-      "/opt/conda/lib/python3.10/site-packages/bitsandbytes/autograd/_functions.py:322: UserWarning: MatMul8bitLt: inputs will be cast from torch.bfloat16 to float16 during quantization\n",
-      "  warnings.warn(f\"MatMul8bitLt: inputs will be cast from {A.dtype} to float16 during quantization\")\n",
-      "/opt/conda/lib/python3.10/site-packages/bitsandbytes/autograd/_functions.py:322: UserWarning: MatMul8bitLt: inputs will be cast from torch.bfloat16 to float16 during quantization\n",
-      "  warnings.warn(f\"MatMul8bitLt: inputs will be cast from {A.dtype} to float16 during quantization\")\n",
-      "/opt/conda/lib/python3.10/site-packages/bitsandbytes/autograd/_functions.py:322: UserWarning: MatMul8bitLt: inputs will be cast from torch.bfloat16 to float16 during quantization\n",
-      "  warnings.warn(f\"MatMul8bitLt: inputs will be cast from {A.dtype} to float16 during quantization\")\n",
-      "{'loss': 2.0448, 'learning_rate': 2e-05, 'epoch': 0.06}                         \n",
-      "  6%|██▌                                         | 1/17 [00:28<07:32, 28.30s/it]\n",
-      "  0%|                                                     | 0/3 [00:00<?, ?it/s]\u001b[A\n",
-      " 67%|██████████████████████████████               | 2/3 [00:03<00:01,  1.85s/it]\u001b[A\n",
-      "                                                                                \u001b[A\n",
-      "\u001b[A{'eval_loss': 1.9694719314575195, 'eval_runtime': 11.391, 'eval_samples_per_second': 1.492, 'eval_steps_per_second': 0.263, 'epoch': 0.06}\n",
-      "  6%|██▌                                         | 1/17 [00:39<07:32, 28.30s/it]\n",
-      "100%|█████████████████████████████████████████████| 3/3 [00:07<00:00,  2.65s/it]\u001b[A\n",
-      "                                                                                \u001b[A[2023-12-28 15:45:35,358] [INFO] [axolotl.callbacks.on_step_end:122] [PID:2812] [RANK:0] GPU memory usage while training: 12.210GB (+4.259GB cache, +0.776GB misc)\u001b[39m\n",
-      " 12%|█████▏                                      | 2/17 [01:04<08:18, 33.20s/it][2023-12-28 15:45:35,358] [INFO] [axolotl.callbacks.on_step_end:122] [PID:2814] [RANK:2] GPU memory usage while training: 12.269GB (+4.522GB cache, +0.743GB misc)\u001b[39m\n",
-      "[2023-12-28 15:45:35,358] [INFO] [axolotl.callbacks.on_step_end:122] [PID:2813] [RANK:1] GPU memory usage while training: 12.283GB (+4.493GB cache, +0.743GB misc)\u001b[39m\n",
-      "{'loss': 2.0022, 'learning_rate': 4e-05, 'epoch': 0.12}                         \n",
-      "{'loss': 2.1054, 'learning_rate': 6e-05, 'epoch': 0.17}                         \n",
-      "{'loss': 1.9004, 'learning_rate': 8e-05, 'epoch': 0.23}                         \n",
-      "{'loss': 1.8794, 'learning_rate': 0.0001, 'epoch': 0.29}                        \n",
-      " 29%|████████████▉                               | 5/17 [02:20<05:23, 26.92s/it]\n",
-      "  0%|                                                     | 0/3 [00:00<?, ?it/s]\u001b[A\n",
-      " 67%|██████████████████████████████               | 2/3 [00:03<00:01,  1.88s/it]\u001b[A\n",
-      "                                                                                \u001b[A\n",
-      "\u001b[A{'eval_loss': 1.7912336587905884, 'eval_runtime': 11.3106, 'eval_samples_per_second': 1.503, 'eval_steps_per_second': 0.265, 'epoch': 0.29}\n",
-      " 29%|████████████▉                               | 5/17 [02:32<05:23, 26.92s/it]\n",
-      "100%|█████████████████████████████████████████████| 3/3 [00:07<00:00,  2.67s/it]\u001b[A\n",
-      "{'loss': 1.7871, 'learning_rate': 0.00012, 'epoch': 0.35}                       \u001b[A\n",
-      "{'loss': 1.7758, 'learning_rate': 0.00014, 'epoch': 0.4}                        \n",
-      "{'loss': 1.4645, 'learning_rate': 0.00016, 'epoch': 0.46}                       \n",
-      "{'loss': 1.4009, 'learning_rate': 0.00018, 'epoch': 0.52}                       \n",
-      "{'loss': 1.3927, 'learning_rate': 0.0002, 'epoch': 0.58}                        \n",
-      " 59%|█████████████████████████▎                 | 10/17 [04:38<03:04, 26.33s/it]\n",
-      "  0%|                                                     | 0/3 [00:00<?, ?it/s]\u001b[A\n",
-      " 67%|██████████████████████████████               | 2/3 [00:03<00:01,  1.89s/it]\u001b[A\n",
-      "                                                                                \u001b[A\n",
-      "\u001b[A{'eval_loss': 1.1426481008529663, 'eval_runtime': 11.3344, 'eval_samples_per_second': 1.5, 'eval_steps_per_second': 0.265, 'epoch': 0.58}\n",
-      " 59%|█████████████████████████▎                 | 10/17 [04:49<03:04, 26.33s/it]\n",
-      "100%|█████████████████████████████████████████████| 3/3 [00:07<00:00,  2.68s/it]\u001b[A\n",
-      "{'loss': 1.0122, 'learning_rate': 0.0001900968867902419, 'epoch': 0.63}         \u001b[A\n",
-      "{'loss': 1.0019, 'learning_rate': 0.00016234898018587337, 'epoch': 0.69}        \n",
-      "{'loss': 0.8976, 'learning_rate': 0.00012225209339563145, 'epoch': 0.75}        \n",
-      "{'loss': 0.9301, 'learning_rate': 7.774790660436858e-05, 'epoch': 0.81}         \n",
-      "{'loss': 0.8595, 'learning_rate': 3.7651019814126654e-05, 'epoch': 0.87}        \n",
-      " 88%|█████████████████████████████████████▉     | 15/17 [06:55<00:52, 26.17s/it]\n",
-      "  0%|                                                     | 0/3 [00:00<?, ?it/s]\u001b[A\n",
-      " 67%|██████████████████████████████               | 2/3 [00:03<00:01,  1.88s/it]\u001b[A\n",
-      "                                                                                \u001b[A\n",
-      "\u001b[A{'eval_loss': 0.8175248503684998, 'eval_runtime': 11.2932, 'eval_samples_per_second': 1.505, 'eval_steps_per_second': 0.266, 'epoch': 0.87}\n",
-      " 88%|█████████████████████████████████████▉     | 15/17 [07:06<00:52, 26.17s/it]\n",
-      "100%|█████████████████████████████████████████████| 3/3 [00:07<00:00,  2.67s/it]\u001b[A\n",
-      "{'loss': 0.7931, 'learning_rate': 9.903113209758096e-06, 'epoch': 0.92}         \u001b[A\n",
-      "{'loss': 0.6909, 'learning_rate': 0.0, 'epoch': 0.98}                           \n",
-      "100%|███████████████████████████████████████████| 17/17 [07:56<00:00, 28.03s/it]/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py:1879: UserWarning: Positional args are being deprecated, use kwargs instead. Refer to https://pytorch.org/docs/master/generated/torch.nn.Module.html#torch.nn.Module.state_dict for details.\n",
-      "  warnings.warn(\n",
-      "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py:1879: UserWarning: Positional args are being deprecated, use kwargs instead. Refer to https://pytorch.org/docs/master/generated/torch.nn.Module.html#torch.nn.Module.state_dict for details.\n",
-      "  warnings.warn(\n",
-      "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py:1879: UserWarning: Positional args are being deprecated, use kwargs instead. Refer to https://pytorch.org/docs/master/generated/torch.nn.Module.html#torch.nn.Module.state_dict for details.\n",
-      "  warnings.warn(\n",
-      "{'train_runtime': 489.0649, 'train_samples_per_second': 0.63, 'train_steps_per_second': 0.035, 'train_loss': 1.408153467318591, 'epoch': 0.98}\n",
-      "100%|███████████████████████████████████████████| 17/17 [08:09<00:00, 28.77s/it]\n",
-      "[2023-12-28 15:52:39,488] [INFO] [axolotl.train.log:60] [PID:2812] [RANK:0] Training Completed!!! Saving pre-trained model to ./out\u001b[39m\n",
-      "\u001b[0m\u001b[0m\u001b[0m"
-     ]
-    }
-   ],
-   "source": [
-    "\"\"\"\n",
-    "Training using the config.yml file and using deepspeed:zero3_bf16 the most aggressive optimization out of zero1,zero2,zero3 stages which partitions \n",
-    "not only optimizer states but also gradients and parameters across GPUs. The bf16 indicate mixed precision training using bfloat16.\n",
-    "For more information read axolotl's readme\n",
-    "\"\"\"\n",
-    "!accelerate launch -m axolotl.cli.train /folder/config.yml --deepspeed deepspeed_configs/zero3_bf16.json"
-   ]
-  }
- ],
- "metadata": {
-  "kernelspec": {
-   "display_name": "Python 3 (ipykernel)",
-   "language": "python",
-   "name": "python3"
-  },
-  "language_info": {
-   "codemirror_mode": {
-    "name": "ipython",
-    "version": 3
-   },
-   "file_extension": ".py",
-   "mimetype": "text/x-python",
-   "name": "python",
-   "nbconvert_exporter": "python",
-   "pygments_lexer": "ipython3",
-   "version": "3.10.13"
-  }
- },
- "nbformat": 4,
- "nbformat_minor": 5
-}
--- a/examples/mistral/Mistral-7b-example/config.yml
+++ b/examples/mistral/Mistral-7b-example/config.yml
@@ -1,74 +0,0 @@
-#Mistral-7b
-base_model: mistralai/Mistral-7B-v0.1
-model_type: MistralForCausalLM
-tokenizer_type: LlamaTokenizer
-
-load_in_8bit: true
-load_in_4bit: false
-strict: false
-
-datasets:
-  - path: tilemachos/Demo-Dataset #Path to json dataset file in huggingface
-    #for type,conversation arguments read axolotl readme and pick what is suited for your project, I wanted a chatbot and put sharegpt and chatml
-    type: sharegpt
-    conversation: chatml
-dataset_prepared_path: tilemachos/Demo-Dataset #Path to json dataset file in huggingface
-val_set_size: 0.05
-output_dir: ./out
-
-#using lora for lower cost
-adapter: lora
-lora_r: 8
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules:
-  - q_proj
-  - v_proj
-
-sequence_len: 512
-sample_packing: false
-pad_to_sequence_len: true
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-#only 2 epochs because of small dataset
-gradient_accumulation_steps: 3
-micro_batch_size: 2
-num_epochs: 2
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: true
-fp16: false
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_steps: 10
-evals_per_epoch: 4
-eval_table_size:
-eval_max_new_tokens: 128
-saves_per_epoch: 1
-debug:
-#default deepspeed, can use more aggresive if needed like zero2, zero3
-deepspeed: deepspeed_configs/zero1.json
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"
--- a/examples/mistral/Mistral-7b-example/data.jsonl
+++ b/examples/mistral/Mistral-7b-example/data.jsonl
@@ -1,10 +0,0 @@
-{"conversations": [{"from": "Customer", "value": "\"<Customer>: Who is the Founder of Apple\""}, {"from": "gpt", "value": "\"<Chatbot>: The founder of Apple is Steve Jobs\""}]}
-{"conversations": [{"from": "Customer", "value": "\"<Customer>: What is the capital of France?\""}, {"from": "gpt", "value": "\"<Chatbot>: The capital of France is Paris.\""}]}
-{"conversations": [{"from": "Customer", "value": "\"<Customer>: How far is the Moon from Earth?\""}, {"from": "gpt", "value": "\"<Chatbot>: The Moon is approximately 384,400 kilometers from Earth.\""}]}
-{"conversations": [{"from": "Customer", "value": "\"<Customer>: What is the tallest mountain in the world?\""}, {"from": "gpt", "value": "\"<Chatbot>: The tallest mountain in the world is Mount Everest.\""}]}
-{"conversations": [{"from": "Customer", "value": "\"<Customer>: Who wrote Romeo and Juliet?\""}, {"from": "gpt", "value": "\"<Chatbot>: Romeo and Juliet was written by William Shakespeare.\""}]}
-{"conversations": [{"from": "Customer", "value": "\"<Customer>: What is the boiling point of water?\""}, {"from": "gpt", "value": "\"<Chatbot>: The boiling point of water is 100 degrees Celsius.\""}]}
-{"conversations": [{"from": "Customer", "value": "\"<Customer>: When was the first man on the moon?\""}, {"from": "gpt", "value": "\"<Chatbot>: The first man landed on the moon in 1969.\""}]}
-{"conversations": [{"from": "Customer", "value": "\"<Customer>: What is the largest ocean?\""}, {"from": "gpt", "value": "\"<Chatbot>: The largest ocean is the Pacific Ocean.\""}]}
-{"conversations": [{"from": "Customer", "value": "\"<Customer>: Who invented the telephone?\""}, {"from": "gpt", "value": "\"<Chatbot>: The telephone was invented by Alexander Graham Bell.\""}]}
-{"conversations": [{"from": "Customer", "value": "\"<Customer>: What is the formula for water?\""}, {"from": "gpt", "value": "\"<Chatbot>: The chemical formula for water is H2O.\""}]}
--- a/examples/mistral/README.md
+++ b/examples/mistral/README.md
@@ -8,5 +8,5 @@ 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_configs/zero2.json
+accelerate launch -m axolotl.cli.train examples/mistral/config.yml --deepspeed deepspeed/zero2.json
 ```
--- a/examples/mistral/config.yml
+++ b/examples/mistral/config.yml
@@ -1,6 +1,7 @@
 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
@@ -33,8 +34,8 @@ learning_rate: 0.000005

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
@@ -48,7 +49,7 @@ flash_attention: true
 warmup_steps: 10
 evals_per_epoch: 4
 eval_table_size:
-eval_max_new_tokens: 128
+eval_table_max_new_tokens: 128
 saves_per_epoch: 1
 debug:
 deepspeed:
--- a/examples/mistral/lora-mps.yml
+++ b/examples/mistral/lora-mps.yml
@@ -1,79 +0,0 @@
-base_model: mistralai/Mistral-7B-v0.1
-model_type: MistralForCausalLM
-tokenizer_type: LlamaTokenizer
-
-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
-output_dir: ./lora-out
-eval_sample_packing: false
-
-adapter: lora
-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_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: 8
-micro_batch_size: 1
-num_epochs: 2
-optimizer: adamw_torch
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: auto
-fp16: false
-tf32: true
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: false
-sdp_attention: true
-
-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:
--- a/examples/mistral/mixtral-qlora-fsdp.yml
+++ b/examples/mistral/mixtral-qlora-fsdp.yml
@@ -1,74 +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: 0.02
-output_dir: ./qlora-out
-
-model_config:
-  output_router_logits: true
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 1024
-sample_packing: false
-pad_to_sequence_len: false
-
-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: 1
-optimizer: paged_adamw_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: auto
-fp16:
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-loss_watchdog_threshold: 5.0
-loss_watchdog_patience: 3
-
-warmup_steps: 10
-evals_per_epoch: 4
-eval_table_size:
-eval_max_new_tokens: 128
-saves_per_epoch: 1
-debug:
-weight_decay: 0.0
-fsdp:
-  - full_shard
-fsdp_config:
-  fsdp_transformer_layer_cls_to_wrap: MixtralSparseMoeBlock
-special_tokens:
--- a/examples/mistral/mixtral.yml
+++ b/examples/mistral/mixtral.yml
@@ -16,12 +16,12 @@ output_dir: ./qlora-out

 ## You can optionally freeze the entire model and unfreeze a subset of parameters
 unfrozen_parameters:
-#  - ^lm_head.weight$
-#  - ^model.embed_tokens.weight$[:32000]
-#  - 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
+#  - 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
@@ -63,8 +63,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
@@ -81,10 +81,10 @@ loss_watchdog_patience: 3
 warmup_steps: 10
 evals_per_epoch: 4
 eval_table_size:
-eval_max_new_tokens: 128
+eval_table_max_new_tokens: 128
 saves_per_epoch: 1
 debug:
-deepspeed: deepspeed_configs/zero2.json
+deepspeed: deepspeed/zero2.json
 weight_decay: 0.0
 fsdp:
 fsdp_config:
--- a/examples/mistral/qlora.yml
+++ b/examples/mistral/qlora.yml
@@ -1,6 +1,7 @@
 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
@@ -49,8 +50,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
@@ -67,7 +68,7 @@ loss_watchdog_patience: 3
 warmup_steps: 10
 evals_per_epoch: 4
 eval_table_size:
-eval_max_new_tokens: 128
+eval_table_max_new_tokens: 128
 saves_per_epoch: 1
 debug:
 deepspeed:
--- a/examples/mpt-7b/config.yml
+++ b/examples/mpt-7b/config.yml
@@ -33,7 +33,7 @@ lr_scheduler: cosine
 learning_rate: 0.0000002
 train_on_inputs: false
 group_by_length: false
-bf16: auto
+bf16: true
 tf32: true
 early_stopping_patience:
 resume_from_checkpoint:
--- a/examples/openllama-3b/lora.yml
+++ b/examples/openllama-3b/lora.yml
@@ -52,7 +52,6 @@ logging_steps: 1
 xformers_attention:
 flash_attention: true
 gptq_groupsize:
-s2_attention:
 gptq_model_v1:
 warmup_steps: 20
 evals_per_epoch: 4
--- a/examples/phi/README.md
+++ b/examples/phi/README.md
@@ -3,7 +3,7 @@
 Due to some nuances with the phi code, please use deepspeed when training phi for full finetune.

 ```shell
-accelerate launch -m axolotl.cli.train examples/phi/phi-ft.yml --deepspeed deepspeed_configs/zero1.json
+accelerate launch -m axolotl.cli.train examples/phi/phi-ft.yml --deepspeed deepspeed/zero1.json

 # OR

--- a/examples/phi/phi-ft.yml
+++ b/examples/phi/phi-ft.yml
@@ -1,6 +1,8 @@
 base_model: microsoft/phi-1_5
-model_type: AutoModelForCausalLM
+model_type: PhiForCausalLM
 tokenizer_type: AutoTokenizer
+is_llama_derived_model: false
+trust_remote_code: true

 load_in_8bit: false
 load_in_4bit: false
@@ -16,7 +18,7 @@ output_dir: ./phi-sft-out

 sequence_len: 2048
 sample_packing: true
-pad_to_sequence_len: true
+pad_to_sequence_len:

 adapter:
 lora_model_dir:
@@ -33,7 +35,7 @@ wandb_name:
 wandb_log_model:

 gradient_accumulation_steps: 1
-micro_batch_size: 2
+micro_batch_size: 1
 num_epochs: 4
 optimizer: adamw_torch
 adam_beta2: 0.95
@@ -43,20 +45,18 @@ lr_scheduler: cosine
 learning_rate: 0.000003

 train_on_inputs: false
-group_by_length: false
-bf16: auto
-fp16:
+group_by_length: true
+bf16: true
+fp16: false
 tf32: true

-gradient_checkpointing: true
-gradient_checkpointing_kwargs:
-  use_reentrant: True
+gradient_checkpointing:
 early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
 logging_steps: 1
 xformers_attention:
-flash_attention: true
+flash_attention:

 warmup_steps: 100
 evals_per_epoch: 4
@@ -68,4 +68,7 @@ fsdp:
 fsdp_config:
 resize_token_embeddings_to_32x: true
 special_tokens:
+  bos_token: "<|endoftext|>"
+  eos_token: "<|endoftext|>"
+  unk_token: "<|endoftext|>"
  pad_token: "<|endoftext|>"
--- a/examples/phi/phi-qlora.yml
+++ b/examples/phi/phi-qlora.yml
@@ -1,6 +1,8 @@
 base_model: microsoft/phi-1_5
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
+is_llama_derived_model: false
+trust_remote_code: true

 load_in_8bit: false
 load_in_4bit: true
@@ -14,9 +16,9 @@ dataset_prepared_path:
 val_set_size: 0.05
 output_dir: ./phi-sft-out

-sequence_len: 2048
-sample_packing: true
-pad_to_sequence_len: true
+sequence_len: 1024
+sample_packing: false  # not CURRENTLY compatible with LoRAs
+pad_to_sequence_len:

 adapter: qlora
 lora_model_dir:
@@ -33,7 +35,7 @@ wandb_name:
 wandb_log_model:

 gradient_accumulation_steps: 1
-micro_batch_size: 2
+micro_batch_size: 1
 num_epochs: 4
 optimizer: adamw_torch
 adam_beta2: 0.95
@@ -43,20 +45,18 @@ lr_scheduler: cosine
 learning_rate: 0.000003

 train_on_inputs: false
-group_by_length: false
-bf16: auto
-fp16:
+group_by_length: true
+bf16: true
+fp16: false
 tf32: true

-gradient_checkpointing: true
-gradient_checkpointing_kwargs:
-  use_reentrant: True
+gradient_checkpointing:
 early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
 logging_steps: 1
 xformers_attention:
-flash_attention: true
+flash_attention:

 warmup_steps: 100
 evals_per_epoch: 4
@@ -68,4 +68,7 @@ fsdp:
 fsdp_config:
 resize_token_embeddings_to_32x: true
 special_tokens:
+  bos_token: "<|endoftext|>"
+  eos_token: "<|endoftext|>"
+  unk_token: "<|endoftext|>"
  pad_token: "<|endoftext|>"
--- a/examples/phi/phi2-ft.yml
+++ b/examples/phi/phi2-ft.yml
@@ -1,6 +1,8 @@
 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
@@ -15,16 +17,19 @@ val_set_size: 0.05
 output_dir: ./phi-sft-out

 sequence_len: 2048
-sample_packing: true
-pad_to_sequence_len: true
+sample_packing: false  # currently unsupported
+pad_to_sequence_len:

 adapter:
 lora_model_dir:
-lora_r:
-lora_alpha:
-lora_dropout:
-lora_target_linear:
+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:
@@ -33,24 +38,22 @@ wandb_name:
 wandb_log_model:

 gradient_accumulation_steps: 1
-micro_batch_size: 2
+micro_batch_size: 1
 num_epochs: 4
-optimizer: adamw_torch
+optimizer: paged_adamw_8bit
 adam_beta2: 0.95
 adam_epsilon: 0.00001
 max_grad_norm: 1.0
 lr_scheduler: cosine
-learning_rate: 0.000003
+learning_rate: 1e-5

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: true

 gradient_checkpointing: true
-gradient_checkpointing_kwargs:
-  use_reentrant: True
 early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
--- a/examples/pythia/lora.yml
+++ b/examples/pythia/lora.yml
@@ -27,7 +27,7 @@ num_epochs: 4
 learning_rate: 0.00001
 train_on_inputs: false
 group_by_length: false
-bf16: auto
+bf16: true
 tf32: true
 early_stopping_patience:
 resume_from_checkpoint:
--- a/examples/qwen/lora.yml
+++ b/examples/qwen/lora.yml
@@ -2,6 +2,7 @@ base_model: Qwen/Qwen-7B
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer

+is_qwen_derived_model: true
 trust_remote_code: true

 load_in_8bit: true
@@ -42,8 +43,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: false
@@ -57,7 +58,7 @@ flash_attention:
 warmup_steps: 10
 evals_per_epoch: 4
 eval_table_size:
-eval_max_new_tokens: 128
+eval_table_max_new_tokens: 128
 saves_per_epoch: 1
 debug:
 deepspeed:
--- a/examples/qwen/qlora.yml
+++ b/examples/qwen/qlora.yml
@@ -2,6 +2,7 @@ base_model: Qwen/Qwen-7B
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer

+is_qwen_derived_model: true
 trust_remote_code: true

 load_in_8bit: false
@@ -42,8 +43,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: false
@@ -57,7 +58,7 @@ flash_attention:
 warmup_steps: 10
 evals_per_epoch: 4
 eval_table_size:
-eval_max_new_tokens: 128
+eval_table_max_new_tokens: 128
 saves_per_epoch: 1
 debug:
 deepspeed:
--- a/examples/redpajama/config-3b.yml
+++ b/examples/redpajama/config-3b.yml
@@ -34,7 +34,7 @@ lr_scheduler: cosine
 learning_rate: 0.0000002
 train_on_inputs: false
 group_by_length: false
-bf16: auto
+bf16: true
 tf32: true
 early_stopping_patience:
 resume_from_checkpoint:
--- a/examples/replit-3b/config-lora.yml
+++ b/examples/replit-3b/config-lora.yml
@@ -33,7 +33,7 @@ lr_scheduler:
 learning_rate: 0.00001
 train_on_inputs: false
 group_by_length: false
-bf16: auto
+bf16: true
 tf32: true
 gradient_checkpointing:
 early_stopping_patience:
--- a/examples/stablelm-2/1.6b/fft.yml
+++ b/examples/stablelm-2/1.6b/fft.yml
@@ -1,69 +0,0 @@
-base_model: stabilityai/stablelm-2-1_6b
-model_type: AutoModelForCausalLM
-tokenizer_type: AutoTokenizer
-trust_remote_code: 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: auto
-fp16:
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-flash_attn_cross_entropy: false
-flash_attn_rms_norm: true
-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_configs/zero2.json # multi-gpu only
-weight_decay: 0.1
-fsdp:
-fsdp_config:
-special_tokens:
--- a/examples/stablelm-2/1.6b/lora.yml
+++ b/examples/stablelm-2/1.6b/lora.yml
@@ -1,66 +0,0 @@
-base_model: stabilityai/stablelm-2-1_6b
-model_type: AutoModelForCausalLM
-tokenizer_type: AutoTokenizer
-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: 4096
-sample_packing: true
-pad_to_sequence_len: true
-
-adapter: lora
-lora_model_dir:
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-lora_fan_in_fan_out:
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 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: auto
-fp16:
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-flash_attn_cross_entropy: false
-flash_attn_rms_norm: true
-
-warmup_steps: 10
-evals_per_epoch: 4
-saves_per_epoch: 1
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
--- a/examples/stablelm-2/README.md
+++ b/examples/stablelm-2/README.md
@@ -1,36 +0,0 @@
-# StableLM 2
-
-This repository contains examples for training and processing using StableLM-2. It also includes a section to help you estimate the GPU requirements for your specific use case.
-
-## Estimating GPU Requirements
-
-| type          | deepspeed | batch size | context length | vRAM GPU (GBs) |
-|---------------|-----------|------------|----------------|----------------|
-| full finetune | N/A       | 1          | 4096           | ~21.5GBs       |
-| full finetune | zero2     | 1          | 4096           | ~20GBs         |
-| lora          | N/A       | 1          | 4096           | ~16.6GBs       |
-
-The above are estimates and might differ slight depending on the setup for example whether you pack your sequence lengths or not (the above assumes you do to length 4096).
-
-This blog post from Hamel Husain was a great resource for estimating these numbers: https://hamel.dev/notes/llm/03_estimating_vram.html
-
-## Training
-We have example scripts here for both full finetuning and lora using the popular alpaca dataset:
-
-```shell
-# preprocess the dataset
-CUDA_VISIBLE_DEVICES="" python -m axolotl.cli.preprocess examples/stablelm-2/1.6b/lora.yml
-```
-
-Single GPU Training:
-```shell
-python -m axolotl.cli.train examples/stablelm-2/fft.yml --deepspeed deepspeed_configs/zero2.json
-# OR
-python -m axolotl.cli.train examples/stablelm-2/1.6b/lora.yml
-```
-
-Multinode GPU Training with `accelerate`:
-```shell
-# make sure you've configured accelerate properly
-accelerate launch -m axolotl.cli.train examples/stablelm-2/1.6b/fft.yml --deepspeed deepspeed_configs/zero2.json
-```
--- a/examples/starcoder2/qlora.yml
+++ b/examples/starcoder2/qlora.yml
@@ -1,69 +0,0 @@
-base_model: bigcode/starcoder2-3b
-
-load_in_8bit: false
-load_in_4bit: true
-strict: false
-
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-
-
-dataset_prepared_path:
-val_set_size: 0.2
-output_dir: ./qlora
-
-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_modules:
-lora_target_linear: true
-lora_fan_in_fan_out:
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-
-gradient_accumulation_steps: 8
-micro_batch_size: 2
-num_epochs: 3
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 2e-5
-
-train_on_inputs: false
-group_by_length: false
-bf16: auto
-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: 20
-evals_per_epoch: 4
-eval_steps:
-eval_table_size:
-saves_per_epoch: 4
-save_steps:
-save_total_limit: 2
-debug:
-deepspeed:
-weight_decay:
-fsdp:
-fsdp_config:
-special_tokens:
--- a/examples/tiny-llama/lora-mps.yml
+++ b/examples/tiny-llama/lora-mps.yml
@@ -1,64 +0,0 @@
-base_model: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T
-model_type: LlamaForCausalLM
-tokenizer_type: LlamaTokenizer
-
-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
-output_dir: ./lora-out
-
-sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
-eval_sample_packing: false
-
-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_torch
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: auto
-fp16: false
-tf32: true
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: false
-
-warmup_steps: 10
-evals_per_epoch: 0
-saves_per_epoch: 1
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
--- a/examples/tiny-llama/lora.yml
+++ b/examples/tiny-llama/lora.yml
@@ -1,6 +1,7 @@
 base_model: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
+is_llama_derived_model: true

 load_in_8bit: true
 load_in_4bit: false
@@ -15,7 +16,6 @@ output_dir: ./lora-out

 sequence_len: 4096
 sample_packing: true
-eval_sample_packing: false
 pad_to_sequence_len: true

 adapter: lora
@@ -41,8 +41,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
--- a/examples/tiny-llama/pretrain.yml
+++ b/examples/tiny-llama/pretrain.yml
@@ -2,6 +2,7 @@ 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
@@ -11,7 +12,6 @@ max_steps: 200
 pretraining_dataset:
  path: c4
  name: en
-  type: pretrain
 dataset_prepared_path:
 val_set_size: 0.0
 output_dir: ./model-out
@@ -34,8 +34,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
--- a/examples/tiny-llama/qlora.yml
+++ b/examples/tiny-llama/qlora.yml
@@ -1,6 +1,7 @@
 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
@@ -42,8 +43,8 @@ learning_rate: 0.0002

 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false

 gradient_checkpointing: true
--- a/examples/xgen-7b/xgen-7b-8k-qlora.yml
+++ b/examples/xgen-7b/xgen-7b-8k-qlora.yml
@@ -62,8 +62,8 @@ lr_scheduler: cosine
 learning_rate: 0.00002
 train_on_inputs: false
 group_by_length: false
-bf16: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false
 gradient_checkpointing: true
 # stop training after this many evaluation losses have increased in a row
--- a/examples/yi-34B-chat/qlora.yml
+++ b/examples/yi-34B-chat/qlora.yml
@@ -1,13 +1,14 @@
 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: auto
-fp16:
+bf16: true
+fp16: false
 tf32: false
 flash_attention: true
 special_tokens:
@@ -28,7 +29,7 @@ num_epochs: 1
 val_set_size: 0.1
 evals_per_epoch: 5
 eval_table_size:
-eval_max_new_tokens: 128
+eval_table_max_new_tokens: 128
 eval_sample_packing: false
 eval_batch_size: 1

--- a/requirements-dev.txt
+++ b/requirements-dev.txt
@@ -1,4 +1,3 @@
 pre-commit
 black
 mypy
-types-requests
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,34 +1,35 @@
 --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/
 packaging==23.2
-peft==0.9.0
-transformers==4.38.2
+peft==0.7.0
+transformers @ git+https://github.com/huggingface/transformers.git@3cefac1d974db5e2825a0cb2b842883a628be7a0
 tokenizers==0.15.0
-bitsandbytes>=0.43.0
-accelerate==0.26.1
-deepspeed==0.13.1
-pydantic==2.6.3
+bitsandbytes>=0.41.1
+accelerate @ git+https://github.com/huggingface/accelerate.git@0d2280dadc6a93413a5496613b7fdda3a4d2551b
+deepspeed
 addict
 fire
 PyYAML>=6.0
-requests
 datasets>=2.15.0
-flash-attn==2.5.5
+flash-attn==2.3.3
 sentencepiece
 wandb
 einops
 xformers==0.0.22
-optimum==1.16.2
+optimum==1.13.2
 hf_transfer
 colorama
 numba
 numpy>=1.24.4
+mlflow
 # qlora things
-evaluate==0.4.1
+bert-score==0.3.13
+evaluate==0.4.0
+rouge-score==0.1.2
 scipy
 scikit-learn==1.2.2
 pynvml
 art
-fschat==0.2.36
+fschat==0.2.34
 gradio==3.50.2
 tensorboard

@@ -40,4 +41,3 @@ gcsfs
 # adlfs

 trl>=0.7.9
-fastcore>=1.5.29
--- a/scripts/cloud-entrypoint.sh
+++ b/scripts/cloud-entrypoint.sh
@@ -5,24 +5,16 @@ echo "Exporting environment variables..."
 printenv | grep -E '^RUNPOD_|^PATH=|^_=' | sed 's/^\(.*\)=\(.*\)$/export \1="\2"/' >> /etc/rp_environment
 echo 'source /etc/rp_environment' >> ~/.bashrc

-if [[ $PUBLIC_KEY ]]; then
-    # runpod
+if [[ $PUBLIC_KEY ]]
+then
    mkdir -p ~/.ssh
    chmod 700 ~/.ssh
    echo $PUBLIC_KEY >> ~/.ssh/authorized_keys
    chmod 700 -R ~/.ssh
    # Start the SSH service in the background
    service ssh start
-elif [ -n "$SSH_KEY" ]; then
-    # latitude.sh
-    mkdir -p ~/.ssh
-    chmod 700 ~/.ssh
-    echo $SSH_KEY >> ~/.ssh/authorized_keys
-    chmod 700 -R ~/.ssh
-    # Start the SSH service in the background
-    service ssh start
 else
-    echo "No PUBLIC_KEY or SSH_KEY environment variable provided, not starting openSSH daemon"
+    echo "No PUBLIC_KEY ENV variable provided, not starting openSSH daemon"
 fi

 # Check if JUPYTER_PASSWORD is set and not empty
@@ -33,7 +25,7 @@ fi

 if [ "$JUPYTER_DISABLE" != "1" ]; then
    # Run Jupyter Lab in the background
-    jupyter lab --port=8888 --ip=* --allow-root --ServerApp.allow_origin=* --ServerApp.preferred_dir=/workspace &
+    jupyter lab --allow-root --ip 0.0.0.0 &
 fi

 # Execute the passed arguments (CMD)
--- a/scripts/motd
+++ b/scripts/motd
@@ -1,17 +0,0 @@
-
-                                 dP            dP   dP
-                                 88            88   88
-      .d8888b. dP.  .dP .d8888b. 88 .d8888b. d8888P 88
-      88'  `88  `8bd8'  88'  `88 88 88'  `88   88   88
-      88.  .88  .d88b.  88.  .88 88 88.  .88   88   88
-      `88888P8 dP'  `dP `88888P' dP `88888P'   dP   dP
-
-Welcome to the axolotl cloud image! If the you've mounted a disk to /workspace and the axolotl directory ie empty, run the following commands:
-
-```
-cd /workspace
-rm -rf /workspace/axolotl
-git clone https://github.com/OpenAccess-AI-Collective/axolotl.git
-cd axolotl
-pip install --no-deps -e .
-```
--- a/setup.py
+++ b/setup.py
@@ -1,7 +1,5 @@
 """setup.py for axolotl"""

-import platform
-import re
 from importlib.metadata import PackageNotFoundError, version

 from setuptools import find_packages, setup
@@ -18,7 +16,6 @@ def parse_requirements():
                or "flash-attention" in line
                or "deepspeed" in line
                or "mamba-ssm" in line
-                or "lion-pytorch" in line
            )
            if line.startswith("--extra-index-url"):
                # Handle custom index URLs
@@ -29,25 +26,10 @@ def parse_requirements():
                _install_requires.append(line)

    try:
-        if "Darwin" in platform.system():
+        torch_version = version("torch")
+        if torch_version.startswith("2.1.1"):
            _install_requires.pop(_install_requires.index("xformers==0.0.22"))
-        else:
-            torch_version = version("torch")
-            _install_requires.append(f"torch=={torch_version}")
-
-            version_match = re.match(r"^(\d+)\.(\d+)(?:\.(\d+))?", torch_version)
-            if version_match:
-                major, minor, patch = version_match.groups()
-                major, minor = int(major), int(minor)
-                patch = (
-                    int(patch) if patch is not None else 0
-                )  # Default patch to 0 if not present
-            else:
-                raise ValueError("Invalid version format")
-
-            if (major, minor) >= (2, 1):
-                _install_requires.pop(_install_requires.index("xformers==0.0.22"))
-                _install_requires.append("xformers>=0.0.23")
+            _install_requires.append("xformers==0.0.23")
    except PackageNotFoundError:
        pass

@@ -59,7 +41,7 @@ install_requires, dependency_links = parse_requirements()

 setup(
    name="axolotl",
-    version="0.4.0",
+    version="0.3.0",
    description="LLM Trainer",
    long_description="Axolotl is a tool designed to streamline the fine-tuning of various AI models, offering support for multiple configurations and architectures.",
    package_dir={"": "src"},
@@ -68,14 +50,13 @@ setup(
    dependency_links=dependency_links,
    extras_require={
        "flash-attn": [
-            "flash-attn==2.5.5",
+            "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",
        ],
        "deepspeed": [
-            "deepspeed==0.13.1",
-            "deepspeed-kernels",
+            "deepspeed",
        ],
        "mamba-ssm": [
            "mamba-ssm==1.0.1",
@@ -83,11 +64,5 @@ setup(
        "auto-gptq": [
            "auto-gptq==0.5.1",
        ],
-        "mlflow": [
-            "mlflow",
-        ],
-        "lion-pytorch": [
-            "lion-pytorch==0.1.2",
-        ],
    },
 )
--- a/src/axolotl/cli/init.py
+++ b/src/axolotl/cli/init.py
@@ -1,29 +1,26 @@
 """Prepare and train a model on a dataset. Can also infer from a model or merge lora"""

 import importlib
-import json
 import logging
 import math
 import os
 import random
 import sys
-import tempfile
 from pathlib import Path
 from threading import Thread
 from typing import Any, Dict, List, Optional, Union
-from urllib.parse import urlparse

-import requests
+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.utils import is_torch_bf16_gpu_available

 from axolotl.common.cli import TrainerCliArgs, load_model_and_tokenizer
 from axolotl.logging_config import configure_logging
@@ -33,7 +30,7 @@ from axolotl.utils.config import (
    normalize_config,
    validate_config,
 )
-from axolotl.utils.data import load_prepare_dpo_datasets, prepare_dataset
+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
@@ -63,52 +60,6 @@ def print_axolotl_text_art(suffix=None):
        print(ascii_art)


-def check_remote_config(config: Union[str, Path]):
-    # Check if the config is a valid HTTPS URL to a .yml or .yaml file
-    if not (isinstance(config, str) and config.startswith("https://")):
-        return config  # Return the original value if it's not a valid URL
-
-    filename = os.path.basename(urlparse(config).path)
-    temp_dir = tempfile.mkdtemp()
-
-    try:
-        response = requests.get(config, timeout=30)
-        response.raise_for_status()  # Check for HTTP errors
-
-        content = response.content
-        try:
-            # Try parsing as JSON first to catch cases where JSON content is mistakenly considered YAML
-            json.loads(content)
-            # Log a warning but do not raise an error; JSON is technically valid YAML - this can happen when you forget to point to a raw github link
-            LOG.warning(
-                f"Warning: The content of the file at {config} is JSON, which is technically valid YAML but might not be intended."
-            )
-        except json.JSONDecodeError:
-            # If it's not valid JSON, verify it's valid YAML
-            try:
-                yaml.safe_load(content)
-            except yaml.YAMLError as err:
-                raise ValueError(
-                    f"Failed to parse the content at {config} as YAML: {err}"
-                ) from err
-
-        # Write the content to a file if it's valid YAML (or JSON treated as YAML)
-        output_path = Path(temp_dir) / filename
-        with open(output_path, "wb") as file:
-            file.write(content)
-        LOG.info(
-            f"Using the following config obtained from {config}:\n\n{content.decode('utf-8')}\n"
-        )
-        return output_path
-
-    except requests.RequestException as err:
-        # This catches all requests-related exceptions including HTTPError
-        raise RuntimeError(f"Failed to download {config}: {err}") from err
-    except Exception as err:
-        # Catch-all for any other exceptions
-        raise err
-
-
 def get_multi_line_input() -> Optional[str]:
    print("Give me an instruction (Ctrl + D to submit): ")
    instruction = ""
@@ -128,11 +79,7 @@ def do_merge_lora(

    LOG.info("running merge of LoRA with base model")
    model = model.merge_and_unload(progressbar=True)
-    try:
-        model.to(dtype=cfg.torch_dtype)
-    except RuntimeError:
-        pass
-    model.generation_config.do_sample = True
+    model.to(dtype=cfg.torch_dtype)

    if cfg.local_rank == 0:
        LOG.info(f"saving merged model to: {str(Path(cfg.output_dir) / 'merged')}")
@@ -214,8 +161,6 @@ def do_inference_gradio(
    cfg: DictDefault,
    cli_args: TrainerCliArgs,
 ):
-    import gradio as gr
-
    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>"}
@@ -322,14 +267,14 @@ def check_not_in(list1: List[str], list2: Union[Dict[str, Any], List[str]]) -> b
    return not any(el in list2 for el in list1)


-def load_cfg(config: Union[str, Path] = Path("examples/"), **kwargs):
-    config = check_remote_config(config)
+def load_cfg(config: Path = Path("examples/"), **kwargs):
    if Path(config).is_dir():
-        config = choose_config(Path(config))
+        config = choose_config(config)

    # load the config from the yaml file
    with open(config, encoding="utf-8") as file:
        cfg: DictDefault = DictDefault(yaml.safe_load(file))
+    cfg.axolotl_config_path = config
    # if there are any options passed in the cli, if it is something that seems valid from the yaml,
    # then overwrite the value
    cfg_keys = cfg.keys()
@@ -342,22 +287,7 @@ def load_cfg(config: Union[str, Path] = Path("examples/"), **kwargs):
            else:
                cfg[k] = kwargs[k]

-    cfg.axolotl_config_path = config
-
-    try:
-        device_props = torch.cuda.get_device_properties("cuda")
-        gpu_version = "sm_" + str(device_props.major) + str(device_props.minor)
-    except:  # pylint: disable=bare-except # noqa: E722
-        gpu_version = None
-
-    cfg = validate_config(
-        cfg,
-        capabilities={
-            "bf16": is_torch_bf16_gpu_available(),
-            "n_gpu": os.environ.get("WORLD_SIZE", 1),
-            "compute_capability": gpu_version,
-        },
-    )
+    validate_config(cfg)

    prepare_optim_env(cfg)

@@ -413,7 +343,78 @@ def load_rl_datasets(
    cfg: DictDefault,
    cli_args: TrainerCliArgs,  # pylint: disable=unused-argument
 ) -> TrainDatasetMeta:
-    train_dataset, eval_dataset = load_prepare_dpo_datasets(cfg)
+    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)
    )
@@ -433,13 +434,6 @@ def check_accelerate_default_config():


 def check_user_token():
-    # Skip check if HF_HUB_OFFLINE is set to True
-    if os.getenv("HF_HUB_OFFLINE") == "1":
-        LOG.info(
-            "Skipping HuggingFace token verification because HF_HUB_OFFLINE is set to True. Only local files will be used."
-        )
-        return True
-
    # Verify if token is valid
    api = HfApi()
    try:
--- a/src/axolotl/cli/preprocess.py
+++ b/src/axolotl/cli/preprocess.py
@@ -3,7 +3,6 @@ CLI to run training on a model
 """
 import logging
 from pathlib import Path
-from typing import Union

 import fire
 import transformers
@@ -14,21 +13,18 @@ from axolotl.cli import (
    check_user_token,
    load_cfg,
    load_datasets,
-    load_rl_datasets,
    print_axolotl_text_art,
 )
 from axolotl.common.cli import PreprocessCliArgs
 from axolotl.common.const import DEFAULT_DATASET_PREPARED_PATH
-from axolotl.prompt_strategies.sharegpt import register_chatml_template

 LOG = logging.getLogger("axolotl.cli.preprocess")


-def do_cli(config: Union[Path, str] = Path("examples/"), **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.is_preprocess = True
    check_accelerate_default_config()
    check_user_token()
    parser = transformers.HfArgumentParser((PreprocessCliArgs))
@@ -36,14 +32,6 @@ def do_cli(config: Union[Path, str] = Path("examples/"), **kwargs):
        return_remaining_strings=True
    )

-    if parsed_cfg.chat_template == "chatml" and parsed_cfg.default_system_message:
-        LOG.info(
-            f"ChatML set. Adding default system message: {parsed_cfg.default_system_message}"
-        )
-        register_chatml_template(parsed_cfg.default_system_message)
-    else:
-        register_chatml_template()
-
    if not parsed_cfg.dataset_prepared_path:
        msg = (
            Fore.RED
@@ -54,11 +42,7 @@ def do_cli(config: Union[Path, str] = Path("examples/"), **kwargs):
        LOG.warning(msg)
        parsed_cfg.dataset_prepared_path = DEFAULT_DATASET_PREPARED_PATH

-    if parsed_cfg.rl:
-        load_rl_datasets(cfg=parsed_cfg, cli_args=parsed_cli_args)
-    else:
-        load_datasets(cfg=parsed_cfg, cli_args=parsed_cli_args)
-
+    _ = 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}`"
--- a/src/axolotl/cli/shard.py
+++ b/src/axolotl/cli/shard.py
@@ -3,7 +3,6 @@ CLI to shard a trained model into 10GiB chunks
 """
 import logging
 from pathlib import Path
-from typing import Union

 import fire
 import transformers
@@ -26,7 +25,7 @@ def shard(
    model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)


-def do_cli(config: Union[Path, str] = Path("examples/"), **kwargs):
+def do_cli(config: Path = Path("examples/"), **kwargs):
    # pylint: disable=duplicate-code
    print_axolotl_text_art()
    parsed_cfg = load_cfg(config, **kwargs)
--- a/src/axolotl/cli/train.py
+++ b/src/axolotl/cli/train.py
@@ -3,12 +3,9 @@ CLI to run training on a model
 """
 import logging
 from pathlib import Path
-from typing import Tuple, Union

 import fire
-from transformers.hf_argparser import HfArgumentParser
-from transformers.modeling_utils import PreTrainedModel
-from transformers.tokenization_utils import PreTrainedTokenizer
+import transformers

 from axolotl.cli import (
    check_accelerate_default_config,
@@ -19,40 +16,27 @@ from axolotl.cli import (
    print_axolotl_text_art,
 )
 from axolotl.common.cli import TrainerCliArgs
-from axolotl.prompt_strategies.sharegpt import register_chatml_template
 from axolotl.train import train

 LOG = logging.getLogger("axolotl.cli.train")


-def do_cli(config: Union[Path, str] = Path("examples/"), **kwargs):
+def do_cli(config: Path = Path("examples/"), **kwargs):
    # pylint: disable=duplicate-code
    parsed_cfg = load_cfg(config, **kwargs)
-    parser = HfArgumentParser((TrainerCliArgs))
-    parsed_cli_args, _ = parser.parse_args_into_dataclasses(
-        return_remaining_strings=True
-    )
-    return do_train(parsed_cfg, parsed_cli_args)
-
-
-def do_train(cfg, cli_args) -> Tuple[PreTrainedModel, PreTrainedTokenizer]:
    print_axolotl_text_art()
    check_accelerate_default_config()
    check_user_token()
-    if cfg.chat_template == "chatml" and cfg.default_system_message:
-        LOG.info(
-            f"ChatML set. Adding default system message: {cfg.default_system_message}"
-        )
-        register_chatml_template(cfg.default_system_message)
-    else:
-        register_chatml_template()
+    parser = transformers.HfArgumentParser((TrainerCliArgs))
+    parsed_cli_args, _ = parser.parse_args_into_dataclasses(
+        return_remaining_strings=True
+    )

-    if cfg.rl:
-        dataset_meta = load_rl_datasets(cfg=cfg, cli_args=cli_args)
+    if parsed_cfg.rl:
+        dataset_meta = load_rl_datasets(cfg=parsed_cfg, cli_args=parsed_cli_args)
    else:
-        dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
-
-    return train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
+        dataset_meta = load_datasets(cfg=parsed_cfg, cli_args=parsed_cli_args)
+    train(cfg=parsed_cfg, cli_args=parsed_cli_args, dataset_meta=dataset_meta)


 if __name__ == "__main__":
--- a/src/axolotl/common/cli.py
+++ b/src/axolotl/common/cli.py
@@ -6,7 +6,6 @@ import logging
 from dataclasses import dataclass, field
 from typing import Optional

-import axolotl.monkeypatch.data.batch_dataset_fetcher  # pylint: disable=unused-import  # noqa: F401
 from axolotl.logging_config import configure_logging
 from axolotl.utils.dict import DictDefault
 from axolotl.utils.models import load_model, load_tokenizer
--- a/src/axolotl/core/policies/init.py
+++ b/src/axolotl/core/policies/init.py
--- a/src/axolotl/core/policies/auto_wrap.py
+++ b/src/axolotl/core/policies/auto_wrap.py
@@ -1,55 +0,0 @@
-"""module for building the auto wrap policy for FSDP"""
-import functools
-
-from peft import PrefixEncoder, PromptEmbedding, PromptEncoder
-from torch.distributed.fsdp.wrap import (
-    _or_policy,
-    lambda_auto_wrap_policy,
-    transformer_auto_wrap_policy,
-)
-from transformers.models.llama.modeling_llama import LlamaDecoderLayer
-from transformers.models.mistral.modeling_mistral import MistralDecoderLayer
-from transformers.models.mixtral.modeling_mixtral import MixtralDecoderLayer
-
-SUPPORTED_AUTO_WRAP_MODEL_TYPES = [
-    "llama",
-    "mistral",
-    "mixtral",
-]
-
-
-def get_wrapping_policy_factory(model_type):
-    if model_type == "llama":
-        layer_to_wrap = LlamaDecoderLayer
-    elif model_type == "mistral":
-        layer_to_wrap = MistralDecoderLayer
-    elif model_type == "mixtral":
-        layer_to_wrap = MixtralDecoderLayer
-
-    def get_wrapping_policy():
-        """This checks for lora layers (has weight and requires_grad)"""
-
-        def lambda_policy_fn(module):
-            return (
-                len(list(module.named_children())) == 0
-                and getattr(module, "weight", None) is not None
-                and module.weight.requires_grad
-            )
-
-        lambda_policy = functools.partial(
-            lambda_auto_wrap_policy, lambda_fn=lambda_policy_fn
-        )
-        transformer_layer_name = layer_to_wrap
-        transformer_wrap_policy = functools.partial(
-            transformer_auto_wrap_policy,
-            transformer_layer_cls=(
-                PrefixEncoder,
-                PromptEncoder,
-                PromptEmbedding,
-                transformer_layer_name,
-            ),
-        )
-        policies = [lambda_policy, transformer_wrap_policy]
-        return functools.partial(_or_policy, policies=policies)
-
-    return get_wrapping_policy
--- a/src/axolotl/core/trainer_builder.py
+++ b/src/axolotl/core/trainer_builder.py
@@ -5,38 +5,24 @@ Builder for the training args and trainer

 import abc
 import importlib
-import importlib.util
 import logging
 import math
-import os
 import sys
 from abc import abstractmethod
 from dataclasses import dataclass, field
 from functools import wraps
 from pathlib import Path
-from typing import List, Optional, Type, Union
+from typing import Optional

 import torch
 import transformers
-from accelerate import FullyShardedDataParallelPlugin
-from accelerate.utils import str_to_bool
 from datasets import Dataset
-from torch.distributed.fsdp import MixedPrecision
 from torch.optim.lr_scheduler import OneCycleLR
 from torch.utils.data import BatchSampler, DataLoader, RandomSampler, SequentialSampler
-from transformers import (
-    EarlyStoppingCallback,
-    Trainer,
-    TrainerCallback,
-    TrainingArguments,
-)
+from transformers import EarlyStoppingCallback, Trainer, TrainingArguments
 from transformers.trainer_utils import seed_worker
-from transformers.utils import is_sagemaker_mp_enabled
 from trl import DPOTrainer

-from axolotl.core.policies.auto_wrap import get_wrapping_policy_factory
-from axolotl.loraplus import create_loraplus_optimizer
-from axolotl.monkeypatch.multipack import SUPPORTED_MULTIPACK_MODEL_TYPES
 from axolotl.monkeypatch.relora import ReLoRACallback, ReLoRAScheduler
 from axolotl.utils.callbacks import (
    EvalFirstStepCallback,
@@ -45,25 +31,19 @@ from axolotl.utils.callbacks import (
    SaveAxolotlConfigtoWandBCallback,
    SaveBetterTransformerModelCallback,
    bench_eval_callback_factory,
-    causal_lm_bench_eval_callback_factory,
    log_prediction_callback_factory,
 )
 from axolotl.utils.collators import (
    BatchSamplerDataCollatorForSeq2Seq,
    DataCollatorForSeq2Seq,
    MambaDataCollator,
-    V2BatchSamplerDataCollatorForSeq2Seq,
 )
 from axolotl.utils.samplers import MultipackBatchSampler, get_dataset_lengths
 from axolotl.utils.schedulers import (
    get_cosine_schedule_with_min_lr,
    get_cosine_schedule_with_quadratic_warmup,
-    get_cosine_schedule_with_warmup_decay_constant,
 )

-if is_sagemaker_mp_enabled():
-    import smdistributed.modelparallel.torch as smp
-
 try:
    import torch._dynamo  # pylint: disable=ungrouped-imports
 except ImportError:
@@ -72,26 +52,6 @@ except ImportError:
 LOG = logging.getLogger("axolotl.core.trainer_builder")


-def is_mlflow_available():
-    return importlib.util.find_spec("mlflow") is not None
-
-
-def _sanitize_kwargs_for_tagging(tag_names, kwargs=None):
-    if isinstance(tag_names, str):
-        tag_names = [tag_names]
-
-    if kwargs is not None:
-        if "tags" not in kwargs:
-            kwargs["tags"] = tag_names
-        elif "tags" in kwargs and isinstance(kwargs["tags"], list):
-            kwargs["tags"].extend(tag_names)
-        elif "tags" in kwargs and isinstance(kwargs["tags"], str):
-            tag_names.append(kwargs["tags"])
-            kwargs["tags"] = tag_names
-
-    return kwargs
-
-
@dataclass
 class AxolotlTrainingArguments(TrainingArguments):
    """
@@ -115,10 +75,6 @@ class AxolotlTrainingArguments(TrainingArguments):
        default=False,
        metadata={"help": "Use sample packing for efficient training."},
    )
-    multipack_real_batches: bool = field(
-        default=False,
-        metadata={"help": "Use real batches for efficient training."},
-    )
    eval_sample_packing: Optional[bool] = field(
        default=None,
        metadata={"help": "Use sample packing for efficient evals."},
@@ -143,14 +99,6 @@ class AxolotlTrainingArguments(TrainingArguments):
        default=None,
        metadata={"help": "how many warmup steps to take after reset for ReLoRA"},
    )
-    relora_anneal_steps: Optional[int] = field(
-        default=None,
-        metadata={"help": "how many warmup steps to take after reset for ReLoRA"},
-    )
-    relora_prune_ratio: Optional[float] = field(
-        default=0.9,
-        metadata={"help": "prune ratio for magnitude pruning of the optimizer"},
-    )
    bench_split: Optional[str] = field(
        default="eval", metadata={"help": "The benchmark split to run on"}
    )
@@ -163,9 +111,6 @@ class AxolotlTrainingArguments(TrainingArguments):
    do_bench_eval: Optional[bool] = field(
        default=False, metadata={"help": "Whether to run the Benchmark evaluation."}
    )
-    do_causal_lm_eval: Optional[bool] = field(
-        default=False, metadata={"help": "Whether to run the Causal LM evaluation."}
-    )
    max_bench_samples: Optional[int] = field(
        default=None,
        metadata={
@@ -183,23 +128,6 @@ class AxolotlTrainingArguments(TrainingArguments):
        default=None,
        metadata={"help": "Minimum learning rate is min_lr_ratio * learning_rate"},
    )
-    cosine_constant_lr_ratio: Optional[float] = field(
-        default=None,
-        metadata={
-            "help": "Starting constant learning rate step is cosine_constant_lr_ratio * max_steps"
-        },
-    )
-    loraplus_lr_ratio: Optional[float] = field(
-        default=None, metadata={"help": "loraplus learning rate ratio lr_B / lr_A."}
-    )
-    loraplus_lr_embedding: Optional[float] = field(
-        default=1e-6,
-        metadata={"help": "loraplus learning rate for lora embedding layers."},
-    )
-    qlora: bool = field(
-        default=False,
-        metadata={"help": "whether this is a qlora training"},
-    )


 class AxolotlTrainer(Trainer):
@@ -224,33 +152,6 @@ class AxolotlTrainer(Trainer):
        super().__init__(*_args, **kwargs)
        self.train_data_collator = self.data_collator

-    def create_optimizer(self):
-        if self.args.loraplus_lr_ratio is None:
-            return super().create_optimizer()
-
-        opt_model = self.model_wrapped if is_sagemaker_mp_enabled() else self.model
-        if self.optimizer is None:  # pylint: disable=access-member-before-definition
-            optimizer_cls, optimizer_kwargs = Trainer.get_optimizer_cls_and_kwargs(
-                self.args,
-            )
-
-            loraplus_lr_ratio = getattr(self.args, "loraplus_lr_ratio", None)
-            loraplus_lr_embedding = getattr(self.args, "loraplus_lr_embedding", None)
-            self.optimizer = create_loraplus_optimizer(  # pylint: disable=attribute-defined-outside-init
-                opt_model,
-                optimizer_cls,
-                optimizer_kwargs,
-                loraplus_lr_ratio,
-                loraplus_lr_embedding,
-            )
-
-        if is_sagemaker_mp_enabled():
-            self.optimizer = smp.DistributedOptimizer(  # pylint: disable=attribute-defined-outside-init
-                self.optimizer
-            )
-
-        return self.optimizer
-
    def create_scheduler(
        self, num_training_steps: int, optimizer: torch.optim.Optimizer = None
    ):
@@ -262,40 +163,24 @@ class AxolotlTrainer(Trainer):
            num_training_steps (int): The number of training steps to do.
            optimizer (torch.optim.Optimizer): The training optimizer
        """
-        use_cosine_quadratic = (
-            self.args.lr_scheduler_type == "cosine"
-            and self.args.lr_quadratic_warmup is True
-        )
-
-        use_cosine_min_lr = (
-            self.args.lr_scheduler_type == "cosine"
-            and self.args.cosine_min_lr_ratio is not None
-        )

        # fmt: off
        if self.lr_scheduler is None:  # type: ignore  # pylint: disable=access-member-before-definition
            # fmt: on
-            if use_cosine_quadratic:
-                if use_cosine_min_lr:
-                    LOG.warning("Both cosine quadratic warmup and min lr detected. Using quadratic warmup.")
-
+            if (
+                self.args.lr_scheduler_type == "cosine"
+                and self.args.lr_quadratic_warmup is True
+            ):
                self.lr_scheduler = get_cosine_schedule_with_quadratic_warmup(  # pylint: disable=attribute-defined-outside-init
                    optimizer,
                    num_warmup_steps=self.args.get_warmup_steps(num_training_steps),
                    num_training_steps=num_training_steps,
                )
-            elif self.args.cosine_min_lr_ratio and self.args.cosine_constant_lr_ratio and use_cosine_min_lr:
-                assert 0 <= self.args.cosine_min_lr_ratio <= 1.0, "cosine_min_lr_ratio must be between 0.0 and 1.0"
-                assert 0 <= self.args.cosine_constant_lr_ratio <= 1.0, "cosine_constant_lr_ratio must be between 0.0 and 1.0"
-                self.lr_scheduler = get_cosine_schedule_with_warmup_decay_constant(  # pylint: disable=attribute-defined-outside-init
-                    optimizer,
-                    num_warmup_steps=self.args.get_warmup_steps(num_training_steps),
-                    num_training_steps=num_training_steps,
-                    min_lr_ratio=self.args.cosine_min_lr_ratio,
-                    constant_lr_ratio=self.args.cosine_constant_lr_ratio,
-                )
-            elif self.args.cosine_min_lr_ratio and use_cosine_min_lr:
+            elif self.args.lr_scheduler_type == "cosine" and self.args.cosine_min_lr_ratio is not None:
                assert 0 <= self.args.cosine_min_lr_ratio <= 1.0, "cosine_min_lr_ratio must be between 0.0 and 1.0"
+                if self.args.deepspeed:
+                    LOG.warning("Using cosine scheduler with deepspeed. This may be ignored if a scheduler is set \
+                                in the deepspeed JSON")
                self.lr_scheduler = get_cosine_schedule_with_min_lr(  # pylint: disable=attribute-defined-outside-init
                    optimizer,
                    num_warmup_steps=self.args.get_warmup_steps(num_training_steps),
@@ -304,30 +189,15 @@ class AxolotlTrainer(Trainer):
                )
            else:
                return super().create_scheduler(num_training_steps, optimizer)
-        else:
-            if use_cosine_quadratic:
-                LOG.warning("axolotl's cosine scheduler with quadratic warmup not used (e.g., because of deepspeed).")
-
-            if use_cosine_min_lr:
-                LOG.warning("axolotl's cosine scheduler with min lr not used (e.g., because of deepspeed).")
-
        return self.lr_scheduler

    def _get_train_sampler(self) -> Optional[torch.utils.data.Sampler]:
        if self.args.sample_packing and not self.args.pretraining:
-            if self.args.multipack_real_batches:
-                batch_size = self.args.per_device_train_batch_size
-                batch_max_len = self.args.max_seq_length
-            else:
-                batch_size = 1
-                batch_max_len = (
-                    self.args.per_device_train_batch_size * self.args.max_seq_length
-                )
            return MultipackBatchSampler(
                RandomSampler(self.train_dataset),
-                batch_size=batch_size,
+                self.args.train_batch_size,
                drop_last=True,
-                batch_max_len=batch_max_len,
+                batch_max_len=self._train_batch_size * self.args.max_seq_length,
                lengths=get_dataset_lengths(self.train_dataset),
                packing_efficiency_estimate=self.args.sample_packing_efficiency,
            )
@@ -337,19 +207,11 @@ class AxolotlTrainer(Trainer):
        self, eval_dataset: Dataset
    ) -> Optional[torch.utils.data.Sampler]:
        if self.args.sample_packing and self.args.eval_sample_packing is not False:
-            if self.args.multipack_real_batches:
-                batch_size = self.args.per_device_eval_batch_size
-                batch_max_len = self.args.max_seq_length
-            else:
-                batch_size = 1
-                batch_max_len = (
-                    self.args.per_device_eval_batch_size * self.args.max_seq_length
-                )
            return MultipackBatchSampler(
                SequentialSampler(eval_dataset),
-                batch_size=batch_size,
+                self.args.per_device_eval_batch_size,
                drop_last=True,
-                batch_max_len=batch_max_len,
+                batch_max_len=self.args.eval_batch_size * self.args.max_seq_length,
                lengths=get_dataset_lengths(eval_dataset),
                packing_efficiency_estimate=self.args.sample_packing_efficiency,
            )
@@ -358,8 +220,7 @@ class AxolotlTrainer(Trainer):
    def get_train_dataloader(self) -> DataLoader:
        if self.args.sample_packing and not self.args.pretraining:
            train_dataset = self.train_dataset
-            if "length" in train_dataset.features.keys():
-                train_dataset = train_dataset.remove_columns(["length"])
+            train_dataset = train_dataset.remove_columns(["length"])
            data_collator = self.data_collator
            dataloader_params = {
                "batch_size": self._train_batch_size,
@@ -467,65 +328,32 @@ class AxolotlTrainer(Trainer):
        #     return (loss, outputs) if return_outputs else loss
        return super().compute_loss(model, inputs, return_outputs=return_outputs)

+    def _sanitize_kwargs_for_tagging(self, tag_names, kwargs=None):
+        if isinstance(tag_names, str):
+            tag_names = [tag_names]
+
+        if kwargs is not None:
+            if "tags" not in kwargs:
+                kwargs["tags"] = tag_names
+            elif "tags" in kwargs and isinstance(kwargs["tags"], list):
+                kwargs["tags"].extend(tag_names)
+            elif "tags" in kwargs and isinstance(kwargs["tags"], str):
+                tag_names.append(kwargs["tags"])
+                kwargs["tags"] = tag_names
+
+        return kwargs
+
    @wraps(Trainer.push_to_hub)
    def push_to_hub(self, *args, **kwargs) -> str:
        """
        Overwrite the `push_to_hub` method in order to force-add the tags when pushing the
        model on the Hub. Please refer to `~transformers.Trainer.push_to_hub` for more details.
        """
-        kwargs = _sanitize_kwargs_for_tagging(tag_names=self.tag_names, kwargs=kwargs)
-
-        return super().push_to_hub(*args, **kwargs)
-
-    @wraps(Trainer.create_accelerator_and_postprocess)
-    def create_accelerator_and_postprocess(self):
-        rank = int(os.environ.get("LOCAL_RANK", 0))
-        res = super().create_accelerator_and_postprocess()
-
-        if self.args.qlora is False:
-            return res
-
-        # the rest of this method override is specific to fsdp + qlora (for now)
-        sync_module_states = (
-            str_to_bool(os.environ.get("FSDP_SYNC_MODULE_STATES", "True")) == 1
+        kwargs = self._sanitize_kwargs_for_tagging(
+            tag_names=self.tag_names, kwargs=kwargs
        )

-        mp_policy = None
-        amp = os.environ["ACCELERATE_MIXED_PRECISION"]
-        if amp == "fp16":
-            mp_policy = MixedPrecision(
-                param_dtype=torch.float32,
-                reduce_dtype=torch.float32,
-                buffer_dtype=torch.float32,
-            )
-        elif amp == "bf16":
-            mp_policy = MixedPrecision(
-                param_dtype=torch.float32,
-                reduce_dtype=torch.float32,
-                buffer_dtype=torch.float32,
-            )
-
-        # If somehow we figure out how we want to parameterize we want to autocast buffers...
-        # mp_policy = MixedPrecision(param_dtype=torch.bfloat16, reduce_dtype=torch.bfloat16, buffer_dtype=torch.float32)
-        # load_param_skip_names = ['inv_freq']
-
-        if self.is_fsdp_enabled:
-            wrapping_policy = get_wrapping_policy_factory(self.args.model_type)
-            fsdp_plugin = FullyShardedDataParallelPlugin(
-                auto_wrap_policy=wrapping_policy(),
-                cpu_offload=False,
-                use_orig_params=False,
-                limit_all_gathers=True,
-                param_init_fn=lambda module: module.to_empty(
-                    device=torch.device("cuda"), recurse=False
-                )
-                if (rank != 0 and sync_module_states)
-                else None,
-                mixed_precision_policy=mp_policy,
-            )
-            self.accelerator.state.fsdp_plugin = fsdp_plugin
-
-        return res
+        return super().push_to_hub(*args, **kwargs)


 class AxolotlMambaTrainer(AxolotlTrainer):
@@ -610,14 +438,10 @@ class ReLoRATrainer(AxolotlTrainer):
            warmup_steps = (
                self.args.relora_warmup_steps if self.args.relora_warmup_steps else 10
            )
-            anneal_steps = (
-                self.args.relora_anneal_steps if self.args.relora_anneal_steps else 1
-            )
            self.lr_scheduler = ReLoRAScheduler(
                optimizer,
                lr_scheduler,
                self.args.relora_steps,
-                anneal_steps,
                warmup_steps,
            )
        else:
@@ -626,24 +450,6 @@ class ReLoRATrainer(AxolotlTrainer):
        return self.lr_scheduler


-class AxolotlDPOTrainer(DPOTrainer):
-    """
-    Extend the base DPOTrainer for axolotl helpers
-    """
-
-    tag_names = ["axolotl", "dpo"]
-
-    @wraps(DPOTrainer.push_to_hub)
-    def push_to_hub(self, *args, **kwargs) -> str:
-        """
-        Overwrite the `push_to_hub` method in order to force-add the tags when pushing the
-        model on the Hub. Please refer to `~transformers.Trainer.push_to_hub` for more details.
-        """
-        kwargs = _sanitize_kwargs_for_tagging(tag_names=self.tag_names, kwargs=kwargs)
-
-        return super().push_to_hub(*args, **kwargs)
-
-
 class TrainerBuilderBase(abc.ABC):
    """
    Base class for trainer builder
@@ -652,7 +458,6 @@ class TrainerBuilderBase(abc.ABC):
    _train_dataset = None
    _eval_dataset = None
    _model_ref = None
-    _peft_config = None

    def __init__(self, cfg, model, tokenizer):
        self.cfg = cfg
@@ -683,26 +488,13 @@ class TrainerBuilderBase(abc.ABC):
    def eval_dataset(self, dataset):
        self._eval_dataset = dataset

-    @property
-    def peft_config(self):
-        return self._peft_config
-
-    @peft_config.setter
-    def peft_config(self, peft_config):
-        self._peft_config = peft_config
-
    @abstractmethod
    def build(self, total_num_steps):
        pass

-    def get_callbacks(self) -> List[TrainerCallback]:
-        callbacks = []
-        if self.cfg.use_wandb:
-            callbacks.append(
-                SaveAxolotlConfigtoWandBCallback(self.cfg.axolotl_config_path)
-            )
-
-        return callbacks
+    @abstractmethod
+    def get_callbacks(self):
+        pass

    @abstractmethod
    def get_post_trainer_create_callbacks(self, trainer):
@@ -710,6 +502,12 @@ class TrainerBuilderBase(abc.ABC):
        Callbacks added after the trainer is created, usually b/c these need access to the trainer
        """

+
+class HFCausalTrainerBuilder(TrainerBuilderBase):
+    """
+    Build the HuggingFace training args/trainer for Causal models
+    """
+
    def hook_pre_create_training_args(self, training_arguments_kwargs):
        # TODO
        return training_arguments_kwargs
@@ -726,16 +524,10 @@ class TrainerBuilderBase(abc.ABC):
        # TODO
        return trainer

-
-class HFCausalTrainerBuilder(TrainerBuilderBase):
-    """
-    Build the HuggingFace training args/trainer for Causal models
-    """
-
    def get_callbacks(self):
-        callbacks = super().get_callbacks()
+        callbacks = []
        callbacks.append(GPUStatsCallback(self.cfg))
-        callbacks.append(EvalFirstStepCallback())
+        callbacks.append(EvalFirstStepCallback)

        if self.cfg.relora_steps:
            callbacks.append(ReLoRACallback(self.cfg))
@@ -744,20 +536,12 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
            hasattr(self.model, "use_bettertransformer")
            and self.model.use_bettertransformer is True
        ):
-            callbacks.append(SaveBetterTransformerModelCallback())
+            callbacks.append(SaveBetterTransformerModelCallback)

        if self.cfg.use_wandb:
            callbacks.append(
                SaveAxolotlConfigtoWandBCallback(self.cfg.axolotl_config_path)
            )
-        if self.cfg.use_mlflow and is_mlflow_available():
-            from axolotl.utils.callbacks.mlflow_ import (
-                SaveAxolotlConfigtoMlflowCallback,
-            )
-
-            callbacks.append(
-                SaveAxolotlConfigtoMlflowCallback(self.cfg.axolotl_config_path)
-            )

        if self.cfg.loss_watchdog_threshold is not None:
            callbacks.append(LossWatchDogCallback(self.cfg))
@@ -774,11 +558,6 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):

        if self.cfg.do_bench_eval:
            callbacks.append(bench_eval_callback_factory(trainer, self.tokenizer))
-        if self.cfg.do_causal_lm_eval:
-            CausalLMBenchEvalCallback = causal_lm_bench_eval_callback_factory(
-                trainer, self.tokenizer
-            )
-            callbacks.append(CausalLMBenchEvalCallback(self.cfg))

        if self.cfg.early_stopping_patience:
            early_stop_cb = EarlyStoppingCallback(
@@ -833,7 +612,7 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
            training_arguments_kwargs[
                "gradient_checkpointing"
            ] = self.cfg.gradient_checkpointing
-            if self.cfg.gradient_checkpointing_kwargs is not None:
+            if self.cfg.gradient_checkpointing_kwargs:
                training_arguments_kwargs[
                    "gradient_checkpointing_kwargs"
                ] = self.cfg.gradient_checkpointing_kwargs
@@ -846,9 +625,6 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
            if self.cfg.fsdp_config:
                training_arguments_kwargs["fsdp_config"] = dict(self.cfg.fsdp_config)

-        if self.cfg.adapter == "qlora":
-            training_arguments_kwargs["qlora"] = True
-
        # deepspeed
        if self.cfg.deepspeed:
            training_arguments_kwargs["deepspeed"] = self.cfg.deepspeed
@@ -903,7 +679,7 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
        elif self.cfg.sample_packing and self.cfg.eval_sample_packing is False:
            training_arguments_kwargs["dataloader_drop_last"] = True

-        if not self.cfg.test_datasets and self.cfg.val_set_size == 0:
+        if self.cfg.val_set_size == 0:
            # no eval set, so don't eval
            training_arguments_kwargs["evaluation_strategy"] = "no"
        elif self.cfg.eval_steps:
@@ -930,8 +706,6 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
            training_arguments_kwargs["do_bench_eval"] = self.cfg.do_bench_eval
            if self.cfg.bench_dataset:
                training_arguments_kwargs["bench_dataset"] = self.cfg.bench_dataset
-        if self.cfg.do_causal_lm_eval:
-            training_arguments_kwargs["do_causal_lm_eval"] = self.cfg.do_causal_lm_eval
        if self.cfg.metric_for_best_model:
            training_arguments_kwargs[
                "metric_for_best_model"
@@ -971,10 +745,9 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
        training_arguments_kwargs[
            "per_device_train_batch_size"
        ] = self.cfg.micro_batch_size
-        if self.cfg.eval_batch_size:
-            training_arguments_kwargs[
-                "per_device_eval_batch_size"
-            ] = self.cfg.eval_batch_size
+        training_arguments_kwargs[
+            "per_device_eval_batch_size"
+        ] = self.cfg.eval_batch_size
        training_arguments_kwargs[
            "gradient_accumulation_steps"
        ] = self.cfg.gradient_accumulation_steps
@@ -992,10 +765,7 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
                self.cfg.load_best_model_at_end is not False
                or self.cfg.early_stopping_patience
            )
-            and (
-                (not self.cfg.test_datasets and self.cfg.val_set_size > 0)
-                or (self.cfg.test_datasets and self.cfg.val_set_size == 0)
-            )
+            and self.cfg.val_set_size > 0
            and self.cfg.save_steps
            and self.cfg.eval_steps
            and self.cfg.save_steps % self.cfg.eval_steps == 0
@@ -1016,10 +786,6 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
        training_arguments_kwargs["optim"] = (
            self.cfg.optimizer if self.cfg.optimizer else "adamw_hf"
        )
-        training_arguments_kwargs["loraplus_lr_ratio"] = self.cfg.loraplus_lr_ratio
-        training_arguments_kwargs[
-            "loraplus_lr_embedding"
-        ] = self.cfg.loraplus_lr_embedding
        training_arguments_kwargs["lr_scheduler_type"] = (
            self.cfg.lr_scheduler
            if self.cfg.lr_scheduler
@@ -1030,18 +796,12 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
            self.cfg.lr_scheduler_kwargs if self.cfg.lr_scheduler_kwargs else {}
        )
        training_arguments_kwargs["cosine_min_lr_ratio"] = self.cfg.cosine_min_lr_ratio
-        training_arguments_kwargs[
-            "cosine_constant_lr_ratio"
-        ] = self.cfg.cosine_constant_lr_ratio
        training_arguments_kwargs["weight_decay"] = (
            self.cfg.weight_decay if self.cfg.weight_decay is not None else 0.0
        )
        training_arguments_kwargs["sample_packing"] = (
            self.cfg.sample_packing if self.cfg.sample_packing else False
        )
-        training_arguments_kwargs["multipack_real_batches"] = (
-            self.cfg.flash_attention is not True
-        )
        training_arguments_kwargs["eval_sample_packing"] = (
            self.cfg.sample_packing
            if self.cfg.eval_sample_packing is not False
@@ -1050,20 +810,8 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
        training_arguments_kwargs[
            "sample_packing_seq_len_multiplier"
        ] = self.cfg.micro_batch_size
-        if self.cfg.relora_steps:
-            training_arguments_kwargs["relora_steps"] = self.cfg.relora_steps
-            training_arguments_kwargs[
-                "relora_warmup_steps"
-            ] = self.cfg.relora_warmup_steps
-            if self.cfg.relora_anneal_steps:
-                training_arguments_kwargs[
-                    "relora_anneal_steps"
-                ] = self.cfg.relora_anneal_steps
-            if self.cfg.relora_prune_ratio:
-                training_arguments_kwargs[
-                    "relora_prune_ratio"
-                ] = self.cfg.relora_prune_ratio
-
+        training_arguments_kwargs["relora_steps"] = self.cfg.relora_steps
+        training_arguments_kwargs["relora_warmup_steps"] = self.cfg.relora_warmup_steps
        training_arguments_kwargs = self.hook_pre_create_training_args(
            training_arguments_kwargs
        )
@@ -1075,42 +823,18 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
                "neftune_noise_alpha"
            ] = self.cfg.neftune_noise_alpha

-        trainer_kwargs = {}
-
-        if self.cfg.optimizer == "lion_pytorch":
-            from lion_pytorch import Lion
-
-            lion_kwargs = {"lr": training_arguments_kwargs["learning_rate"]}
-            if "weight_decay" in training_arguments_kwargs:
-                lion_kwargs["weight_decay"] = training_arguments_kwargs["weight_decay"]
-
-            if (
-                "adam_beta1" in training_arguments_kwargs
-                and "adam_beta2" in training_arguments_kwargs
-            ):
-                lion_kwargs["betas"] = (
-                    training_arguments_kwargs["adam_beta1"],
-                    training_arguments_kwargs["adam_beta2"],
-                )
-
-            trainer_kwargs["optimizers"] = (
-                Lion(params=self.model.parameters(), **lion_kwargs),
-                None,
-            )
-            # Set default so transformers doesn't throw
-            training_arguments_kwargs["optim"] = "adamw_hf"
-
-        if self.cfg.optimizer == "adamw_anyprecision":
-            if Path(self.cfg.torchdistx_path).exists():
-                sys.path.append(self.cfg.torchdistx_path)
-                importlib.import_module("torchdistx")
-
        training_args = (
            AxolotlTrainingArguments(  # pylint: disable=unexpected-keyword-arg
                **training_arguments_kwargs,
            )
        )
        training_args = self.hook_post_create_training_args(training_args)
+        trainer_kwargs = {}
+
+        if self.cfg.optimizer == "adamw_anyprecision":
+            if Path(self.cfg.torchdistx_path).exists():
+                sys.path.append(self.cfg.torchdistx_path)
+                importlib.import_module("torchdistx")

        data_collator_kwargs = {
            "padding": True,  # True/"longest" is the default
@@ -1172,27 +896,14 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
        if is_eval and training_args.eval_sample_packing:
            use_batch_sampler_collator = True

-        collator: Type[
-            Union[
-                V2BatchSamplerDataCollatorForSeq2Seq,
-                BatchSamplerDataCollatorForSeq2Seq,
-                DataCollatorForSeq2Seq,
-            ]
-        ]
        if use_batch_sampler_collator:
-            if self.cfg.model_config_type in SUPPORTED_MULTIPACK_MODEL_TYPES:
-                collator = V2BatchSamplerDataCollatorForSeq2Seq
-            elif (
-                self.cfg.model_config_type in ["llama"]
-                and self.cfg.flash_attention is not True
-            ):
-                collator = V2BatchSamplerDataCollatorForSeq2Seq
-            else:
-                collator = BatchSamplerDataCollatorForSeq2Seq
-        else:
-            collator = DataCollatorForSeq2Seq
+            return BatchSamplerDataCollatorForSeq2Seq(
+                self.tokenizer,
+                return_tensors="pt",
+                **kwargs,
+            )

-        return collator(
+        return DataCollatorForSeq2Seq(
            self.tokenizer,
            return_tensors="pt",
            **kwargs,
@@ -1205,7 +916,7 @@ class HFDPOTrainerBuilder(TrainerBuilderBase):
    """

    def get_callbacks(self):
-        callbacks = super().get_callbacks()
+        callbacks = []
        return callbacks

    def get_post_trainer_create_callbacks(self, trainer):
@@ -1223,82 +934,21 @@ class HFDPOTrainerBuilder(TrainerBuilderBase):
        ]:
            if hasattr(self.cfg, arg) and getattr(self.cfg, arg) is not None:
                training_args_kwargs[arg] = getattr(self.cfg, arg)
-
-        if self.cfg.hub_model_id:
-            training_args_kwargs["hub_model_id"] = self.cfg.hub_model_id
-            training_args_kwargs["push_to_hub"] = True
-            training_args_kwargs["hub_private_repo"] = True
-            training_args_kwargs["hub_always_push"] = True
-
-            if self.cfg.hub_strategy:
-                training_args_kwargs["hub_strategy"] = self.cfg.hub_strategy
-
-        if self.cfg.save_safetensors is not None:
-            training_args_kwargs["save_safetensors"] = self.cfg.save_safetensors
-
-        if self.eval_dataset:
-            training_args_kwargs["evaluation_strategy"] = "steps"
-            training_args_kwargs["eval_steps"] = self.cfg.eval_steps
-        else:
-            training_args_kwargs["evaluation_strategy"] = "no"
-        if self.cfg.bf16 or self.cfg.bfloat16:
-            training_args_kwargs["bf16"] = True
-
-        training_args_kwargs["lr_scheduler_type"] = (
-            self.cfg.lr_scheduler if self.cfg.lr_scheduler else "cosine"
-        )
-        training_args_kwargs["lr_scheduler_kwargs"] = (
-            self.cfg.lr_scheduler_kwargs if self.cfg.lr_scheduler_kwargs else {}
-        )
-        if self.cfg.remove_unused_columns is not None:
-            training_args_kwargs[
-                "remove_unused_columns"
-            ] = self.cfg.remove_unused_columns
-        else:
-            training_args_kwargs["remove_unused_columns"] = False
-
-        if self.cfg.dataloader_pin_memory is not None:
-            training_args_kwargs[
-                "dataloader_pin_memory"
-            ] = self.cfg.dataloader_pin_memory
-        if self.cfg.dataloader_num_workers is not None:
-            training_args_kwargs[
-                "dataloader_num_workers"
-            ] = self.cfg.dataloader_num_workers
-        if self.cfg.dataloader_prefetch_factor is not None:
-            training_args_kwargs[
-                "dataloader_prefetch_factor"
-            ] = self.cfg.dataloader_prefetch_factor
-        if self.cfg.gradient_checkpointing:
-            training_args_kwargs[
-                "gradient_checkpointing"
-            ] = self.cfg.gradient_checkpointing
-            if self.cfg.gradient_checkpointing_kwargs is not None:
-                training_args_kwargs[
-                    "gradient_checkpointing_kwargs"
-                ] = self.cfg.gradient_checkpointing_kwargs
-            else:
-                training_args_kwargs["gradient_checkpointing_kwargs"] = {
-                    "use_reentrant": False
-                }
-
-        # set save_strategy and save_steps
-        if self.cfg.save_steps:
-            training_args_kwargs["save_strategy"] = "steps"
-            training_args_kwargs["save_steps"] = self.cfg.save_steps
-        elif self.cfg.save_strategy:
-            training_args_kwargs["save_strategy"] = self.cfg.save_strategy
-        else:
-            # default to saving each epoch if not defined
-            training_args_kwargs["save_strategy"] = "epoch"
-
        training_args = TrainingArguments(
            per_device_train_batch_size=self.cfg.micro_batch_size,
-            max_steps=self.cfg.max_steps or total_num_steps,
+            max_steps=total_num_steps,
+            remove_unused_columns=False,
            gradient_accumulation_steps=self.cfg.gradient_accumulation_steps,
            learning_rate=self.cfg.learning_rate,
+            evaluation_strategy="no",
+            # eval_steps=self.cfg.eval_steps,
+            save_strategy="steps",
+            save_steps=self.cfg.save_steps,
            output_dir=self.cfg.output_dir,
            warmup_steps=self.cfg.warmup_steps,
+            bf16=True,
+            gradient_checkpointing=self.cfg.gradient_checkpointing,
+            gradient_checkpointing_kwargs={"use_reentrant": False},
            logging_first_step=True,
            logging_steps=1,
            optim=self.cfg.optimizer,
@@ -1317,31 +967,22 @@ class HFDPOTrainerBuilder(TrainerBuilderBase):
                dpo_trainer_kwargs["label_smoothing"] = self.cfg.dpo_label_smoothing
        elif self.cfg.rl == "kto_pair":
            dpo_trainer_kwargs["loss_type"] = "kto_pair"
-        if self.eval_dataset:
-            dpo_trainer_kwargs["eval_dataset"] = self.eval_dataset
-        if self.cfg.adapter and self.peft_config:
-            dpo_trainer_kwargs["peft_config"] = self.peft_config
-        if self.cfg.precompute_ref_log_probs is not None:
-            dpo_trainer_kwargs[
-                "precompute_ref_log_probs"
-            ] = self.cfg.precompute_ref_log_probs
-        dpo_trainer = AxolotlDPOTrainer(
+
+        dpo_trainer = DPOTrainer(
            self.model,
            self.model_ref,
            args=training_args,
            beta=self.cfg.dpo_beta or 0.1,
            train_dataset=self.train_dataset,
+            # eval_dataset=self.eval_dataset,
+            eval_dataset=None,
            tokenizer=self.tokenizer,
            max_length=self.cfg.sequence_len,
            max_target_length=None,
            max_prompt_length=self.cfg.sequence_len,
            generate_during_eval=True,
-            callbacks=self.get_callbacks(),
            **dpo_trainer_kwargs,
        )
-        dpo_trainer = self.hook_post_create_trainer(dpo_trainer)
-        for callback in self.get_post_trainer_create_callbacks(dpo_trainer):
-            dpo_trainer.add_callback(callback)

        return dpo_trainer

--- a/src/axolotl/datasets.py
+++ b/src/axolotl/datasets.py
@@ -24,30 +24,26 @@ class TokenizedPromptDataset(Dataset):
        Args:
            prompt_tokenizer (PromptTokenizingStrategy): The prompt tokenizing method for processing the data.
            dataset (dataset.Dataset): Dataset with text files.
-            process_count (int): Number of processes to use for tokenizing.
-            keep_in_memory (bool): Whether to keep the tokenized dataset in memory.
    """

    def __init__(  # pylint: disable=super-init-not-called
        self,
        prompt_tokenizer: PromptTokenizingStrategy,
-        dataset: Dataset,
+        dataset: IterableDataset,
        process_count: Optional[int] = None,
-        keep_in_memory: Optional[bool] = False,
        **kwargs,
    ):
        self.prompt_tokenizer = prompt_tokenizer
        self.process_count = process_count
-        self.keep_in_memory = keep_in_memory
-        super().__init__(
-            self.process(dataset).data,
-            **kwargs,
-        )
+        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 os.cpu_count())
-
+        num_proc = (
+            min(64, self.process_count)
+            if self.process_count
+            else min(64, os.cpu_count())
+        )
        map_kwargs = {}
        if self.prompt_tokenizer.supports_batched:
            map_kwargs["batched"] = True
@@ -56,8 +52,6 @@ class TokenizedPromptDataset(Dataset):
            self.prompt_tokenizer.tokenize_prompt,
            num_proc=num_proc,
            remove_columns=features,
-            keep_in_memory=self.keep_in_memory,
-            desc="Tokenizing Prompts",
            **map_kwargs,
        )

--- a/src/axolotl/logging_config.py
+++ b/src/axolotl/logging_config.py
@@ -30,7 +30,6 @@ class ColorfulFormatter(Formatter):

 DEFAULT_LOGGING_CONFIG: Dict[str, Any] = {
    "version": 1,
-    "disable_existing_loggers": False,
    "formatters": {
        "simple": {
            "format": "[%(asctime)s] [%(levelname)s] [%(name)s.%(funcName)s:%(lineno)d] [PID:%(process)d] %(message)s",
--- a/src/axolotl/loraplus.py
+++ b/src/axolotl/loraplus.py
@@ -1,133 +0,0 @@
-"""Module for LoRA+"""
-
-# MIT License
-#
-# Copyright (c) 2024 nikhil-ghosh-berkeley
-# https://github.com/nikhil-ghosh-berkeley/loraplus
-
-import logging
-from functools import reduce
-
-from peft.tuners import lora
-from torch import nn
-from transformers.pytorch_utils import ALL_LAYERNORM_LAYERS
-from transformers.trainer_pt_utils import get_parameter_names
-
-LOG = logging.getLogger("axolotl.loraplus")
-
-
-def get_module(name, opt_model):
-    """
-    Retrieve a module from a model using its parameter name.
-    Args:
-        name (str): Full name of the parameter, typically including module path.
-        opt_model (torch.nn.Module): The model from which to retrieve the module.
-
-    Returns:
-        Module corresponding to the given name.
-    """
-    parent_idx = 2 if "lora" in name else 1
-    module_names = name.split(sep=".")[:-parent_idx]
-    module = reduce(getattr, module_names, opt_model)
-    return module
-
-
-def create_loraplus_optimizer(
-    opt_model,
-    optimizer_cls,
-    optimizer_kwargs,
-    loraplus_lr_ratio,
-    loraplus_lr_embedding=None,
-):
-    """
-    Creates an optimizer for the given model, applying LoRA-specific learning rate adjustments to different parameter groups.
-
-    Args:
-        opt_model (torch.nn.Module): The model for which the optimizer is being created.
-        optimizer_cls (class): The class of the optimizer to be used (e.g., torch.optim.Adam).
-        optimizer_kwargs (dict): A dictionary of keyword arguments for the optimizer's initialization.
-        loraplus_lr_ratio (float): The learning rate ratio to be applied to LoRA parameters.
-        loraplus_lr_embedding (float, optional): A specific learning rate for embedding parameters, with a default value if not provided.
-
-    Returns:
-        An instance of the specified optimizer class configured with the model's parameters organized into groups with custom learning rates.
-    """
-
-    assert loraplus_lr_ratio is not None, "loraplus_lr_ratio must be provided."
-
-    if loraplus_lr_embedding is None:
-        loraplus_lr_embedding = 1e-6
-
-    decay_parameters = get_parameter_names(opt_model, ALL_LAYERNORM_LAYERS)
-    decay_parameters = [name for name in decay_parameters if "bias" not in name]
-    param_groups = {
-        "groupA": {},
-        "groupB": {},
-        "groupB_no_decay": {},
-        "embedding": {},
-    }
-
-    for name, param in opt_model.named_parameters():
-        if not param.requires_grad:
-            continue
-
-        module = get_module(name, opt_model)
-        if isinstance(module, lora.Embedding):
-            param_groups["embedding"][name] = param
-        elif "lora_B" in name or param.ndim == 1:
-            if name in decay_parameters:
-                param_groups["groupB"][name] = param
-            else:
-                param_groups["groupB_no_decay"][name] = param
-        else:
-            param_groups["groupA"][name] = param
-
-    assigned_param_groups = ""
-    for group, group_params in param_groups.items():
-        assigned_param_groups += f"{group}\n {list(group_params.keys())}\n\n"
-    LOG.info(assigned_param_groups)
-
-    lr = optimizer_kwargs["lr"]  # pylint: disable=invalid-name
-    weight_decay = optimizer_kwargs.get("weight_decay", 0.0)
-
-    optimizer_grouped_parameters = [
-        {
-            "params": list(param_groups["groupA"].values()),
-            "weight_decay": weight_decay,
-            "lr": lr,
-        },
-        {
-            "params": list(param_groups["embedding"].values()),
-            "weight_decay": weight_decay,
-            "lr": loraplus_lr_embedding,
-        },
-        {
-            "params": list(param_groups["groupB"].values()),
-            "weight_decay": weight_decay,
-            "lr": lr * loraplus_lr_ratio,
-        },
-        {
-            "params": list(param_groups["groupB_no_decay"].values()),
-            "weight_decay": 0.0,
-            "lr": lr * loraplus_lr_ratio,
-        },
-    ]
-
-    optimizer = optimizer_cls(optimizer_grouped_parameters, **optimizer_kwargs)
-    if optimizer_cls.__name__ == "Adam8bit":
-        import bitsandbytes
-
-        manager = bitsandbytes.optim.GlobalOptimManager.get_instance()
-
-        skipped = 0
-        for module in opt_model.modules():
-            if isinstance(module, nn.Embedding):
-                skipped += sum(
-                    {p.data_ptr(): p.numel() for p in module.parameters()}.values()
-                )
-                LOG.info(f"skipped {module}: {skipped/2**20}M params")
-                manager.register_module_override(module, "weight", {"optim_bits": 32})
-                LOG.debug(f"bitsandbytes: will optimize {module} in fp32")
-        LOG.info(f"skipped: {skipped/2**20}M params")
-
-    return optimizer
--- a/src/axolotl/models/phi/init.py
+++ b/src/axolotl/models/phi/init.py
@@ -0,0 +1,8 @@
+"""
+MixFormers model architecture used for phi models
+"""
+
+from .configuration_mixformer_sequential import MixFormerSequentialConfig  # noqa
+from .configuration_phi import PhiConfig  # noqa
+from .modeling_mixformer_sequential import MixFormerSequentialForCausalLM  # noqa
+from .modeling_phi import PhiForCausalLM  # noqa
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Wing Lian	1b33588f09	use low_cpu_mem_usage with ds zero 1 or 2	2024-01-16 19:33:44 -05:00
Wing Lian	1b59a3e698	use low_cpu_mem_usage when using deepspeed	2024-01-16 07:44:35 -05:00