tweak to make it work when we have no explicit test split

update pip install command for apex

.bandit

@@ -0,0 +1,3 @@
+[bandit]
+exclude = tests
+skips = B101

.flake8

@@ -0,0 +1,5 @@
+[flake8]
+max-line-length = 88
+
+select = C,E,F,W,B,B950
+extend-ignore = E203, E501, W503

.github/release-drafter.yml

@@ -0,0 +1,31 @@
+name-template: 'v$RESOLVED_VERSION'
+tag-template: 'v$RESOLVED_VERSION'
+categories:
+  - title: '🚀 Features'
+    labels:
+      - 'feature'
+      - 'enhancement'
+  - title: '🐛 Bug Fixes'
+    labels:
+      - 'fix'
+      - 'bugfix'
+      - 'bug'
+  - title: '🧰 Maintenance'
+    label: 'chore'
+change-template: '- $TITLE @$AUTHOR (#$NUMBER)'
+change-title-escapes: '\<*_&' # You can add # and @ to disable mentions, and add ` to disable code blocks.
+version-resolver:
+  major:
+    labels:
+      - 'major'
+  minor:
+    labels:
+      - 'minor'
+  patch:
+    labels:
+      - 'patch'
+  default: patch
+template: |
+  ## What’s Changed
+
+  $CHANGES

.github/workflows/base.yml

@@ -0,0 +1,68 @@
+name: ci-cd-base
+
+on:
+  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: self-hosted
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - cuda: "118"
+            cuda_version: 11.8.0
+            python_version: "3.9"
+            pytorch: 2.0.0
+            axolotl_extras:
+          - cuda: "118"
+            cuda_version: 11.8.0
+            python_version: "3.10"
+            pytorch: 2.0.0
+            axolotl_extras:
+          - cuda: "117"
+            cuda_version: 11.7.1
+            python_version: "3.9"
+            pytorch: 1.13.1
+            axolotl_extras:
+          - cuda: "118"
+            cuda_version: 11.8.0
+            python_version: "3.9"
+            pytorch: 2.0.0
+            axolotl_extras: gptq
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+      - name: Docker metadata
+        id: metadata
+        uses: docker/metadata-action@v3
+        with:
+          images: winglian/axolotl-base
+      - name: Login to Docker Hub
+        uses: docker/login-action@v2
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v2
+      - name: Build
+        uses: docker/build-push-action@v4
+        with:
+          context: .
+          file: ./docker/Dockerfile-base
+          push: ${{ github.event_name != 'pull_request' }}
+          tags: ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+          labels: ${{ steps.metadata.outputs.labels }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max
+          build-args: |
+            CUDA_VERSION=${{ matrix.cuda_version }}
+            CUDA=${{ matrix.cuda }}
+            PYTHON_VERSION=${{ matrix.python_version }}
+            PYTORCH_VERSION=${{ matrix.pytorch }}
+            AXOLOTL_EXTRAS=${{ matrix.axolotl_extras }}

.github/workflows/main.yml

@@ -0,0 +1,119 @@
+name: ci-cd
+
+on:
+  push:
+    branches:
+      - "main"
+      - "dev"
+
+jobs:
+  build-axolotl:
+    if: github.repository_owner == 'OpenAccess-AI-Collective'
+    # this job needs to be run on self-hosted GPU runners...
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - cuda: cu118
+            cuda_version: 11.8.0
+            python_version: "3.9"
+            pytorch: 2.0.0
+            axolotl_extras:
+          - cuda: cu118
+            cuda_version: 11.8.0
+            python_version: "3.10"
+            pytorch: 2.0.0
+            axolotl_extras:
+          - cuda: cu118
+            cuda_version: 11.8.0
+            python_version: "3.9"
+            pytorch: 2.0.0
+            axolotl_extras: gptq
+          - cuda: cu117
+            cuda_version: 11.7.1
+            python_version: "3.9"
+            pytorch: 1.13.1
+            axolotl_extras:
+    runs-on: self-hosted
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+      - name: Docker metadata
+        id: metadata
+        uses: docker/metadata-action@v3
+        with:
+          images: winglian/axolotl
+      - name: Login to Docker Hub
+        uses: docker/login-action@v2
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v2
+      - name: Build
+        uses: docker/build-push-action@v4
+        with:
+          context: .
+          build-args: |
+            BASE_TAG=${{ github.ref_name }}-base-py${{ matrix.python_version }}-${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+          file: ./docker/Dockerfile
+          push: ${{ github.event_name != 'pull_request' }}
+          tags: ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+          labels: ${{ steps.metadata.outputs.labels }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max
+  build-axolotl-runpod:
+    needs: build-axolotl
+    if: github.repository_owner == 'OpenAccess-AI-Collective'
+    # this job needs to be run on self-hosted GPU runners...
+    strategy:
+      matrix:
+        include:
+          - cuda: cu118
+            cuda_version: 11.8.0
+            python_version: "3.9"
+            pytorch: 2.0.0
+            axolotl_extras:
+          - cuda: cu118
+            cuda_version: 11.8.0
+            python_version: "3.10"
+            pytorch: 2.0.0
+            axolotl_extras:
+          - cuda: cu118
+            cuda_version: 11.8.0
+            python_version: "3.9"
+            pytorch: 2.0.0
+            axolotl_extras: gptq
+          - cuda: cu117
+            cuda_version: 11.7.1
+            python_version: "3.9"
+            pytorch: 1.13.1
+            axolotl_extras:
+    runs-on: self-hosted
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+      - name: Docker metadata
+        id: metadata
+        uses: docker/metadata-action@v3
+        with:
+          images: winglian/axolotl-runpod
+      - name: Login to Docker Hub
+        uses: docker/login-action@v2
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v2
+      - name: Build
+        uses: docker/build-push-action@v4
+        with:
+          context: .
+          build-args: |
+            BASE_TAG=${{ github.ref_name }}-py${{ matrix.python_version }}-${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+          file: ./docker/Dockerfile-runpod
+          push: ${{ github.event_name != 'pull_request' }}
+          tags: ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+          labels: ${{ steps.metadata.outputs.labels }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max

.github/workflows/pre-commit.yml

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

.github/workflows/tests.yml

@@ -0,0 +1,32 @@
+name: PyTest
+on:
+  push:
+  pull_request:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python_version: ["3.9", "3.10"]
+    timeout-minutes: 10
+
+    steps:
+      - name: Check out repository code
+        uses: actions/checkout@v3
+
+      - name: Setup Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: ${{ matrix.python_version }}
+          cache: 'pip' # caching pip dependencies
+
+      - name: Install dependencies
+        run: |
+          pip install -e .
+          pip install -r requirements-tests.txt
+
+      - name: Run tests
+        run: |
+          pytest tests/

.gitignore

@@ -1,3 +1,163 @@
 **/axolotl.egg-info
-**/__pycache__
-.idea
+configs
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+.idea/

.isort.cfg

@@ -0,0 +1,2 @@
+[settings]
+profile=black

.mypy.ini

@@ -0,0 +1,39 @@
+[mypy]
+
+exclude = venv
+
+[mypy-alpaca_lora_4bit.*]
+ignore_missing_imports = True
+
+[mypy-axolotl.monkeypatch.*]
+ignore_errors = True
+
+[mypy-flash_attn.*]
+ignore_missing_imports = True
+
+[mypy-huggingface_hub]
+ignore_missing_imports = True
+
+[mypy-transformers.*]
+ignore_missing_imports = True
+
+[mypy-peft]
+ignore_missing_imports = True
+
+[mypy-bitsandbytes]
+ignore_missing_imports = True
+
+[mypy-datasets]
+ignore_missing_imports = True
+
+[mypy-fire]
+ignore_missing_imports = True
+
+[mypy-setuptools]
+ignore_missing_imports = True
+
+[mypy-addict]
+ignore_missing_imports = True
+
+[mypy-xformers.*]
+ignore_missing_imports = True

.pre-commit-config.yaml

@@ -0,0 +1,42 @@
+default_language_version:
+    python: python3
+
+repos:
+-   repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v4.4.0
+    hooks:
+    -   id: check-yaml
+    -   id: end-of-file-fixer
+    -   id: trailing-whitespace
+-   repo: https://github.com/psf/black
+    rev: 23.3.0
+    hooks:
+    -   id: black
+-   repo: https://github.com/pycqa/isort
+    rev: 5.12.0
+    hooks:
+      - id: isort
+-   repo: https://github.com/PyCQA/flake8
+    rev: 6.0.0
+    hooks:
+    - id: flake8
+-   repo: https://github.com/PyCQA/pylint
+    rev: v2.17.4
+    hooks:
+    - id: pylint
+-   repo: https://github.com/pre-commit/mirrors-mypy
+    rev: v1.3.0
+    hooks:
+    - id: mypy
+      additional_dependencies:
+        [
+            'types-PyYAML',
+        ]
+-   repo: https://github.com/PyCQA/bandit
+    rev: 1.7.5
+    hooks:
+    -   id: bandit
+        args: [
+            '--ini',
+            '.bandit',
+        ]

.pylintrc

@@ -0,0 +1,14 @@
+[MASTER]
+init-hook="from pylint.config import find_pylintrc; import os, sys; sys.path.append(os.path.dirname(find_pylintrc()))"
+
+[TYPECHECK]
+
+# List of members which are set dynamically and missed by Pylint inference
+# system, and so shouldn't trigger E1101 when accessed.
+generated-members=numpy.*, torch.*
+
+
+[pylint.messages_control]
+disable=missing-function-docstring, line-too-long, import-error,
+    too-many-arguments, too-many-locals, too-many-statements, too-many-branches, too-few-public-methods,
+    too-many-instance-attributes, fixme, import-outside-toplevel, logging-fstring-interpolation,

FAQS.md

@@ -0,0 +1,7 @@
+# FAQs
+
+- Can you train StableLM with this? Yes, but only with a single GPU atm. Multi GPU support is coming soon! Just waiting on this [PR](https://github.com/huggingface/transformers/pull/22874)
+- Will this work with Deepspeed? That's still a WIP, but setting `export ACCELERATE_USE_DEEPSPEED=true` should work in some cases
+- `Error invalid argument at line 359 in file /workspace/bitsandbytes/csrc/pythonInterface.c`
+`/arrow/cpp/src/arrow/filesystem/s3fs.cc:2598:  arrow::fs::FinalizeS3 was not called even though S3 was initialized.`
+This could lead to a segmentation fault at exit. Try reinstalling bitsandbytes and transformers from source.

README.md

@@ -1,47 +1,602 @@
-~~# Axolotl
+# Axolotl
 
-#### You know you're going to axolotl questions
+<div align="center">
+  <img src="image/axolotl.png" alt="axolotl" width="160">
+  <div>
+    <p>
+      <b>One repo to finetune them all! </b>
+    </p>
+    <p>
+      Go ahead and axolotl questions!!
+    </p>
+    <img src="https://github.com/OpenAccess-AI-Collective/axolotl/actions/workflows/pre-commit.yml/badge.svg?branch=main" alt="pre-commit">
+    <img alt="PyTest Status" src="https://github.com/OpenAccess-AI-Collective/axolotl/actions/workflows/tests.yml/badge.svg?branch=main">
+  </div>
+</div>
 
-## Getting Started
+## Axolotl supports
 
-- Point the config you are using to a huggingface hub dataset (see [configs/llama_7B_4bit.yml](https://github.com/winglian/axolotl/blob/main/configs/llama_7B_4bit.yml#L6-L8))
+|          | fp16/fp32 | lora | qlora | gptq | gptq w/ lora | gptq w/flash attn | flash attn | xformers attn |
+|----------|:----------|:-----|-------|------|:-------------|-------------------|------------|---------------|
+| llama    | ✅         | ✅    | ✅     | ✅    | ✅             | ✅                 | ✅          | ✅             |
+| Pythia   | ✅         | ✅    | ✅     | ❌    | ❓            | ❌                 | ❌          | ❓             |
+| cerebras | ✅         | ✅    | ✅     | ❌    | ❓            | ❌                 | ❌          | ✅             |
+| mpt      | ✅         | ❌    | ❓     | ❌    | ❓            | ❌                 | ❌          | ❓             |
+| falcon   | ✅         | ✅    | ✅     | ❌    | ❓            | ❌                 | ❌          | ✅             |
+| gpt-j    | ✅         | ✅    | ✅     | ❌    | ❓            | ❌                 | ❓          | ✅             |
+
+
+## Quickstart ⚡
+
+**Requirements**: Python 3.9 and Pytorch 2.0.
+
+```bash
+git clone https://github.com/OpenAccess-AI-Collective/axolotl
+
+pip3 install -e .
+pip3 install -U git+https://github.com/huggingface/peft.git
+
+accelerate config
+
+# finetune lora
+accelerate launch scripts/finetune.py examples/openllama-3b/lora.yml
+
+# inference
+accelerate launch scripts/finetune.py examples/openllama-3b/lora.yml \
+    --inference --lora_model_dir="./lora-out"
+```
+
+## Installation
+
+### Environment
+
+- Docker
+  ```bash
+  docker run --gpus '"all"' --rm -it winglian/axolotl:main-py3.9-cu118-2.0.0
+  ```
+  - `winglian/axolotl-runpod:main-py3.9-cu118-2.0.0`: for runpod
+  - `winglian/axolotl-runpod:main-py3.9-cu118-2.0.0-gptq`: for gptq
+  - `winglian/axolotl:dev`: dev branch (not usually up to date)
+
+  Or run on the current files for development:
+
+  ```sh
+  docker compose up -d
+  ```
+
+- Conda/Pip venv
+  1. Install python **3.9**
+
+  2. Install pytorch stable https://pytorch.org/get-started/locally/
+
+  3. Install python dependencies with ONE of the following:
+      - Recommended, supports QLoRA, NO gptq/int4 support
+        ```bash
+        pip3 install -e .
+        pip3 install -U git+https://github.com/huggingface/peft.git
+        ```
+      - gptq/int4 support, NO QLoRA
+        ```bash
+        pip3 install -e .[gptq]
+        ```
+      - same as above but not recommended
+        ```bash
+        pip3 install -e .[gptq_triton]
+        ```
+
+- LambdaLabs
+  <details>
+
+  <summary>Click to Expand</summary>
+
+  1. Install python
+  ```bash
+  sudo apt update
+  sudo apt install -y python3.9
+
+  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
+
+  ```
+
+  2. Install pip
+  ```bash
+  wget https://bootstrap.pypa.io/get-pip.py
+  python get-pip.py
+  ```
+
+  3. Install torch
+  ```bash
+  pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
+  ```
+
+  4. Axolotl
+  ```bash
+  git clone https://github.com/OpenAccess-AI-Collective/axolotl
+  cd axolotl
+
+  pip3 install -e . # change depend on needs
+  pip3 install protobuf==3.20.3
+  pip3 install -U requests
+  pip3 install -U --ignore-installed psutil
+  pip3 install -U scipy
+  pip3 install git+https://github.com/huggingface/peft.git # not for gptq
+  ```
+
+  5. Set path
+  ```bash
+  export LD_LIBRARY_PATH=/usr/lib/x86_64-linux-gnu:$LD_LIBRARY_PATH
+  ```
+  </details>
+
+### Dataset
+
+Have dataset(s) in one of the following format (JSONL recommended):
+
+- `alpaca`: instruction; input(optional)
+  ```json
+  {"instruction": "...", "input": "...", "output": "..."}
+  ```
+- `sharegpt:chat`: conversations
+  ```json
+  {"conversations": [{"from": "...", "value": "..."}]}
+  ```
+- `completion`: raw corpus
+  ```json
+  {"text": "..."}
+  ```
+
+<details>
+
+<summary>See other formats</summary>
+
+- `jeopardy`: question and answer
+  ```json
+  {"question": "...", "category": "...", "answer": "..."}
+  ```
+- `oasst`: instruction
+  ```json
+  {"INSTRUCTION": "...", "RESPONSE": "..."}
+  ```
+- `gpteacher`: instruction; input(optional)
+  ```json
+  {"instruction": "...", "input": "...", "response": "..."}
+  ```
+- `reflection`: instruction with reflect; input(optional)
+  ```json
+  {"instruction": "...", "input": "...", "output": "...", "reflection": "...", "corrected": "..."}
+  ```
+- `explainchoice`: question, choices, (solution OR explanation)
+  ```json
+  {"question": "...", "choices": ["..."], "solution": "...", "explanation": "..."}
+  ```
+- `concisechoice`: question, choices, (solution OR explanation)
+  ```json
+  {"question": "...", "choices": ["..."], "solution": "...", "explanation": "..."}
+  ```
+- `summarizetldr`: article and summary
+  ```json
+  {"article": "...", "summary": "..."}
+  ```
+- `alpaca_chat`: basic instruct for alpaca chat
+  ```json
+  {"instruction": "...", "input": "...", "response": "..."}
+  ```
+- `alpaca_chat.load_qa`: question and answer for alpaca chat
+  ```json
+  {"question": "...", "answer": "..."}
+  ```
+- `alpaca_chat.load_concise`: question and answer for alpaca chat, for concise answers
+  ```json
+  {"instruction": "...", "input": "...", "response": "..."}
+  ```
+- `alpaca_chat.load_camel_ai`: question and answer for alpaca chat, for load_camel_ai
+  ```json
+  {"message_1": "...", "message_2": "..."}
+  ```
+- `alpaca_w_system.load_open_orca`: support for open orca datasets with included system prompts, instruct
+  ```json
+  {"system_prompt": "...", "question": "...", "response": "..."}
+  ```
+- `context_qa`: in context question answering from an article
+  ```json
+  {"article": "...", "question": "...", "answer": "..."}
+  ```
+- `context_qa.load_404`: in context question answering from an article, with default response for no answer from context
+  ```json
+  {"article": "...", "unanswerable_question": "..."}
+  ```
+- `creative_acr.load_answer`: instruction and revision
+  ```json
+  {"instruction": "...", "revision": "..."}
+  ```
+- `creative_acr.load_critique`: critique
+  ```json
+  {"scores": "...", "critiques": "...", "instruction": "...", "answer": "..."}
+  ```
+- `creative_acr.load_revise`: critique and revise
+  ```json
+  {"scores": "...", "critiques": "...", "instruction": "...", "answer": "...", "revision": "..."}
+  ```
+- `pygmalion`: pygmalion
+  ```json
+  {"conversations": [{"role": "...", "value": "..."}]}
+  ```
+- `sharegpt_simple.load_role`: conversations where `role` is used instead of `from`
+  ```json
+  {"conversations": [{"role": "...", "value": "..."}]}
+  ```
+- `sharegpt_jokes`: creates a chat where bot is asked to tell a joke, then explain why the joke is funny
+  ```json
+  {"conversations": [{"title": "...", "text": "...", "explanation": "..."}]}
+  ```
+
+</details>
+
+#### How to add custom prompts
+
+  1. Add your method to a file in [prompt_strategies](src/axolotl/prompt_strategies). Please see other files as example.
+  2. Use your custom file name as the dataset type.
+
+Optionally, download some datasets, see [data/README.md](data/README.md)
+
+
+
+### Config
+
+See sample configs in [configs](configs) folder or [examples](examples) for quick start. It is recommended to duplicate and modify to your needs. The most important options are:
+
+- model
+  ```yaml
+  base_model: ./llama-7b-hf # local or huggingface repo
+  ```
+  Note: The code will load the right architecture.
+
+- dataset
+  ```yaml
+  datasets:
+    - path: vicgalle/alpaca-gpt4 # local or huggingface repo
+      type: alpaca # format from earlier
+  sequence_len: 2048 # max token length / prompt
+  ```
+
+- loading
+  ```yaml
+  load_in_4bit: true
+  load_in_8bit: true
+  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
+  ```
+  Note: Repo does not do 4-bit quantization.
+
+- lora
+  ```yaml
+  adapter: lora # qlora or leave blank for full finetune
+  lora_r: 8
+  lora_alpha: 16
+  lora_dropout: 0.05
+  lora_target_modules:
+    - q_proj
+    - v_proj
+  ```
+
+<details>
+
+<summary>All yaml options</summary>
 
 ```yaml
+# this is the huggingface model that contains *.pt, *.safetensors, or *.bin files
+# this can also be a relative path to a model on disk
+base_model: ./llama-7b-hf
+# you can specify an ignore pattern if the model repo contains more than 1 model type (*.pt, etc)
+base_model_ignore_patterns:
+# if the base_model repo on hf hub doesn't include configuration .json files,
+# you can set that here, or leave this empty to default to base_model
+base_model_config: ./llama-7b-hf
+# 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
+model_type: AutoModelForCausalLM
+# Corresponding tokenizer for the model AutoTokenizer is a good choice
+tokenizer_type: AutoTokenizer
+# Trust remote code for untrusted source
+trust_remote_code:
+# use_fast option for tokenizer loading from_pretrained, default to True
+tokenizer_use_fast:
+
+# 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
+# use bitsandbytes 4 bit
+load_in_4bit:
+
+# Use CUDA bf16
+bf16: true # bool or 'full' for `bf16_full_eval`. require >=ampere
+# Use CUDA fp16
+fp16: true
+# Use CUDA tf32
+tf32: true # require >=ampere
+
+# a list of one or more datasets to finetune the model with
 datasets:
+  # this can be either a hf dataset, or relative path
   - path: vicgalle/alpaca-gpt4
-    type: alpaca
+  # The type of prompt to use for training. [alpaca, sharegpt, gpteacher, oasst, reflection]
+    type: alpaca # format OR format:prompt_style (chat/instruct)
+    data_files: # path to source data files
+    shards: # number of shards to split data into
+
+# 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
+# push prepared dataset to hub
+push_dataset_to_hub: # repo path
+# push checkpoints to hub
+push_to_hub_model_id: # repo path
+# whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets
+# required to be true when used in combination with `push_dataset_to_hub`
+hf_use_auth_token: # boolean
+# How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc
+val_set_size: 0.04
+# Num shards for whole dataset
+dataset_shard_num:
+# Index of shard to use for whole dataset
+dataset_shard_idx:
+
+# the maximum length of an input to train with, this should typically be less than 2048
+# as most models have a token/context limit of 2048
+sequence_len: 2048
+# max sequence length to concatenate training samples together up to
+# inspired by StackLLaMA. see https://huggingface.co/blog/stackllama#supervised-fine-tuning
+max_packed_sequence_len: 1024
+
+# if you want to use 'lora' or 'qlora' or leave blank to train all parameters in original model
+adapter: lora
+# if you already have a lora model trained that you want to load, put that here
+# lora hyperparameters
+lora_model_dir:
+lora_r: 8
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_modules:
+  - q_proj
+  - v_proj
+#  - k_proj
+#  - o_proj
+#  - gate_proj
+#  - down_proj
+#  - up_proj
+lora_target_linear: # if true, will target all linear layers
+lora_modules_to_save:
+#  - embed_tokens
+#  - lm_head
+lora_out_dir:
+lora_fan_in_fan_out: false
+
+# wandb configuration if you're using it
+wandb_mode:
+wandb_project:
+wandb_watch:
+wandb_run_id:
+wandb_log_model: # 'checkpoint'
+
+# where to save the finished model to
+output_dir: ./completed-model
+
+# training hyperparameters
+gradient_accumulation_steps: 1
+micro_batch_size: 2
+eval_batch_size: 2
+num_epochs: 3
+warmup_steps: 100
+learning_rate: 0.00003
+logging_steps:
+save_steps:
+eval_steps:
+
+# whether to mask out or include the human's prompt from the training labels
+train_on_inputs: false
+# don't use this, leads to wonky training (according to someone on the internet)
+group_by_length: false
+
+# Whether to use gradient checkpointing https://huggingface.co/docs/transformers/v4.18.0/en/performance#gradient-checkpointing
+gradient_checkpointing: false
+
+# stop training after this many evaluation losses have increased in a row
+# https://huggingface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppingCallback
+early_stopping_patience: 3
+
+# specify a scheduler and kwargs to use with the optimizer
+lr_scheduler: # 'one_cycle' | 'log_sweep' | empty for cosine
+lr_scheduler_kwargs:
+
+# 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
+optimizer:
+# specify weight decay
+weight_decay:
+# adamw hyperparams
+adam_beta1:
+adam_beta2:
+adam_epsilon:
+# Gradient clipping max norm
+max_grad_norm:
+
+# whether to bettertransformers
+flash_optimum:
+# whether to use xformers attention patch https://github.com/facebookresearch/xformers:
+xformers_attention:
+# whether to use flash attention patch https://github.com/HazyResearch/flash-attention:
+flash_attention:  # require a100 for llama
+# whether to use scaled-dot-product attention
+# https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html
+sdp_attention:
+# Landmark attention (only llama)
+landmark_attention:
+# xpos RoPE see https://github.com/kaiokendev/cutoff-len-is-context-len/blob/main/util/xpos_rope_llama_monkey_patch.py
+# llama only
+xpos_rope:
+
+# 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
+# be careful with this being turned on between different models
+auto_resume_from_checkpoints: false
+
+# don't mess with this, it's here for accelerate and torchrun
+local_rank:
+
+# add or change special tokens
+special_tokens:
+  # bos_token: "<s>"
+  # eos_token: "</s>"
+  # unk_token: "<unk>"
+# add extra tokens
+tokens:
+
+# FSDP
+fsdp:
+fsdp_config:
+
+# Deepspeed
+deepspeed:
+
+# Path to torch distx for optim 'adamw_anyprecision'
+torchdistx_path:
+
+# Set padding for data collator to 'longest'
+collator_pad_to_longest:
+
+# Debug mode
+debug:
+
+# Seed
+seed:
+
+# Allow overwrite yml config using from cli
+strict:
 ```
 
-- Optionally Download some datasets, see [data/README.md](data/README.md)
+</details>
 
+### Accelerate
 
-- Create a new or update the existing YAML config [config/pythia_1_2B_alpaca.yml](config/pythia_1_2B_alpaca.yml)
-- Install python dependencies with ONE of the following:
+Configure accelerate
 
-    - `pip3 install -e .[int4]` (recommended)
-    - `pip3 install -e .[int4_triton]`
-    - `pip3 install -e .`
--
-- If not using `int4` or `int4_triton`, run `pip install "peft @ git+https://github.com/huggingface/peft.git"`
-- Configure accelerate `accelerate config` or update `~/.cache/huggingface/accelerate/default_config.yaml`
+```bash
+accelerate config
 
-```yaml
-compute_environment: LOCAL_MACHINE
-distributed_type: MULTI_GPU
-downcast_bf16: 'no'
-gpu_ids: all
-machine_rank: 0
-main_training_function: main
-mixed_precision: bf16
-num_machines: 1
-num_processes: 4
-rdzv_backend: static
-same_network: true
-tpu_env: []
-tpu_use_cluster: false
-tpu_use_sudo: false
-use_cpu: false
+# Edit manually
+# nano ~/.cache/huggingface/accelerate/default_config.yaml
 ```
 
-- Train! `accelerate launch scripts/finetune.py`, make sure to choose the correct YAML config file
-- Alternatively you can pass in the config file like: `accelerate launch scripts/finetune.py configs/llama_7B_alpaca.yml`~~
+### Train
+
+Run
+```bash
+accelerate launch scripts/finetune.py configs/your_config.yml
+```
+
+### Inference
+
+Pass the appropriate flag to the train command:
+
+- Pretrained LORA:
+  ```bash
+  --inference --lora_model_dir="./lora-output-dir"
+  ```
+- Full weights finetune:
+  ```bash
+  --inference --base_model="./completed-model"
+  ```
+- Full weights finetune w/ a prompt from a text file:
+  ```bash
+  cat /tmp/prompt.txt | python scripts/finetune.py configs/your_config.yml \
+    --base_model="./completed-model" --inference --prompter=None --load_in_8bit=True
+  ```
+
+### Merge LORA to base
+
+Add below flag to train command above
+
+```bash
+--merge_lora --lora_model_dir="./completed-model" --load_in_8bit=False --load_in_4bit=False
+```
+
+If you run out of CUDA memory, you can try to merge in system RAM with
+
+```bash
+CUDA_VISIBLE_DEVICES="" python3 scripts/finetune.py ...
+```
+
+## Common Errors 🧰
+
+> Cuda out of memory
+
+Please reduce any below
+  - `micro_batch_size`
+  - `eval_batch_size`
+  - `gradient_accumulation_steps`
+  - `sequence_len`
+
+> RuntimeError: expected scalar type Float but found Half
+
+Try set `fp16: true`
+
+> NotImplementedError: No operator found for `memory_efficient_attention_forward` ...
+
+Try to turn off xformers.
+
+## Need help? 🙋♂️
+
+Join our [Discord server](https://discord.gg/HhrNrHJPRb) where we can help you
+
+## Badge ❤🏷️
+
+Building something cool with Axolotl? Consider adding a badge to your model card.
+
+```markdown
+[<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
+```
+
+[<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
+
+## Community Showcase
+
+Open Access AI Collective
+- [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)
+
+PocketDoc Labs
+- [Dan's PersonalityEngine 13b LoRA](https://huggingface.co/PocketDoc/Dans-PersonalityEngine-13b-LoRA)
+
+## Contributing 🤝
+
+Bugs? Please check for open issue else create a new [Issue](https://github.com/OpenAccess-AI-Collective/axolotl/issues/new).
+
+PRs are **greatly welcome**!
+
+Please run below to setup env
+```bash
+pip3 install -r requirements-dev.txt -r requirements-tests.txt
+pre-commit install
+
+# test
+pytest tests/
+```

TODO.md

@@ -0,0 +1,10 @@
+# todo list
+
+- [] Validation of parameters for combinations that won't work
+
+
+
+## things that are known not to work
+
+- FSDP offload and gradient_checkpointing - https://github.com/pytorch/pytorch/issues/82203
+- adamw_bnb_8bit doesn't play well with FSDP offload

configs/cerebras_1_3B_alpaca.yml

@@ -1,40 +0,0 @@
-base_model: cerebras/Cerebras-GPT-1.3B
-model_type: AutoModelForCausalLM
-tokenizer_type: AutoTokenizer
-load_in_8bit: true
-datasets:
-  - path: data/alpaca_data_gpt4.jsonl
-    type: alpaca
-  - path: data/vicuna_cleaned.jsonl
-    type: sharegpt
-  - path: data/gpt4-instruct-similarity-0.6-dataset.jsonl
-    type: gpteacher
-  - path: data/roleplay-similarity_0.6-instruct-dataset.jsonl
-    type: gpteacher
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.05
-adapter: lora
-sequence_len: 2048
-lora_r: 8
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules:
-  - c_attn
-lora_fan_in_fan_out: false
-wandb_project: pythia-1.4b-lora
-wandb_watch:
-wandb_run_id:
-wandb_log_model: checkpoint
-output_dir: ./lora-alpaca
-batch_size: 32
-micro_batch_size: 4
-num_epochs: 5
-learning_rate: 0.0003
-train_on_inputs: false
-group_by_length: false
-bf16: True
-tf32: True
-gradient_checkpointing:
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:

configs/llama_65B_alpaca.yml

@@ -1,41 +0,0 @@
-base_model: huggyllama/llama-65b
-model_type: LlamaForCausalLM
-tokenizer_type: LlamaTokenizer
-load_in_8bit: true
-datasets:
-  - path: data/alpaca_data_gpt4.jsonl
-    type: alpaca
-  - path: data/vicuna_cleaned.jsonl
-    type: sharegpt
-  - path: data/gpt4-instruct-similarity-0.6-dataset.jsonl
-    type: gpteacher
-  - path: data/roleplay-similarity_0.6-instruct-dataset.jsonl
-    type: gpteacher
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.04
-adapter: lora
-lora_model_dir:
-sequence_len: 2048
-lora_r: 8
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules:
-  - q_proj
-  - v_proj
-lora_fan_in_fan_out: false
-wandb_project: llama-65b-lora
-wandb_watch:
-wandb_run_id:
-wandb_log_model: checkpoint
-output_dir: ./lora-llama-alpaca
-batch_size: 128
-micro_batch_size: 16
-num_epochs: 5
-learning_rate: 0.00003
-train_on_inputs: false
-group_by_length: false
-bf16: true
-tf32: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:

configs/llama_7B_4bit.yml

@@ -1,41 +0,0 @@
-base_model: decapoda-research/llama-7b-hf-int4
-base_model_config: decapoda-research/llama-7b-hf
-model_type: LlamaForCausalLM
-tokenizer_type: LlamaTokenizer
-load_in_8bit: true
-datasets:
-  - path: vicgalle/alpaca-gpt4
-    type: alpaca
-dataset_prepared_path: data/last_run_prepared
-val_set_size: 0.04
-adapter: lora
-lora_model_dir:
-sequence_len: 2048
-max_packed_sequence_len: 1024
-lora_r: 8
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules:
-  - q_proj
-  - v_proj
-#  - k_proj
-#  - o_proj
-lora_fan_in_fan_out: false
-wandb_project:
-wandb_watch:
-wandb_run_id:
-wandb_log_model: checkpoint
-output_dir: ./lora-test
-batch_size: 8
-micro_batch_size: 2
-num_epochs: 3
-learning_rate: 0.00003
-train_on_inputs: false
-group_by_length: false
-bf16: true
-tf32: true
-gradient_checkpointing: false
-early_stopping_patience: 3
-resume_from_checkpoint:
-local_rank:
-load_4bit: true

configs/llama_7B_alpaca.yml

@@ -1,41 +0,0 @@
-base_model: huggyllama/llama-7b
-model_type: LlamaForCausalLM
-tokenizer_type: LlamaTokenizer
-load_in_8bit: true
-datasets:
-  - path: data/alpaca_data_gpt4.jsonl
-    type: alpaca
-  - path: data/vicuna_cleaned.jsonl
-    type: sharegpt
-  - path: data/gpt4-instruct-similarity-0.6-dataset.jsonl
-    type: gpteacher
-  - path: data/roleplay-similarity_0.6-instruct-dataset.jsonl
-    type: gpteacher
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.04
-adapter: lora
-lora_model_dir:
-sequence_len: 2048
-lora_r: 8
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules:
-  - q_proj
-  - v_proj
-lora_fan_in_fan_out: false
-wandb_project: llama-7b-lora
-wandb_watch:
-wandb_run_id:
-wandb_log_model: checkpoint
-output_dir: ./lora-llama-alpaca
-batch_size: 128
-micro_batch_size: 16
-num_epochs: 5
-learning_rate: 0.00003
-train_on_inputs: false
-group_by_length: false
-bf16: true
-tf32: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:

configs/vicuna_13B_4bit_reflect.yml

@@ -1,45 +0,0 @@
-base_model: anon8231489123/vicuna-13b-GPTQ-4bit-128g
-base_model_config: anon8231489123/vicuna-13b-GPTQ-4bit-128g
-model_type: LlamaForCausalLM
-tokenizer_type: LlamaTokenizer
-load_in_8bit: false
-load_4bit: true
-gptq_groupsize: 128
-gptq_model_v1: false
-datasets:
-# https://github.com/vaguenebula/AlpacaDataReflect/blob/main/alpaca_reflect_pruned.json
-  - path: data/alpaca_reflect_pruned.jsonl
-    type: reflection
-dataset_prepared_path: data/last_run_prepared
-val_set_size: 0.04
-adapter: lora
-lora_model_dir:
-sequence_len: 2048
-max_packed_sequence_len: 2048
-lora_r: 8
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules:
-  - q_proj
-  - v_proj
-#  - k_proj
-#  - o_proj
-lora_fan_in_fan_out: false
-wandb_project:
-wandb_watch:
-wandb_run_id:
-wandb_log_model: checkpoint
-output_dir: ./lora-reflect
-batch_size: 8
-micro_batch_size: 2
-num_epochs: 3
-learning_rate: 0.00003
-train_on_inputs: false
-group_by_length: false
-bf16: true
-tf32: true
-gradient_checkpointing: false
-early_stopping_patience: 3
-resume_from_checkpoint:
-local_rank:
-flash_attention: true

data/README.md

@@ -1,6 +1,5 @@
 
-- Download some datasets
--
+## Download some datasets
 ```shell
 curl https://raw.githubusercontent.com/tloen/alpaca-lora/main/alpaca_data_gpt4.json -o data/raw/alpaca_data_gpt4.json
 curl https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json -L -o data/raw/vicuna_cleaned.json
@@ -8,16 +7,17 @@ curl https://github.com/teknium1/GPTeacher/blob/main/Instruct/gpt4-instruct-simi
 curl https://github.com/teknium1/GPTeacher/blob/main/Roleplay/roleplay-similarity_0.6-instruct-dataset.json?raw=true -L -o data/raw/roleplay-similarity_0.6-instruct-dataset.json
 ```
 
-- Convert the JSON data files to JSONL.
+## Convert the JSON data files to JSONL.
 
 ```shell
-python3 ./scripts/alpaca_json_to_jsonl.py --input data/alpaca_data_gpt4.json > data/alpaca_data_gpt4.jsonl
-python3 ./scripts/alpaca_json_to_jsonl.py --input data/raw/vicuna_cleaned.json > data/vicuna_cleaned.jsonl
-python3 ./scripts/alpaca_json_to_jsonl.py --input data/raw/roleplay-similarity_0.6-instruct-dataset.json > data/roleplay-similarity_0.6-instruct-dataset.jsonl
-python3 ./scripts/alpaca_json_to_jsonl.py --input data/raw/gpt4-instruct-similarity-0.6-dataset.json > data/gpt4-instruct-similarity-0.6-dataset.jsonl
+python3 ./scripts/alpaca_json_to_jsonl.py --file data/alpaca_data_gpt4.json --output data/alpaca_data_gpt4.jsonl
+python3 ./scripts/alpaca_json_to_jsonl.py --file data/raw/vicuna_cleaned.json --output data/vicuna_cleaned.jsonl
+python3 ./scripts/alpaca_json_to_jsonl.py --file data/raw/roleplay-similarity_0.6-instruct-dataset.json --output data/roleplay-similarity_0.6-instruct-dataset.jsonl
+python3 ./scripts/alpaca_json_to_jsonl.py --file data/raw/gpt4-instruct-similarity-0.6-dataset.json --output data/gpt4-instruct-similarity-0.6-dataset.jsonl
 ```
+---
 
-- Using JSONL makes it easier to subset the data if you want a smaller training set, i.e get 2000 random examples.
+Using JSONL makes it easier to subset the data if you want a smaller training set, i.e get 2000 random examples.
 
 ```shell
 shuf -n2000 data/vicuna_cleaned.jsonl > data/vicuna_cleaned.subset0.jsonl

docker-compose.yaml

@@ -0,0 +1,20 @@
+# version: '3.8'
+services:
+  axolotl:
+    build:
+      context: .
+      dockerfile: ./docker/Dockerfile
+    volumes:
+      - .:/workspace/axolotl
+      - ~/.cache/huggingface/:/root/.cache/huggingface/
+    # set environment variables
+    environment:
+      - WANDB_API_KEY=${WANDB_API_KEY}
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              # count: 1
+              capabilities: [gpu]
+    command: tail -f /dev/null

docker/Dockerfile

@@ -0,0 +1,26 @@
+ARG BASE_TAG=main-base
+FROM winglian/axolotl-base:$BASE_TAG
+
+ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
+ARG AXOLOTL_EXTRAS=""
+
+RUN apt-get update && \
+    apt-get install -y vim curl
+
+WORKDIR /workspace
+
+RUN pip3 install --force-reinstall "peft @ git+https://github.com/huggingface/peft.git@main" \
+            "accelerate @ git+https://github.com/huggingface/accelerate.git@main" \
+            "transformers @ git+https://github.com/huggingface/transformers.git@main"
+
+RUN git clone --depth=1 https://github.com/OpenAccess-AI-Collective/axolotl.git
+# If AXOLOTL_EXTRAS is set, append it in brackets
+RUN cd axolotl && \
+    if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
+        pip install -e .[$AXOLOTL_EXTRAS]; \
+    else \
+        pip install -e .; \
+    fi
+
+# helper for huggingface-login cli
+RUN git config --global credential.helper store

docker/Dockerfile-base

@@ -0,0 +1,100 @@
+ARG CUDA_VERSION="11.8.0"
+ARG CUDNN_VERSION="8"
+ARG UBUNTU_VERSION="22.04"
+ARG MAX_JOBS=4
+
+FROM nvidia/cuda:$CUDA_VERSION-cudnn$CUDNN_VERSION-devel-ubuntu$UBUNTU_VERSION as base-builder
+
+ENV PATH="/root/miniconda3/bin:${PATH}"
+
+ARG PYTHON_VERSION="3.9"
+ARG PYTORCH="2.0.0"
+ARG CUDA="118"
+
+ENV PYTHON_VERSION=$PYTHON_VERSION
+
+RUN apt-get update
+RUN apt-get install -y wget git build-essential ninja-build git-lfs libaio-dev && rm -rf /var/lib/apt/lists/*
+
+RUN wget \
+    https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh \
+    && mkdir /root/.conda \
+    && bash Miniconda3-latest-Linux-x86_64.sh -b \
+    && rm -f Miniconda3-latest-Linux-x86_64.sh
+
+RUN conda create -n "py${PYTHON_VERSION}" python="${PYTHON_VERSION}"
+
+ENV PATH="/root/miniconda3/envs/py${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} torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu$CUDA
+
+
+FROM base-builder AS flash-attn-builder
+
+WORKDIR /workspace
+
+ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
+
+RUN git clone https://github.com/HazyResearch/flash-attention.git && \
+    cd flash-attention && \
+    python3 setup.py bdist_wheel && \
+    cd csrc/fused_dense_lib && \
+    python3 setup.py bdist_wheel && \
+    cd ../xentropy && \
+    python3 setup.py bdist_wheel && \
+    cd ../rotary && \
+    python3 setup.py bdist_wheel && \
+    cd ../layer_norm && \
+    python3 setup.py bdist_wheel
+
+FROM base-builder AS deepspeed-builder
+
+ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
+
+WORKDIR /workspace
+
+RUN git clone https://github.com/microsoft/DeepSpeed.git && \
+    cd DeepSpeed && \
+    MAX_CONCURRENCY=8 DS_BUILD_SPARSE_ATTN=0 DS_BUILD_OPS=1 python3 setup.py bdist_wheel
+
+FROM base-builder AS bnb-builder
+
+WORKDIR /workspace
+ARG CUDA="118"
+ENV CUDA=$CUDA
+
+RUN git clone https://github.com/TimDettmers/bitsandbytes.git && \
+    cd bitsandbytes && \
+    CUDA_VERSION=$CUDA make cuda11x && \
+    python setup.py bdist_wheel
+
+FROM base-builder
+
+# recompile apex
+RUN python3 -m pip uninstall -y apex
+RUN git clone https://github.com/NVIDIA/apex
+#  `MAX_JOBS=1` disables parallel building to avoid cpu memory OOM when building image on GitHub Action (standard) runners
+RUN cd apex && MAX_JOBS=1 python3 -m pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" ./
+
+RUN mkdir -p /workspace/builds
+COPY --from=bnb-builder /workspace/bitsandbytes /workspace/builds/bitsandbytes
+
+RUN mkdir -p /workspace/wheels/bitsandbytes
+COPY --from=deepspeed-builder /workspace/DeepSpeed/dist/deepspeed-*.whl wheels
+COPY --from=bnb-builder /workspace/bitsandbytes/dist/bitsandbytes-*.whl wheels
+COPY --from=bnb-builder /workspace/bitsandbytes/bitsandbytes/libbitsandbytes*.so wheels/bitsandbytes
+COPY --from=flash-attn-builder /workspace/flash-attention/dist/flash_attn-*.whl wheels
+COPY --from=flash-attn-builder /workspace/flash-attention/csrc/fused_dense_lib/dist/fused_dense_lib-*.whl wheels
+COPY --from=flash-attn-builder /workspace/flash-attention/csrc/xentropy/dist/xentropy_cuda_lib-*.whl wheels
+COPY --from=flash-attn-builder /workspace/flash-attention/csrc/rotary/dist/rotary_emb-*.whl wheels
+COPY --from=flash-attn-builder /workspace/flash-attention/csrc/layer_norm/dist/dropout_layer_norm-*.whl wheels
+
+RUN pip3 install wheels/deepspeed-*.whl wheels/flash_attn-*.whl wheels/fused_dense_lib-*.whl wheels/xentropy_cuda_lib-*.whl wheels/rotary_emb-*.whl wheels/dropout_layer_norm-*.whl
+RUN cd /workspace/builds/bitsandbytes && python3 setup.py install
+RUN git lfs install --skip-repo
+RUN pip3 install awscli && \
+    # The base image ships with `pydantic==1.8.2` which is not working
+    pip3 install -U --no-cache-dir pydantic

docker/Dockerfile-runpod

@@ -0,0 +1,14 @@
+ARG BASE_TAG=main
+FROM winglian/axolotl:$BASE_TAG
+
+COPY scripts/runpod-entrypoint.sh /root/runpod-entrypoint.sh
+
+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 && \
+    chmod +x /workspace/axolotl/scripts/runpod-entrypoint.sh && \
+    chmod +x /root/runpod-entrypoint.sh
+
+ENTRYPOINT ["/root/runpod-entrypoint.sh"]
+CMD ["sleep", "infinity"]

ds_config.json

@@ -1,37 +1,57 @@
 {
+  "zero_optimization": {
+    "stage": 3,
+    "offload_optimizer": {
+      "device": "cpu",
+      "pin_memory": true
+    },
+    "offload_param": {
+      "device": "cpu",
+      "pin_memory": true
+    },
+    "overlap_comm": true,
+    "contiguous_gradients": true,
+    "sub_group_size": 0,
+    "reduce_bucket_size": "auto",
+    "stage3_prefetch_bucket_size": "auto",
+    "stage3_param_persistence_threshold": "auto",
+    "stage3_max_live_parameters": 0,
+    "stage3_max_reuse_distance": 0,
+    "stage3_gather_16bit_weights_on_model_save": true
+  },
   "bf16": {
     "enabled": "auto"
   },
   "fp16": {
     "enabled": "auto",
+    "auto_cast": false,
     "loss_scale": 0,
+    "initial_scale_power": 32,
     "loss_scale_window": 1000,
-    "initial_scale_power": 16,
     "hysteresis": 2,
     "min_loss_scale": 1
   },
-  "scheduler": {
-    "type": "WarmupLR",
+  "optimizer": {
+    "type": "AdamW",
     "params": {
-      "warmup_min_lr": "auto",
-      "warmup_max_lr": "auto",
-      "warmup_num_steps": "auto"
+      "lr": "auto",
+      "betas": [
+        0.9,
+        0.999
+      ],
+      "eps": 1e-8,
+      "weight_decay": "auto"
     }
   },
-  "zero_optimization": {
-    "stage": 2,
-    "overlap_comm": true,
-    "allgather_partitions": true,
-    "allgather_bucket_size": 5e8,
-    "contiguous_gradients": true,
-    "reduce_bucket_size": "auto",
-    "reduce_scatter": true
+  "scheduler": {
+    "type": "OneCycle",
+    "params": {
+      "cycle_min_lr": 0.00001,
+      "cycle_max_lr": 0.00003,
+      "cycle_first_step_size": 120
+    }
   },
-  "gradient_accumulation_steps": "auto",
-  "gradient_clipping": "auto",
-  "steps_per_print": 5,
   "train_batch_size": "auto",
   "train_micro_batch_size_per_gpu": "auto",
-  "wall_clock_breakdown": false,
-  "round_robin_gradients": true
+  "wall_clock_breakdown": false
 }

examples/cerebras/qlora.yml

@@ -0,0 +1,60 @@
+base_model: cerebras/Cerebras-GPT-1.3B
+base_model_config: cerebras/Cerebras-GPT-1.3B
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+push_dataset_to_hub:
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
+adapter: qlora
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len: 2048
+lora_r: 16
+lora_alpha: 32
+lora_dropout: 0.05
+lora_target_modules:
+  - c_fc
+  - c_attn
+  - c_proj
+lora_target_linear:
+lora_fan_in_fan_out:
+wandb_project:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./qlora-out
+batch_size: 4
+micro_batch_size: 4
+num_epochs: 2
+optimizer: paged_adamw_8bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.0002
+train_on_inputs: false
+group_by_length: true
+bf16: true
+fp16: false
+tf32: true
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention: true
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 10
+eval_steps: 20
+save_steps:
+debug:
+deepspeed:
+weight_decay: 0.1
+fsdp:
+fsdp_config:
+special_tokens:
+  pad_token: "<|endoftext|>"

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

@@ -0,0 +1,63 @@
+base_model: tiiuae/falcon-7b
+base_model_config: tiiuae/falcon-7b
+trust_remote_code: true
+model_type: AutoModelForCausalLM
+tokenizer_type: AutoTokenizer
+load_in_8bit: true
+load_in_4bit: false
+gptq: false
+strict: false
+push_dataset_to_hub:
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca:chat
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
+adapter: lora
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len:
+lora_r: 16
+lora_alpha: 32
+lora_dropout: 0.0
+lora_target_modules:
+lora_target_linear: true
+lora_fan_in_fan_out:
+wandb_project:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./falcon-7b
+batch_size: 2
+micro_batch_size: 1
+num_epochs: 4
+optimizer: adamw_bnb_8bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.00003
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: true
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention: true
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 40
+eval_steps: 5
+save_steps: 43
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:
+  pad_token: "<|endoftext|>"
+  bos_token: ">>ABSTRACT<<"
+  eos_token: "<|endoftext|>"

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

@@ -0,0 +1,92 @@
+# 1b: tiiuae/falcon-rw-1b
+# 40b: tiiuae/falcon-40b
+base_model: tiiuae/falcon-7b
+base_model_config: tiiuae/falcon-7b
+# required by falcon custom model code: https://huggingface.co/tiiuae/falcon-7b/tree/main
+trust_remote_code: true
+model_type: AutoModelForCausalLM
+tokenizer_type: AutoTokenizer
+load_in_8bit: false
+# enable 4bit for QLoRA
+load_in_4bit: true
+gptq: false
+strict: false
+push_dataset_to_hub:
+datasets:
+  - path: QingyiSi/Alpaca-CoT
+    data_files:
+      - Chain-of-Thought/formatted_cot_data/gsm8k_train.json
+    type: "alpaca:chat"
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
+# enable QLoRA
+adapter: qlora
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len:
+
+# hyperparameters from QLoRA paper Appendix B.2
+# "We find hyperparameters to be largely robust across datasets"
+lora_r: 64
+lora_alpha: 16
+# 0.1 for models up to 13B
+# 0.05 for 33B and 65B models
+lora_dropout: 0.05
+# add LoRA modules on all linear layers of the base model
+lora_target_modules:
+lora_target_linear: true
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./qlora-out
+
+# QLoRA paper Table 9
+# - 16 for 7b & 13b
+# - 32 for 33b, 64 for 64b
+# Max size tested on A6000
+# - 7b: 40
+# - 40b: 4
+# decrease if OOM, increase for max VRAM utilization
+micro_batch_size: 1
+gradient_accumulation_steps: 2
+num_epochs: 3
+# Optimizer for QLoRA
+optimizer: paged_adamw_32bit
+torchdistx_path:
+lr_scheduler: cosine
+# QLoRA paper Table 9
+# - 2e-4 for 7b & 13b
+# - 1e-4 for 33b & 64b
+learning_rate: 0.0002
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: true
+gradient_checkpointing: true
+# stop training after this many evaluation losses have increased in a row
+# https://huggingface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppingCallback
+early_stopping_patience: 3
+resume_from_checkpoint:
+auto_resume_from_checkpoints: true
+local_rank:
+logging_steps: 1
+xformers_attention: true
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 10
+eval_steps: 5
+save_steps: 10
+debug:
+deepspeed:
+weight_decay: 0.000001
+fsdp:
+fsdp_config:
+special_tokens:
+  pad_token: "<|endoftext|>"
+  bos_token: ">>ABSTRACT<<"
+  eos_token: "<|endoftext|>"

examples/falcon/config-7b.yml

@@ -0,0 +1,63 @@
+base_model: tiiuae/falcon-7b
+base_model_config: tiiuae/falcon-7b
+trust_remote_code: true
+model_type: AutoModelForCausalLM
+tokenizer_type: AutoTokenizer
+load_in_8bit: false
+load_in_4bit: false
+gptq: false
+strict: false
+push_dataset_to_hub:
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca:chat
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
+adapter:
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len:
+lora_r: 64
+lora_alpha: 32
+lora_dropout: 0.0
+lora_target_modules:
+lora_target_linear: true
+lora_fan_in_fan_out:
+wandb_project:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./falcon-7b
+batch_size: 2
+micro_batch_size: 1
+num_epochs: 4
+optimizer: adamw_bnb_8bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.00003
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: true
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention: true
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 40
+eval_steps: 5
+save_steps: 43
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:
+  pad_token: "<|endoftext|>"
+  bos_token: ">>ABSTRACT<<"
+  eos_token: "<|endoftext|>"

examples/gptj/qlora.yml

@@ -0,0 +1,57 @@
+base_model: EleutherAI/gpt-j-6b
+base_model_config: EleutherAI/gpt-j-6b
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+push_dataset_to_hub:
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
+adapter: qlora
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len:
+lora_r: 8
+lora_alpha: 32
+lora_dropout: 0.05
+lora_target_modules:
+lora_target_linear: true
+lora_fan_in_fan_out:
+wandb_project:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./qlora-out
+gradient_accumulation_steps: 2
+micro_batch_size: 2
+num_epochs: 2
+optimizer: paged_adamw_8bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.0001
+train_on_inputs: false
+group_by_length: true
+bf16: true
+fp16: false
+tf32: true
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention: true
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 10
+eval_steps: 20
+save_steps:
+debug:
+deepspeed:
+weight_decay: 0.1
+fsdp:
+fsdp_config:
+special_tokens:
+  pad_token: "<|endoftext|>"

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

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

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

@@ -0,0 +1,62 @@
+base_model: Neko-Institute-of-Science/LLaMA-7B-4bit-128g
+base_model_config: Neko-Institute-of-Science/LLaMA-7B-4bit-128g
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+trust_remote_code:
+load_in_8bit: true
+gptq: true
+datasets:
+  - path: vicgalle/alpaca-gpt4
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.02
+adapter:
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len:
+lora_r: 8
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_modules:
+  - q_proj
+  - v_proj
+lora_fan_in_fan_out: false
+wandb_project: llama-7b-lora-int4
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./llama-7b-lora-int4
+gradient_accumulation_steps: 1
+micro_batch_size: 1
+num_epochs: 3
+optimizer: adamw_bnb_8bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.0000002
+train_on_inputs: false
+group_by_length: false
+fp16: true
+bf16: false
+tf32: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 5
+xformers_attention:
+flash_attention:
+gradient_checkpointing: true
+gptq_groupsize: 128
+gptq_model_v1: false
+warmup_steps: 20
+eval_steps: 110
+save_steps: 660
+debug:
+deepspeed:
+weight_decay: 0.0001
+fsdp:
+fsdp_config:
+tokens:
+  pad_token: "[PAD]"
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"

examples/jeopardy-bot/config.yml

@@ -0,0 +1,55 @@
+base_model: huggyllama/llama-7b
+base_model_config: huggyllama/llama-7b
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+load_in_8bit: false
+datasets:
+  - path: openaccess-ai-collective/jeopardy
+    type: jeopardy
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.02
+adapter:
+lora_model_dir:
+sequence_len: 512
+max_packed_sequence_len:
+lora_r:
+lora_alpha:
+lora_dropout:
+lora_target_modules:
+lora_fan_in_fan_out: false
+wandb_project:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./jeopardy-bot-7b
+gradient_accumulation_steps: 1
+micro_batch_size: 1
+num_epochs: 3
+optimizer: adamw_bnb_8bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.00003
+train_on_inputs: false
+group_by_length: false
+bf16: true
+tf32: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 5
+xformers_attention: true
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 20
+eval_steps: 110
+save_steps: 660
+debug:
+deepspeed:
+weight_decay: 0.1
+fsdp:
+fsdp_config:
+tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"

examples/mpt-7b/README.md

@@ -0,0 +1,6 @@
+# MPT-7B
+
+```shell
+accelerate launch scripts/finetune.py examples/mpt-7b/config.yml
+
+```

examples/mpt-7b/config.yml

@@ -0,0 +1,58 @@
+base_model: mosaicml/mpt-7b
+base_model_config: mosaicml/mpt-7b
+tokenizer_type: AutoTokenizer
+trust_remote_code: true  # required for mpt as their model class is not merged into transformers yet
+load_in_8bit: false
+datasets:
+  - path: vicgalle/alpaca-gpt4
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.02
+adapter:
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len:
+lora_r: 8
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_modules:
+  - q_proj
+  - v_proj
+lora_fan_in_fan_out: false
+wandb_project: mpt-alpaca-7b
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./mpt-alpaca-7b
+gradient_accumulation_steps: 1
+micro_batch_size: 1
+num_epochs: 3
+optimizer: adamw_bnb_8bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.0000002
+train_on_inputs: false
+group_by_length: false
+bf16: true
+tf32: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 5
+xformers_attention:
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 20
+eval_steps: 110
+save_steps: 660
+debug:
+deepspeed:
+weight_decay: 0.0001
+fsdp:
+fsdp_config:
+tokens:
+  pad_token: "<|padding|>"
+  bos_token: "<|endoftext|>"
+  eos_token: "<|endoftext|>"
+  unk_token: "<|endoftext|>"

examples/openllama-3b/README.md

@@ -0,0 +1,16 @@
+# openllama-3b
+
+Basic full tune
+```shell
+accelerate launch scripts/finetune.py examples/openllama-3b/config.yml
+```
+
+LoRA
+```shell
+accelerate launch scripts/finetune.py examples/openllama-3b/lora.yml
+```
+
+QLoRA
+```shell
+accelerate launch scripts/finetune.py examples/openllama-3b/qlora.yml
+```

examples/openllama-3b/config.yml

@@ -0,0 +1,62 @@
+base_model: openlm-research/open_llama_3b
+base_model_config: openlm-research/open_llama_3b
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+load_in_8bit: false
+load_in_4bit: false
+strict: false
+push_dataset_to_hub:
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.02
+adapter:
+lora_model_dir:
+sequence_len: 256
+max_packed_sequence_len:
+lora_r:
+lora_alpha:
+lora_dropout:
+lora_target_modules:
+lora_target_linear:
+lora_fan_in_fan_out:
+wandb_project:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./openllama-out
+gradient_accumulation_steps: 1
+micro_batch_size: 1
+num_epochs: 3
+optimizer: adamw_bnb_8bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.00001
+train_on_inputs: false
+group_by_length: false
+float16: true
+bf16: false
+fp16: false
+tf32: false
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention: true
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 10
+eval_steps: 50
+save_steps:
+debug:
+deepspeed:
+weight_decay: 0.1
+fsdp:
+fsdp_config:
+special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"

examples/openllama-3b/lora.yml

@@ -0,0 +1,67 @@
+base_model: openlm-research/open_llama_3b
+base_model_config: openlm-research/open_llama_3b
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+load_in_8bit: true
+load_in_4bit: false
+strict: false
+push_dataset_to_hub:
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.02
+adapter: lora
+lora_model_dir:
+sequence_len: 256
+max_packed_sequence_len:
+lora_r: 8
+lora_alpha: 16
+lora_dropout: 0.0
+lora_target_modules:
+  - gate_proj
+  - down_proj
+  - up_proj
+  - q_proj
+  - v_proj
+  - k_proj
+  - o_proj
+lora_fan_in_fan_out:
+wandb_project:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./lora-out
+batch_size: 16
+micro_batch_size: 4
+num_epochs: 3
+optimizer: adamw_bnb_8bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.0002
+train_on_inputs: false
+group_by_length: false
+bf16: false
+fp16: true
+tf32: false
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention: true
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 10
+eval_steps: 50
+save_steps:
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"

examples/openllama-3b/qlora.yml

@@ -0,0 +1,61 @@
+base_model: openlm-research/open_llama_3b
+base_model_config: openlm-research/open_llama_3b
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+push_dataset_to_hub:
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
+adapter: qlora
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len: 2048
+lora_r: 8
+lora_alpha: 32
+lora_dropout: 0.05
+lora_target_modules:
+lora_target_linear: true
+lora_fan_in_fan_out:
+wandb_project:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./qlora-out
+batch_size: 4
+micro_batch_size: 4
+num_epochs: 2
+optimizer: paged_adamw_32bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.0002
+train_on_inputs: false
+group_by_length: true
+bf16: true
+fp16: false
+tf32: true
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention: true
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 10
+eval_steps: 20
+save_steps:
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"

examples/pythia-12b/README.md

@@ -0,0 +1,9 @@
+# Pythia 12B
+
+- Single-GPU A100 only (?)
+
+```shell
+python scripts/finetune.py examples/pythia-12b/config.yml
+```
+
+⚠️ Multiple-GPU A100 - Doesn't seem to work with multi-gpu without causing OOM! ⚠️

examples/pythia-12b/config.yml

@@ -0,0 +1,49 @@
+base_model: EleutherAI/pythia-12b-deduped
+base_model_config: EleutherAI/pythia-12b-deduped
+base_model_ignore_patterns: pytorch*  # prefer safetensors
+model_type: GPTNeoXForCausalLM
+tokenizer_type: AutoTokenizer
+load_in_8bit: false
+load_in_4bit: false
+gptq: false
+device_map: auto
+datasets:
+  - path: vicgalle/alpaca-gpt4
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.05
+adapter:
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len: 2048
+lora_r: 64
+lora_alpha: 32
+lora_dropout: 0.0
+lora_target_modules:
+lora_target_linear: true
+lora_fan_in_fan_out: true  # pythia/GPTNeoX lora specific
+wandb_project:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./pythia-12b
+gradient_accumulation_steps: 1
+micro_batch_size: 1
+num_epochs: 5
+learning_rate: 0.00003
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+train_on_inputs: false
+group_by_length: false
+bf16: false
+fp16: false
+float16: true
+tf32: true
+flash_optimum: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+gradient_checkpointing: true
+fsdp:
+fsdp_config:
+collator_pad_to_longest: true

examples/pythia/lora.yml

@@ -1,36 +1,29 @@
 base_model: EleutherAI/pythia-1.4b-deduped
-model_type: GPTNeoXForCausalLM
-tokenizer_type: AutoTokenizer
+base_model_config: EleutherAI/pythia-1.4b-deduped
 load_in_8bit: true
 datasets:
-  - path: data/alpaca_data_gpt4.jsonl
+  - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-  - path: data/vicuna_cleaned.jsonl
-    type: sharegpt
-  - path: data/gpt4-instruct-similarity-0.6-dataset.jsonl
-    type: gpteacher
-  - path: data/roleplay-similarity_0.6-instruct-dataset.jsonl
-    type: gpteacher
 dataset_prepared_path: last_run_prepared
 val_set_size: 0.05
 adapter: lora
 lora_model_dir:
-sequence_len: 2048
-lora_r: 8
+sequence_len: 512
+lora_r: 16
 lora_alpha: 32
 lora_dropout: 0.05
 lora_target_modules:
   - query_key_value
-#  - xxx
+lora_target_linear:
 lora_fan_in_fan_out: true  # pythia/GPTNeoX lora specific
-wandb_project: pythia-1.4b-lora
+wandb_project:
 wandb_watch:
 wandb_run_id:
-wandb_log_model: checkpoint
-output_dir: ./lora-alpaca
-batch_size: 48
+wandb_log_model:
+output_dir: ./lora-alpaca-pythia
+gradient_accumulation_steps: 1
 micro_batch_size: 4
-num_epochs: 5
+num_epochs: 3
 learning_rate: 0.00001
 train_on_inputs: false
 group_by_length: false
@@ -39,3 +32,6 @@ tf32: True
 early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
+weight_decay: 0.1
+eval_steps: 20
+logging_steps: 1

examples/redpajama/README.md

@@ -0,0 +1,6 @@
+# RedPajama 3B preview release
+
+```shell
+accelerate launch scripts/finetune.py examples/redpajama/config-3b.yml
+
+```

examples/redpajama/config-3b.yml

@@ -0,0 +1,59 @@
+base_model: togethercomputer/RedPajama-INCITE-Chat-3B-v1
+base_model_config: togethercomputer/RedPajama-INCITE-Chat-3B-v1
+model_type: GPTNeoXForCausalLM
+tokenizer_type: AutoTokenizer
+trust_remote_code:
+load_in_8bit: false
+datasets:
+  - path: vicgalle/alpaca-gpt4
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.02
+adapter:
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len:
+lora_r: 8
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_modules:
+  - q_proj
+  - v_proj
+lora_fan_in_fan_out: false
+wandb_project: redpajama-alpaca-3b
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./redpajama-alpaca-3b
+batch_size: 4
+micro_batch_size: 1
+num_epochs: 3
+optimizer: adamw_bnb_8bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.0000002
+train_on_inputs: false
+group_by_length: false
+bf16: true
+tf32: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 5
+xformers_attention:
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 20
+eval_steps: 110
+save_steps: 660
+debug:
+deepspeed:
+weight_decay: 0.0001
+fsdp:
+fsdp_config:
+tokens:
+  pad_token: "<|padding|>"
+  bos_token: "<|endoftext|>"
+  eos_token: "<|endoftext|>"
+  unk_token: "<|endoftext|>"

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

@@ -0,0 +1,55 @@
+base_model: replit/replit-code-v1-3b
+base_model_config: replit/replit-code-v1-3b
+trust_remote_code: true
+load_in_8bit: false
+datasets:
+  - path: vicgalle/alpaca-gpt4
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.05
+adapter: lora
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len:
+lora_r: 8
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_modules:
+  - Wqkv
+  - mlp_up
+  - mlp_down
+lora_fan_in_fan_out:
+wandb_project: lora-replit
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./lora-replit
+batch_size: 8
+micro_batch_size: 1
+num_epochs: 3
+optimizer:
+torchdistx_path:
+lr_scheduler:
+learning_rate: 0.00001
+train_on_inputs: false
+group_by_length: false
+bf16: true
+tf32: true
+gradient_checkpointing:
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 20
+eval_steps: 50
+save_steps:
+debug:
+deepspeed:
+weight_decay: 0
+fsdp:
+fsdp_config:
+#special_tokens:

requirements-dev.txt

@@ -0,0 +1,3 @@
+pre-commit
+black
+mypy

requirements-tests.txt

@@ -0,0 +1 @@
+pytest

requirements.txt

@@ -1,14 +1,20 @@
 peft @ git+https://github.com/huggingface/peft.git
 transformers @ git+https://github.com/huggingface/transformers.git
-attrdict
+bitsandbytes>=0.39.0
+accelerate
+addict
 fire
 PyYAML==6.0
-black
-bitsandbytes
 datasets
-accelerate
+accelerate>=0.19.0
 sentencepiece
 wandb
-flash-attn
-deepspeed
 einops
+xformers
+optimum
+# qlora things
+bert-score==0.3.13
+evaluate==0.4.0
+rouge-score==0.1.2
+scipy
+scikit-learn==1.2.2

scripts/alpaca_json_to_jsonl.py

@@ -1,24 +1,38 @@
+"""Module to convert json file to jsonl"""
+
 import os
 import sys
 from pathlib import Path
+from typing import Optional, Union
 
 import fire
-from typing import Optional
+
+from axolotl.convert import (
+    FileReader,
+    FileWriter,
+    JsonlSerializer,
+    JsonParser,
+    JsonToJsonlConverter,
+    StdoutWriter,
+)
 
 # add src to the pythonpath so we don't need to pip install this
 project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
 src_dir = os.path.join(project_root, "src")
 sys.path.insert(0, src_dir)
 
-from axolotl.convert import *
-
 
 def main(
-    input: Path,
+    file: Path,
     output: Optional[Path] = None,
     to_stdout: Optional[bool] = False,
 ):
+    """
+    Convert a json file to jsonl
+    """
+
     file_reader = FileReader()
+    writer: Union[StdoutWriter, FileWriter]
     if to_stdout or output is None:
         writer = StdoutWriter()
     else:
@@ -28,7 +42,7 @@ def main(
 
     converter = JsonToJsonlConverter(file_reader, writer, json_parser, jsonl_serializer)
 
-    converter.convert(input, output)
+    converter.convert(file, output)
 
 
 if __name__ == "__main__":

scripts/finetune.py

@@ -1,305 +1,136 @@
+"""Prepare and train a model on a dataset. Can also infer from a model or merge lora"""
+
+import importlib
 import logging
-import math
 import os
 import random
 import signal
 import sys
-from hashlib import md5
 from pathlib import Path
+from typing import Any, Dict, List, Optional, Union
 
-import bitsandbytes as bnb
 import fire
 import torch
-import transformers
 import yaml
-from attrdict import AttrDefault
-from datasets import load_dataset, IterableDataset, Dataset, load_from_disk
-from torch import nn
-from transformers import (
-    AutoModelForCausalLM,
-    AutoTokenizer,
-    LlamaForCausalLM,
-    LlamaTokenizer,
-    EarlyStoppingCallback,
-    GenerationConfig,
-)
 
 # add src to the pythonpath so we don't need to pip install this
-from transformers.trainer_pt_utils import get_parameter_names
+from optimum.bettertransformer import BetterTransformer
+from transformers import GenerationConfig, TextStreamer
+
+from axolotl.utils.data import load_prepare_datasets, load_pretraining_dataset
+from axolotl.utils.dict import DictDefault
+from axolotl.utils.models import load_model, load_tokenizer
+from axolotl.utils.tokenization import check_dataset_labels
+from axolotl.utils.trainer import setup_trainer
+from axolotl.utils.validation import validate_config
+from axolotl.utils.wandb import setup_wandb_env_vars
 
 project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
 src_dir = os.path.join(project_root, "src")
 sys.path.insert(0, src_dir)
 
-from axolotl.datasets import TokenizedPromptDataset, ConstantLengthDataset
-from axolotl.prompt_tokenizers import (
-    AlpacaPromptTokenizingStrategy,
-    ShareGPTPromptTokenizingStrategy,
-    LLAMA_DEFAULT_PAD_TOKEN,
-    GPTeacherPromptTokenizingStrategy,
-    OpenAssistantPromptTokenizingStrategy, AlpacaReflectionPTStrategy,
-)
-from axolotl.prompters import AlpacaPrompter, GPTeacherPrompter, ShareGPTPrompter, ReflectAlpacaPrompter
 
 logging.basicConfig(level=os.getenv("LOG_LEVEL", "INFO"))
 DEFAULT_DATASET_PREPARED_PATH = "last_run_prepared"
 
 
-def setup_wandb_env_vars(cfg):
-    if cfg.wandb_project and len(cfg.wandb_project) > 0:
-        os.environ["WANDB_PROJECT"] = cfg.wandb_project
-        cfg.use_wandb = True
-        if cfg.wandb_watch and len(cfg.wandb_watch) > 0:
-            os.environ["WANDB_WATCH"] = cfg.wandb_watch
-        if cfg.wandb_log_model and len(cfg.wandb_log_model) > 0:
-            os.environ["WANDB_LOG_MODEL"] = cfg.wandb_log_model
-        if cfg.wandb_run_id and len(cfg.wandb_run_id) > 0:
-            os.environ["WANDB_RUN_ID"] = cfg.wandb_run_id
-
-
-def load_model(
-    base_model,
-    base_model_config,
-    model_type,
-    tokenizer_type,
-    cfg,
-    adapter="lora",
-    inference: bool = False,
-):
-    # TODO refactor as a kwarg
-    load_in_8bit = cfg.load_in_8bit
-    tokenizer = None
-    is_llama_derived_model = "llama" in base_model or "llama" in cfg.model_type.lower()
-
-    if adapter != "lora":
-        raise NotImplementedError(f"{adapter} peft adapter not available")
-    if is_llama_derived_model and cfg.flash_attention:
-        if cfg.device not in ["mps", "cpu"] and inference is False:
-            from axolotl.flash_attn import replace_llama_attn_with_flash_attn
-
-            logging.info("patching with flash attention")
-            replace_llama_attn_with_flash_attn()
-
-    torch_dtype = (torch.float16 if cfg.load_in_8bit or cfg.fp16 else torch.float32,)
-    try:
-        if cfg.load_4bit:
-            from alpaca_lora_4bit.monkeypatch.peft_tuners_lora_monkey_patch import (
-                replace_peft_model_with_int4_lora_model,
-            )
-
-            replace_peft_model_with_int4_lora_model()
-
-        from peft import (
-            LoraConfig,
-            get_peft_model,
-            prepare_model_for_int8_training,
-            PeftModel,
-        )
-    except Exception as e:
-        logging.exception(e)
-        raise e
-
-    try:
-        if cfg.load_4bit and is_llama_derived_model:
-            from alpaca_lora_4bit.autograd_4bit import load_llama_model_4bit_low_ram
-            from huggingface_hub import snapshot_download
-
-            cache_model_path = Path(snapshot_download(base_model))
-            files = (
-                list(cache_model_path.glob("*.pt"))
-                + list(cache_model_path.glob("*.safetensors"))
-                + list(cache_model_path.glob("*.bin"))
-            )
-            if len(files) > 0:
-                model_path = str(files[0])
-            else:
-                logging.warning(
-                    "unable to find a cached model file, this will likely fail..."
-                )
-                model_path = str(cache_model_path)
-            model, tokenizer = load_llama_model_4bit_low_ram(
-                base_model_config if base_model_config else base_model,
-                model_path,
-                device_map=cfg.device_map,
-                groupsize=cfg.gptq_groupsize if cfg.gptq_groupsize else -1,
-                is_v1_model=cfg.gptq_model_v1
-                if cfg.gptq_model_v1 is not None
-                else True,
-            )
-            load_in_8bit = False
-        elif is_llama_derived_model:
-            model = LlamaForCausalLM.from_pretrained(
-                base_model,
-                load_in_8bit=cfg.load_in_8bit,
-                torch_dtype=torch_dtype,
-                device_map=cfg.device_map,
-            )
-        else:
-            model = getattr(transformers, model_type).from_pretrained(
-                base_model,
-                load_in_8bit=cfg.load_in_8bit,
-                torch_dtype=torch_dtype,
-                device_map=cfg.device_map,
-            )
-    except Exception as e:
-        logging.error(
-            "Exception raised attempting to load model, retrying with AutoModelForCausalLM"
-        )
-        logging.exception(e)
-        model = AutoModelForCausalLM.from_pretrained(
-            base_model,
-            load_in_8bit=cfg.load_in_8bit,
-            torch_dtype=torch_dtype,
-            device_map=cfg.device_map,
-        )
-
-    if not tokenizer:
-        try:
-            if is_llama_derived_model:
-                tokenizer = LlamaTokenizer.from_pretrained(model)
-            else:
-                tokenizer = getattr(transformers, tokenizer_type).from_pretrained(model)
-        except:
-            tokenizer = AutoTokenizer.from_pretrained(base_model)
-
-    logging.debug(f"EOS: {tokenizer.eos_token_id} / {tokenizer.eos_token}")
-    logging.debug(f"BOS: {tokenizer.bos_token_id} / {tokenizer.bos_token}")
-    logging.debug(f"PAD: {tokenizer.pad_token_id} / {tokenizer.pad_token}")
-    logging.debug(f"UNK: {tokenizer.unk_token_id} / {tokenizer.unk_token}")
-
-    if tokenizer.__class__.__name__ in ["LlamaTokenizer", "LlamaTokenizerFast"]:
-        tokenizer.pad_token = LLAMA_DEFAULT_PAD_TOKEN
-
-    if tokenizer.__class__.__name__ == "GPTNeoXTokenizerFast":
-        tokenizer.add_special_tokens({"pad_token": "[PAD]"})
-        os.environ["TOKENIZERS_PARALLELISM"] = "false"
-
-    if load_in_8bit and not cfg.load_4bit:
-        logging.info("converting model w/ prepare_model_for_int8_training")
-        model = prepare_model_for_int8_training(model)
-
-    lora_config = LoraConfig(
-        r=cfg.lora_r,
-        lora_alpha=cfg.lora_alpha,
-        target_modules=cfg.lora_target_modules,
-        lora_dropout=cfg.lora_dropout,
-        fan_in_fan_out=cfg.lora_fan_in_fan_out,
-        bias="none",
-        task_type="CAUSAL_LM",
-    )
-
-    if cfg.lora_model_dir:
-        model = PeftModel.from_pretrained(
-            model,
-            cfg.lora_model_dir,
-            device_map=cfg.device_map,
-            torch_dtype=torch.float16,
-        )
-    else:
-        model = get_peft_model(model, lora_config)
-
-    if cfg.ddp:
-        model.to(f"cuda:{cfg.local_rank}")
-
-    if cfg.load_4bit:
-        # Scales to half
-        logging.info("Fitting 4bit scales and zeros to half")
-        for n, m in model.named_modules():
-            if "Autograd4bitQuantLinear" in str(type(m)) or "Linear4bitLt" in str(
-                type(m)
-            ):
-                if hasattr(m, "is_v1_model") and m.is_v1_model:
-                    m.zeros = m.zeros.half()
-                m.scales = m.scales.half()
-                m.bias = m.bias.half()
-
-    # TODO resume_from_checkpoint handling
-    model.print_trainable_parameters()
-    return model, tokenizer, lora_config
-
-
 def choose_device(cfg):
     def get_device():
-        if torch.cuda.is_available():
-            return "cuda"
-        else:
-            try:
-                if torch.backends.mps.is_available():
-                    return "mps"
-            except:
-                return "cpu"
+        try:
+            if torch.cuda.is_available():
+                return f"cuda:{cfg.local_rank}"
+
+            if torch.backends.mps.is_available():
+                return "mps"
+
+            raise SystemError("No CUDA/mps device found")
+        except Exception:  # pylint: disable=broad-exception-caught
+            return "cpu"
 
     cfg.device = get_device()
-    if cfg.device == "cuda":
-        cfg.device_map = {"": cfg.local_rank}
-    else:
-        cfg.device_map = {"": cfg.device}
+    if cfg.device_map != "auto":
+        if cfg.device.startswith("cuda"):
+            cfg.device_map = {"": cfg.local_rank}
+        else:
+            cfg.device_map = {"": cfg.device}
 
 
-def check_dataset_labels(dataset, tokenizer):
-    from termcolor import colored
-
-    # the dataset is already shuffled, so let's just check the first 5 elements
-    for idx in range(5):
-        # Get the input_ids, labels, and attention_mask from the dataset
-        input_ids = dataset[idx]["input_ids"]
-        labels = dataset[idx]["labels"]
-        attention_mask = dataset[idx]["attention_mask"]
-
-        # You can compare the input_ids and labels element-wise
-        # Remember to ignore positions with IGNORE_TOKEN_ID (if you use it) or attention_mask equal to 0
-        colored_tokens = []
-        for i, (input_id, label_id, mask) in enumerate(
-            zip(input_ids, labels, attention_mask)
-        ):
-            decoded_input_token = tokenizer.decode(input_id)
-            # Choose the color based on whether the label has the ignore value or not
-            color = (
-                "red" if label_id == -100 else ("yellow" if label_id == 0 else "green")
-            )
-            colored_token = colored(decoded_input_token, color) + colored(
-                f"({label_id}, {mask})", "white"
-            )
-            colored_tokens.append(colored_token)
-
-        logging.info(" ".join(colored_tokens))
-        logging.info("\n\n\n")
+def get_multi_line_input() -> Optional[str]:
+    print("Give me an instruction (Ctrl + D to finish): ")
+    instruction = ""
+    for line in sys.stdin:
+        instruction += line  # pylint: disable=consider-using-join
+    # instruction = pathlib.Path("/proc/self/fd/0").read_text()
+    return instruction
 
 
-def do_inference(cfg, model, tokenizer):
-    tokenizer.add_special_tokens({"unk_token": "<unk>"})
-    tokenizer.add_special_tokens({"bos_token": "<s>"})
-    tokenizer.add_special_tokens({"eos_token": "</s>"})
+def do_inference(cfg, model, tokenizer, prompter: Optional[str]):
+    default_tokens = {"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>"}
 
-    instruction = "Tell me a joke about dromedaries."
-    input = ""
-    prompt = "Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.\n\n### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:\n".format(
-        instruction=instruction, input=input
-    )
-    batch = tokenizer(prompt, return_tensors="pt", add_special_tokens=True)
+    for token, symbol in default_tokens.items():
+        # If the token isn't already specified in the config, add it
+        if not (cfg.special_tokens and token in cfg.special_tokens):
+            tokenizer.add_special_tokens({token: symbol})
 
-    model.eval()
-    with torch.no_grad():
-        # gc = GenerationConfig()  # TODO swap out and use this
-        generated = model.generate(
-            inputs=batch["input_ids"].to("cuda"),
-            do_sample=True,
-            use_cache=True,
-            repetition_penalty=1.1,
-            max_new_tokens=100,
-            temperature=0.9,
-            top_p=0.95,
-            top_k=40,
-            return_dict_in_generate=True,
-            output_attentions=False,
-            output_hidden_states=False,
-            output_scores=False,
+    prompter_module = None
+    if prompter:
+        prompter_module = getattr(
+            importlib.import_module("axolotl.prompters"), prompter
         )
-    print(tokenizer.decode(generated["sequences"].cpu().tolist()[0]))
+
+    if cfg.landmark_attention:
+        from axolotl.monkeypatch.llama_landmark_attn import set_model_mem_id
+
+        set_model_mem_id(model, tokenizer)
+        model.set_mem_cache_args(
+            max_seq_len=255, mem_freq=50, top_k=5, max_cache_size=None
+        )
+
+    while True:
+        print("=" * 80)
+        # support for multiline inputs
+        instruction = get_multi_line_input()
+        if not instruction:
+            return
+        if prompter_module:
+            prompt: str = next(
+                prompter_module().build_prompt(instruction=instruction.strip("\n"))
+            )
+        else:
+            prompt = instruction.strip()
+        batch = tokenizer(prompt, return_tensors="pt", add_special_tokens=True)
+
+        print("=" * 40)
+        model.eval()
+        with torch.no_grad():
+            generation_config = GenerationConfig(
+                repetition_penalty=1.1,
+                max_new_tokens=1024,
+                temperature=0.9,
+                top_p=0.95,
+                top_k=40,
+                bos_token_id=tokenizer.bos_token_id,
+                eos_token_id=tokenizer.eos_token_id,
+                pad_token_id=tokenizer.pad_token_id,
+                do_sample=True,
+                use_cache=True,
+                return_dict_in_generate=True,
+                output_attentions=False,
+                output_hidden_states=False,
+                output_scores=False,
+            )
+            streamer = TextStreamer(tokenizer)
+            generated = model.generate(
+                inputs=batch["input_ids"].to(cfg.device),
+                generation_config=generation_config,
+                streamer=streamer,
+            )
+        print("=" * 40)
+        print(tokenizer.decode(generated["sequences"].cpu().tolist()[0]))
 
 
 def choose_config(path: Path):
-    yaml_files = [file for file in path.glob("*.yml")]
+    yaml_files = list(path.glob("*.yml"))
 
     if not yaml_files:
         raise ValueError(
@@ -324,96 +155,8 @@ def choose_config(path: Path):
     return chosen_file
 
 
-def setup_trainer(cfg, train_dataset, eval_dataset, model, tokenizer):
-    total_num_steps = int(
-        math.ceil(len(train_dataset) * cfg.num_epochs / cfg.batch_size)
-    )
-    warmup_steps = min(int(0.03 * total_num_steps), 100)
-    logging_steps = max(min(int(0.005 * total_num_steps), 10), 1)
-    save_steps = eval_steps = min(int(0.05 * total_num_steps), 200)
-
-    training_arguments_kwargs = {}
-    if cfg.bf16 == "full":
-        training_arguments_kwargs["bf16_full_eval"] = True
-    else:
-        training_arguments_kwargs["bf16"] = cfg.bf16
-    training_arguments_kwargs["tf32"] = cfg.tf32
-    training_arguments_kwargs["warmup_steps"] = warmup_steps
-    training_arguments_kwargs["logging_steps"] = logging_steps
-    if cfg.gradient_checkpointing is not None:
-        training_arguments_kwargs["gradient_checkpointing"] = cfg.gradient_checkpointing
-
-    training_args = transformers.TrainingArguments(
-        per_device_train_batch_size=cfg.micro_batch_size,
-        gradient_accumulation_steps=cfg.gradient_accumulation_steps,
-        num_train_epochs=cfg.num_epochs,
-        learning_rate=cfg.learning_rate,
-        evaluation_strategy="steps" if cfg.val_set_size > 0 else "no",
-        save_strategy="steps",
-        eval_steps=eval_steps if cfg.val_set_size > 0 else None,
-        save_steps=save_steps,
-        output_dir=cfg.output_dir,
-        save_total_limit=3,
-        load_best_model_at_end=True if cfg.val_set_size > 0 else False,
-        ddp_find_unused_parameters=False if cfg.ddp else None,
-        group_by_length=cfg.group_by_length,
-        report_to="wandb" if cfg.use_wandb else None,
-        run_name=cfg.wandb_run_id if cfg.use_wandb else None,
-        **training_arguments_kwargs,
-    )
-
-    decay_parameters = get_parameter_names(model, [nn.LayerNorm])
-    decay_parameters = [name for name in decay_parameters if "bias" not in name]
-    optimizer_grouped_parameters = [
-        {
-            "params": [p for n, p in model.named_parameters() if n in decay_parameters],
-            "weight_decay": training_args.weight_decay,
-        },
-        {
-            "params": [
-                p for n, p in model.named_parameters() if n not in decay_parameters
-            ],
-            "weight_decay": 0.0,
-        },
-    ]
-
-    trainer_kwargs = {}
-
-    if cfg.load_in_8bit and not cfg.load_4bit:
-        adam_bnb_optim = bnb.optim.Adam8bit(
-            optimizer_grouped_parameters,
-            betas=(training_args.adam_beta1, training_args.adam_beta2),
-            eps=training_args.adam_epsilon,
-            lr=training_args.learning_rate,
-        )
-
-        # TODO optionally use torch.optim.OneCycleLR
-        lr_scheduler = transformers.get_cosine_schedule_with_warmup(
-            adam_bnb_optim,
-            training_args.warmup_steps,
-            total_num_steps,
-        )
-        trainer_kwargs["optimizers"] = (adam_bnb_optim, lr_scheduler)
-
-    # TODO on_save callback to sync checkpoints to GCP/AWS in background
-    if cfg.early_stopping_patience:
-        early_stop_cb = EarlyStoppingCallback(
-            cfg.early_stopping_patience,
-        )
-        trainer_kwargs["callbacks"] = [early_stop_cb]
-
-    trainer = transformers.Trainer(
-        model=model,
-        train_dataset=train_dataset,
-        eval_dataset=eval_dataset,
-        args=training_args,
-        data_collator=transformers.DataCollatorForSeq2Seq(
-            tokenizer, pad_to_multiple_of=8, return_tensors="pt", padding=True
-        ),
-        **trainer_kwargs,
-    )
-
-    return trainer
+def check_not_in(list1: List[str], list2: Union[Dict[str, Any], List[str]]) -> bool:
+    return not any(el in list2 for el in list1)
 
 
 def train(
@@ -425,30 +168,37 @@ def train(
         config = choose_config(config)
 
     # load the config from the yaml file
-    with open(config, "r") as f:
-        cfg: AttrDefault = AttrDefault(lambda: None, yaml.load(f, Loader=yaml.Loader))
+    with open(config, encoding="utf-8") as file:
+        cfg: DictDefault = DictDefault(yaml.safe_load(file))
     # if there are any options passed in the cli, if it is something that seems valid from the yaml,
     # then overwrite the value
-    cfg_keys = dict(cfg).keys()
-    for k in kwargs:
-        if k in cfg_keys:
+    cfg_keys = cfg.keys()
+    for k, _ in kwargs.items():
+        # if not strict, allow writing to cfg even if it's not in the yml already
+        if k in cfg_keys or not cfg.strict:
             # handle booleans
             if isinstance(cfg[k], bool):
                 cfg[k] = bool(kwargs[k])
             else:
                 cfg[k] = kwargs[k]
 
+    validate_config(cfg)
+
     # setup some derived config / hyperparams
-    cfg.gradient_accumulation_steps = cfg.batch_size // cfg.micro_batch_size
+    cfg.gradient_accumulation_steps = cfg.gradient_accumulation_steps or (
+        cfg.batch_size // cfg.micro_batch_size
+    )
+    cfg.batch_size = (
+        cfg.batch_size or cfg.micro_batch_size * cfg.gradient_accumulation_steps
+    )
     cfg.world_size = int(os.environ.get("WORLD_SIZE", 1))
     cfg.local_rank = int(os.environ.get("LOCAL_RANK", 0))
     choose_device(cfg)
-    cfg.ddp = cfg.world_size != 1
+    cfg.ddp = cfg.ddp if cfg.ddp is not None else cfg.world_size != 1
     if cfg.ddp:
         cfg.device_map = {"": int(os.environ.get("LOCAL_RANK", 0))}
-        cfg.gradient_accumulation_steps = (
-            cfg.gradient_accumulation_steps // cfg.world_size
-        )
+        cfg.batch_size = cfg.batch_size * cfg.world_size
+
     setup_wandb_env_vars(cfg)
     if cfg.device == "mps":
         cfg.load_in_8bit = False
@@ -457,133 +207,80 @@ def train(
             cfg.fp16 = True
         cfg.bf16 = False
 
+    if cfg.tf32:
+        torch.backends.cuda.matmul.allow_tf32 = True
+
+    # load the tokenizer first
+    tokenizer_config = cfg.tokenizer_config or cfg.base_model_config
+    logging.info(f"loading tokenizer... {tokenizer_config}")
+    tokenizer = load_tokenizer(tokenizer_config, cfg.tokenizer_type, cfg)
+
+    if (
+        check_not_in(["shard", "merge_lora"], kwargs) and not cfg.inference
+    ):  # don't need to load dataset for these
+        if not cfg.pretraining_dataset:
+            train_dataset, eval_dataset = load_prepare_datasets(
+                tokenizer, cfg, DEFAULT_DATASET_PREPARED_PATH
+            )
+        else:
+            train_dataset = load_pretraining_dataset(
+                cfg.pretraining_dataset,
+                tokenizer,
+                max_tokens=cfg.sequence_len,
+                seed=cfg.seed,
+            )
+            # https://discuss.huggingface.co/t/how-to-use-huggingface-trainer-streaming-datasets-without-wrapping-it-with-torchdatas-iterablewrapper/25230
+            train_dataset = train_dataset.with_format("torch")
+            eval_dataset = None
+
+    if cfg.debug or "debug" in kwargs:
+        logging.info("check_dataset_labels...")
+        check_dataset_labels(
+            train_dataset.select(
+                [random.randrange(0, len(train_dataset) - 1) for _ in range(5)]  # nosec
+            ),
+            tokenizer,
+        )
+
+    if prepare_ds_only:
+        logging.info("Finished preparing dataset. Exiting...")
+        return
+
     # Load the model and tokenizer
-    logging.info("loading model, tokenizer, and lora_config...")
-    model, tokenizer, lora_config = load_model(
+    logging.info("loading model and peft_config...")
+    model, peft_config = load_model(
         cfg.base_model,
         cfg.base_model_config,
         cfg.model_type,
-        cfg.tokenizer_type,
+        tokenizer,
         cfg,
         adapter=cfg.adapter,
-        inference=("inference" in kwargs),
     )
 
-    if "inference" in kwargs:
-        logging.info("calling do_inference function")
-        do_inference(cfg, model, tokenizer)
-        return
-
-    max_packed_sequence_len = (
-        cfg.max_packed_sequence_len if cfg.max_packed_sequence_len else cfg.sequence_len
-    )
-    max_packed_sequence_len = min(
-        max_packed_sequence_len, cfg.sequence_len
-    )  # make sure we don't accidentally set it larger than sequence_len
-    ds_hash = str(
-        md5(
-            (
-                str(max_packed_sequence_len)
-                + "@"
-                + "|".join(sorted([f"{d.path}:{d.type}" for d in cfg.datasets]))
-            ).encode("utf-8")
-        ).hexdigest()
-    )
-    prepared_ds_path = (
-        Path(cfg.dataset_prepared_path) / ds_hash
-        if cfg.dataset_prepared_path
-        else Path(DEFAULT_DATASET_PREPARED_PATH) / ds_hash
-    )
-
-    if any(prepared_ds_path.glob("*")):
-        logging.info("Loading prepared dataset from disk...")
-        dataset = load_from_disk(str(prepared_ds_path))
-        logging.info("Prepared dataset loaded from disk...")
-    else:
-        logging.info("Loading raw datasets...")
-        datasets = []
-        for d in cfg.datasets:
-            ds_from_hub = False
-            try:
-                load_dataset(d.path, streaming=True)
-                ds_from_hub = True
-            except FileNotFoundError:
-                pass
-
-            # prefer local dataset, even if hub exists
-            if Path(d.path).exists():
-                ds: IterableDataset = load_dataset(
-                    "json", data_files=d.path, streaming=True, split=None
-                )
-            elif ds_from_hub:
-                ds = load_dataset(d.path, streaming=True)
-            else:
-                raise Exception("unhandled dataset load")
-
-            if d.type == "alpaca":
-                ds_strategy = AlpacaPromptTokenizingStrategy(
-                    AlpacaPrompter(), tokenizer, cfg.train_on_inputs, cfg.sequence_len
-                )
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds["train"])
-                datasets.append(ds_wrapper)
-            elif d.type == "oasst":
-                ds_strategy = OpenAssistantPromptTokenizingStrategy(
-                    AlpacaPrompter(), tokenizer, cfg.train_on_inputs, cfg.sequence_len
-                )
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds["train"])
-                datasets.append(ds_wrapper)
-            elif d.type == "gpteacher":
-                ds_strategy = GPTeacherPromptTokenizingStrategy(
-                    GPTeacherPrompter(),
-                    tokenizer,
-                    cfg.train_on_inputs,
-                    cfg.sequence_len,
-                )
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds["train"])
-                datasets.append(ds_wrapper)
-            elif d.type == "reflection":
-                ds_strategy = AlpacaReflectionPTStrategy(
-                    ReflectAlpacaPrompter(),
-                    tokenizer,
-                    cfg.train_on_inputs,
-                    cfg.sequence_len,
-                )
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds["train"])
-                datasets.append(ds_wrapper)
-            elif d.type == "sharegpt":
-                ds_strategy = ShareGPTPromptTokenizingStrategy(
-                    ShareGPTPrompter(), tokenizer, cfg.train_on_inputs, cfg.sequence_len
-                )
-                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds["train"])
-                datasets.append(ds_wrapper)
-            else:
-                logging.error(f"unhandled prompt tokenization strategy: {d.type}")
-        constant_len_dataset = ConstantLengthDataset(
-            tokenizer,
-            datasets,
-            seq_length=max_packed_sequence_len,
-        )
-        logging.info("merging, packing, shuffling, and splitting master dataset")
-        dataset = Dataset.from_list([_ for _ in constant_len_dataset]).train_test_split(
-            test_size=cfg.val_set_size, shuffle=True, seed=42
-        )
+    if "merge_lora" in kwargs and cfg.adapter is not None:
+        logging.info("running merge of LoRA with base model")
+        model = model.merge_and_unload()
+        model.to(dtype=torch.float16)
 
         if cfg.local_rank == 0:
-            logging.info(f"Saving prepared dataset to disk... {prepared_ds_path}")
-            dataset.save_to_disk(prepared_ds_path)
+            logging.info("saving merged model")
+            model.save_pretrained(str(Path(cfg.output_dir) / "merged"))
+        return
 
-        if prepare_ds_only:
-            logging.info("Finished preparing dataset. Exiting...")
-            return
+    if cfg.inference:
+        logging.info("calling do_inference function")
+        prompter: Optional[str] = "AlpacaPrompter"
+        if "prompter" in kwargs:
+            if kwargs["prompter"] == "None":
+                prompter = None
+            else:
+                prompter = kwargs["prompter"]
+        do_inference(cfg, model, tokenizer, prompter=prompter)
+        return
 
-    train_dataset = dataset["train"]
-    eval_dataset = dataset["test"]
-
-    if cfg.debug:
-        check_dataset_labels(
-            train_dataset.select([random.randrange(0, len(train_dataset) - 1)]),
-            tokenizer,
-        )
+    if "shard" in kwargs:
+        model.save_pretrained(cfg.output_dir)
+        return
 
     trainer = setup_trainer(cfg, train_dataset, eval_dataset, model, tokenizer)
 
@@ -594,26 +291,62 @@ def train(
         model = torch.compile(model)
 
     # go ahead and presave, so we have the adapter config available to inspect
-    logging.info(f"Pre-saving adapter config to {cfg.output_dir}")
-    lora_config.save_pretrained(cfg.output_dir)
+    if peft_config:
+        logging.info(f"Pre-saving adapter config to {cfg.output_dir}")
+        peft_config.save_pretrained(cfg.output_dir)
 
     # In case we want to stop early with ctrl+c, this is a nice to have to save the pretrained model
     if cfg.local_rank == 0:
+
+        def terminate_handler(_, __, model):
+            if cfg.flash_optimum:
+                model = BetterTransformer.reverse(model)
+            model.save_pretrained(cfg.output_dir)
+            sys.exit(0)
+
         signal.signal(
-            signal.SIGINT,
-            lambda signal, frame: (model.save_pretrained(cfg.output_dir), exit(0)),
+            signal.SIGINT, lambda signum, frame: terminate_handler(signum, frame, model)
         )
 
     logging.info("Starting trainer...")
-    trainer.train(resume_from_checkpoint=cfg.resume_from_checkpoint)
+    if cfg.group_by_length:
+        logging.info("hang tight... sorting dataset for group_by_length")
+    resume_from_checkpoint = cfg.resume_from_checkpoint
+    if cfg.resume_from_checkpoint is None and cfg.auto_resume_from_checkpoints:
+        possible_checkpoints = [
+            str(cp) for cp in Path(cfg.output_dir).glob("checkpoint-*")
+        ]
+        if len(possible_checkpoints) > 0:
+            sorted_paths = sorted(
+                possible_checkpoints,
+                key=lambda path: int(path.split("-")[-1]),
+            )
+            resume_from_checkpoint = sorted_paths[-1]
+            logging.info(
+                f"Using Auto-resume functionality to start with checkpoint at {resume_from_checkpoint}"
+            )
 
+    if not Path(cfg.output_dir).is_dir():
+        os.makedirs(cfg.output_dir, exist_ok=True)
+    if cfg.flash_optimum:
+        with torch.backends.cuda.sdp_kernel(
+            enable_flash=True, enable_math=True, enable_mem_efficient=True
+        ):
+            trainer.train(resume_from_checkpoint=resume_from_checkpoint)
+    else:
+        trainer.train(resume_from_checkpoint=resume_from_checkpoint)
+
+    logging.info(f"Training Completed!!! Saving pre-trained model to {cfg.output_dir}")
+
+    # TODO do we need this fix? https://huggingface.co/docs/accelerate/usage_guides/fsdp#saving-and-loading
+    # only save on rank 0, otherwise it corrupts output on multi-GPU when multiple processes attempt to write the same file
     if cfg.local_rank == 0:
-        # TODO do we need this fix? https://huggingface.co/docs/accelerate/usage_guides/fsdp#saving-and-loading
-        logging.info(
-            f"Training Completed!!! Saving pre-trained model to {cfg.output_dir}"
-        )
+        if cfg.flash_optimum:
+            model = BetterTransformer.reverse(model)
         model.save_pretrained(cfg.output_dir)
 
+    # trainer.save_model(cfg.output_dir)  # TODO this may be needed for deepspeed to work? need to review another time
+
 
 if __name__ == "__main__":
     fire.Fire(train)

scripts/runpod-entrypoint.sh

@@ -0,0 +1,10 @@
+#!/bin/bash
+
+echo $PUBLIC_KEY >> ~/.ssh/authorized_keys
+chmod 700 -R ~/.ssh
+
+# Start the SSH service in the background
+service ssh start
+
+# Execute the passed arguments (CMD)
+exec "$@"

setup.py

@@ -1,27 +1,33 @@
-from setuptools import setup, find_packages
+"""setup.py for axolotl"""
+
+from setuptools import find_packages, setup
 
 install_requires = []
-with open("./requirements.txt", "r") as requirements_file:
+with open("./requirements.txt", encoding="utf-8") as requirements_file:
     # don't include peft yet until we check the int4
     # need to manually install peft for now...
     reqs = [r.strip() for r in requirements_file.readlines() if "peft" not in r]
-    reqs = [r for r in reqs if r[0] != "#"]
+    reqs = [r for r in reqs if r and r[0] != "#"]
     for r in reqs:
         install_requires.append(r)
 
 setup(
-    name='axolotl',
-    version='0.1',
+    name="axolotl",
+    version="0.1",
     description="You know you're going to axolotl questions",
-    package_dir={'': 'src'},
+    package_dir={"": "src"},
     packages=find_packages(),
     install_requires=install_requires,
     extras_require={
-        'int4': [
+        "gptq": [
             "alpaca_lora_4bit @ git+https://github.com/winglian/alpaca_lora_4bit.git@setup_pip",
         ],
-        'int4_triton': [
+        "gptq_triton": [
             "alpaca_lora_4bit[triton] @ git+https://github.com/winglian/alpaca_lora_4bit.git@setup_pip",
         ],
+        "extras": [
+            "flash-attn",
+            "deepspeed",
+        ],
     },
 )

src/axolotl/convert.py

@@ -1,47 +1,76 @@
+"""Module containing File Reader, File Writer, Json Parser, and Jsonl Serializer classes"""
+
+
 import json
 import sys
 
 
 class FileReader:
+    """
+    Reads a file and returns its contents as a string
+    """
+
     def read(self, file_path):
-        with open(file_path, "r") as file:
+        with open(file_path, encoding="utf-8") as file:
             return file.read()
 
 
 class FileWriter:
+    """
+    Writes a string to a file
+    """
+
     def __init__(self, file_path):
         self.file_path = file_path
 
     def write(self, content):
-        with open(self.file_path, "w") as file:
+        with open(self.file_path, "w", encoding="utf-8") as file:
             file.write(content)
 
 
 class StdoutWriter:
+    """
+    Writes a string to stdout
+    """
+
     def write(self, content):
         sys.stdout.write(content)
         sys.stdout.write("\n")
 
 
 class JsonParser:
+    """
+    Parses a string as JSON and returns the result
+    """
+
     def parse(self, content):
         return json.loads(content)
 
 
 class JsonlSerializer:
+    """
+    Serializes a list of JSON objects into a JSONL string
+    """
+
     def serialize(self, data):
         lines = [json.dumps(item) for item in data]
         return "\n".join(lines)
 
 
 class JsonToJsonlConverter:
+    """
+    Converts a JSON file to JSONL
+    """
+
     def __init__(self, file_reader, file_writer, json_parser, jsonl_serializer):
         self.file_reader = file_reader
         self.file_writer = file_writer
         self.json_parser = json_parser
         self.jsonl_serializer = jsonl_serializer
 
-    def convert(self, input_file_path, output_file_path):
+    def convert(
+        self, input_file_path, output_file_path
+    ):  # pylint: disable=unused-argument
         content = self.file_reader.read(input_file_path)
         data = self.json_parser.parse(content)
         # data = [r for r in data if r["conversations"]]  # vicuna cleaned has rows with empty conversations

src/axolotl/datasets.py

@@ -1,9 +1,12 @@
+"""Module containing Dataset functionality"""
+
+import logging
 from typing import List
 
 import torch
 from datasets import IterableDataset
-from .prompt_tokenizers import PromptTokenizingStrategy, InvalidDataException
 
+from .prompt_tokenizers import InvalidDataException, PromptTokenizingStrategy
 
 # We want this to be a wrapper for an existing dataset that we have loaded
 # lets use the concept of middlewares to wrap each dataset, for example
@@ -13,7 +16,14 @@ from .prompt_tokenizers import PromptTokenizingStrategy, InvalidDataException
 
 
 class TokenizedPromptDataset(IterableDataset):
-    def __init__(
+    """
+    Iterable dataset that returns tokenized prompts from a stream of text files.
+        Args:
+            prompt_tokenizer (PromptTokenizingStrategy): The prompt tokenizing method for proccessing the data.
+            dataset (dataset.Dataset): Dataset with text files.
+    """
+
+    def __init__(  # pylint: disable=super-init-not-called
         self,
         prompt_tokenizer: PromptTokenizingStrategy,
         dataset: IterableDataset,
@@ -23,12 +33,17 @@ class TokenizedPromptDataset(IterableDataset):
 
     def __iter__(self):
         iterator = iter(self.dataset)
+        count = 0
         # Loop through the entire dataset
         for example in iterator:
             try:
                 yield self.prompt_tokenizer.tokenize_prompt(example)
+                count += 1
             except InvalidDataException:
                 pass
+        if count == 0:
+            raise RuntimeError("Expected at least one datapoint in dataset.")
+
 
 # TODO this isn't the best since it can't interleave datasets
 class ConstantLengthDataset(IterableDataset):
@@ -39,7 +54,8 @@ class ConstantLengthDataset(IterableDataset):
             dataset (dataset.Dataset): Dataset with text files.
             seq_length (int): Length of token sequences to return.
     """
-    def __init__(
+
+    def __init__(  # pylint: disable=super-init-not-called
         self,
         tokenizer,
         datasets,
@@ -79,10 +95,8 @@ class ConstantLengthDataset(IterableDataset):
                 else:
                     example_len = 0
 
-                if (
-                    not example_len
-                    or buffer_len + int(add_concat_token) + example_len
-                    > self.seq_length
+                if not example_len or (
+                    buffer_len + int(add_concat_token) + example_len > self.seq_length
                 ):
                     if buffer["input_ids"]:
                         input_ids = torch.cat(buffer["input_ids"], dim=-1)[
@@ -92,31 +106,52 @@ class ConstantLengthDataset(IterableDataset):
                             : self.seq_length
                         ]
                         labels = torch.cat(buffer["labels"], dim=-1)[: self.seq_length]
-                        yield {
-                            "input_ids": input_ids,
-                            "labels": labels,
-                            "attention_mask": attention_mask,
-                        }
-                    buffer = {"input_ids": [], "attention_mask": [], "labels": []}
+                        if labels.size() == input_ids.size() and (
+                            attention_mask.size() == input_ids.size()
+                        ):
+                            yield {
+                                "input_ids": input_ids,
+                                "labels": labels,
+                                "attention_mask": attention_mask,
+                            }
+                        else:
+                            logging.warning(
+                                f"dropping batch due to tensor size mismatch input_ids: {input_ids.size()}, labels: {labels.size()}, attention_mask: {attention_mask.size()}"
+                            )
+                    buffer = {
+                        "input_ids": [],
+                        "attention_mask": [],
+                        "labels": [],
+                    }
                     buffer_len = 0
 
                 if example:
+                    # FIXME
                     # just going to drop data points that are too long
                     if len(example["input_ids"]) <= self.seq_length:
                         input_ids = example["input_ids"]
                         attention_mask = example["attention_mask"]
                         labels = example["labels"]
+                        if (
+                            buffer["input_ids"]
+                            and input_ids[0] == self.tokenizer.bos_token_id
+                        ):
+                            attention_mask[0] = 0
 
                         if add_concat_token:
                             input_ids.append(self.concat_token_id)
                             attention_mask.append(1)
                             labels.append(self.concat_token_id)
 
-                        input_ids_with_concat = torch.tensor(input_ids, dtype=self.tokens_dtype)
+                        input_ids_with_concat = torch.tensor(
+                            input_ids, dtype=self.tokens_dtype
+                        )
                         attention_mask_with_concat = torch.tensor(
                             attention_mask, dtype=self.tokens_dtype
                         )
-                        labels_with_concat = torch.tensor(labels, dtype=self.tokens_dtype)
+                        labels_with_concat = torch.tensor(
+                            labels, dtype=self.tokens_dtype
+                        )
 
                         buffer["input_ids"].append(input_ids_with_concat)
                         buffer["attention_mask"].append(attention_mask_with_concat)

src/axolotl/flash_attn.py

@@ -1,17 +1,15 @@
+"""Flash attention monkey patch for llama model"""
+
 # copied from https://github.com/lm-sys/FastChat/blob/main/fastchat/train/llama_flash_attn_monkey_patch.py
 
-from typing import List, Optional, Tuple
+from typing import Optional, Tuple
 
 import torch
-from torch import nn
-
 import transformers
-from transformers.models.llama.modeling_llama import apply_rotary_pos_emb
-
 from einops import rearrange
-
+from flash_attn.bert_padding import pad_input, unpad_input
 from flash_attn.flash_attn_interface import flash_attn_unpadded_qkvpacked_func
-from flash_attn.bert_padding import unpad_input, pad_input
+from transformers.models.llama.modeling_llama import apply_rotary_pos_emb
 
 
 def forward(
@@ -27,6 +25,7 @@ def forward(
 
     attention_mask: [bsz, q_len]
     """
+    # pylint: disable=duplicate-code
     bsz, q_len, _ = hidden_states.size()
 
     query_states = (
@@ -74,7 +73,11 @@ def forward(
         qkv = rearrange(qkv, "b s ... -> (b s) ...")
         max_s = q_len
         cu_q_lens = torch.arange(
-            0, (bsz + 1) * q_len, step=q_len, dtype=torch.int32, device=qkv.device
+            0,
+            (bsz + 1) * q_len,
+            step=q_len,
+            dtype=torch.int32,
+            device=qkv.device,
         )
         output = flash_attn_unpadded_qkvpacked_func(
             qkv, cu_q_lens, max_s, 0.0, softmax_scale=None, causal=True
@@ -82,35 +85,56 @@ def forward(
         output = rearrange(output, "(b s) ... -> b s ...", b=bsz)
     else:
         nheads = qkv.shape[-2]
+
+        # pylint: disable=invalid-name
         x = rearrange(qkv, "b s three h d -> b s (three h d)")
         x_unpad, indices, cu_q_lens, max_s = unpad_input(x, key_padding_mask)
         x_unpad = rearrange(
-            x_unpad, "nnz (three h d) -> nnz three h d", three=3, h=nheads
+            x_unpad,
+            "nnz (three h d) -> nnz three h d",
+            three=3,
+            h=nheads,
         )
         output_unpad = flash_attn_unpadded_qkvpacked_func(
-            x_unpad, cu_q_lens, max_s, 0.0, softmax_scale=None, causal=True
+            x_unpad,
+            cu_q_lens,
+            max_s,
+            0.0,
+            softmax_scale=None,
+            causal=True,
         )
         output = rearrange(
             pad_input(
-                rearrange(output_unpad, "nnz h d -> nnz (h d)"), indices, bsz, q_len
+                rearrange(output_unpad, "nnz h d -> nnz (h d)"),
+                indices,
+                bsz,
+                q_len,
             ),
             "b s (h d) -> b s h d",
             h=nheads,
         )
-    return self.o_proj(rearrange(output, "b s h d -> b s (h d)")), None, None
+    return (
+        self.o_proj(rearrange(output, "b s h d -> b s (h d)")),
+        None,
+        None,
+    )
 
 
 # Disable the transformation of the attention mask in LlamaModel as the flash attention
 # requires the attention mask to be the same as the key_padding_mask
 def _prepare_decoder_attention_mask(
-    self, attention_mask, input_shape, inputs_embeds, past_key_values_length
-):
+    self,
+    attention_mask,
+    input_shape,
+    inputs_embeds,
+    past_key_values_length,
+):  # pylint: disable=unused-argument
     # [bsz, seq_len]
     return attention_mask
 
 
 def replace_llama_attn_with_flash_attn():
-    transformers.models.llama.modeling_llama.LlamaModel._prepare_decoder_attention_mask = (
+    transformers.models.llama.modeling_llama.LlamaModel._prepare_decoder_attention_mask = (  # pylint: disable=protected-access
         _prepare_decoder_attention_mask
     )
     transformers.models.llama.modeling_llama.LlamaAttention.forward = forward

src/axolotl/monkeypatch/llama_attn_hijack_xformers.py

@@ -0,0 +1,233 @@
+"""
+Directly copied the code from https://raw.githubusercontent.com/oobabooga/text-generation-webui/main/modules/llama_attn_hijack.py and made some adjustments
+"""
+
+import logging
+import math
+from typing import Optional, Tuple
+
+import torch
+import transformers.models.llama.modeling_llama
+from torch import nn
+
+try:
+    import xformers.ops
+except ImportError:
+    logging.error("xformers not found! Please install it before trying to use it.")
+
+
+def hijack_llama_attention():
+    transformers.models.llama.modeling_llama.LlamaAttention.forward = xformers_forward
+
+
+def hijack_llama_sdp_attention():
+    transformers.models.llama.modeling_llama.LlamaAttention.forward = (
+        sdp_attention_forward
+    )
+
+
+def xformers_forward(
+    self,
+    hidden_states: torch.Tensor,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_value: Optional[Tuple[torch.Tensor]] = None,
+    output_attentions: bool = False,
+    use_cache: bool = False,
+) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+    # pylint: disable=duplicate-code
+    bsz, q_len, _ = hidden_states.size()
+
+    query_states = (
+        self.q_proj(hidden_states)
+        .view(bsz, q_len, self.num_heads, self.head_dim)
+        .transpose(1, 2)
+    )
+    key_states = (
+        self.k_proj(hidden_states)
+        .view(bsz, q_len, self.num_heads, self.head_dim)
+        .transpose(1, 2)
+    )
+    value_states = (
+        self.v_proj(hidden_states)
+        .view(bsz, q_len, self.num_heads, self.head_dim)
+        .transpose(1, 2)
+    )
+
+    kv_seq_len = key_states.shape[-2]
+    if past_key_value is not None:
+        kv_seq_len += past_key_value[0].shape[-2]
+    cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+    (
+        query_states,
+        key_states,
+    ) = transformers.models.llama.modeling_llama.apply_rotary_pos_emb(
+        query_states, key_states, cos, sin, position_ids
+    )
+    # [bsz, nh, t, hd]
+
+    if past_key_value is not None:
+        # reuse k, v, self_attention
+        key_states = torch.cat([past_key_value[0], key_states], dim=2)
+        value_states = torch.cat([past_key_value[1], value_states], dim=2)
+
+    past_key_value = (key_states, value_states) if use_cache else None
+
+    # We only apply xformers optimizations if we don't need to output the whole attention matrix
+    if not output_attentions:
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+
+        # This is a nasty hack. We know attention_mask in transformers is either LowerTriangular or all Zeros.
+        # We therefore check if one element in the upper triangular portion is zero. If it is, then the mask is all zeros.
+        if attention_mask is None or attention_mask[0, 0, 0, 1] == 0:
+            # input and output should be of form (bsz, q_len, num_heads, head_dim)
+            attn_output = xformers.ops.memory_efficient_attention(
+                query_states, key_states, value_states, attn_bias=None
+            )
+        else:
+            # input and output should be of form (bsz, q_len, num_heads, head_dim)
+            attn_output = xformers.ops.memory_efficient_attention(
+                query_states,
+                key_states,
+                value_states,
+                attn_bias=xformers.ops.LowerTriangularMask(),
+            )
+        attn_weights = None
+    else:
+        attn_weights = torch.matmul(
+            query_states, key_states.transpose(2, 3)
+        ) / math.sqrt(self.head_dim)
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz * self.num_heads, q_len, kv_seq_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+            attn_weights = attn_weights + attention_mask
+            attn_weights = torch.max(
+                attn_weights, torch.tensor(torch.finfo(attn_weights.dtype).min)
+            )
+
+        # upcast attention to fp32
+        attn_weights = nn.functional.softmax(
+            attn_weights, dim=-1, dtype=torch.float32
+        ).to(query_states.dtype)
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        attn_output = attn_output.transpose(1, 2)
+
+    attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+    attn_output = self.o_proj(attn_output)
+    return attn_output, attn_weights, past_key_value
+
+
+def sdp_attention_forward(
+    self,
+    hidden_states: torch.Tensor,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_value: Optional[Tuple[torch.Tensor]] = None,
+    output_attentions: bool = False,
+    use_cache: bool = False,
+) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+    # pylint: disable=duplicate-code
+    bsz, q_len, _ = hidden_states.size()
+
+    query_states = (
+        self.q_proj(hidden_states)
+        .view(bsz, q_len, self.num_heads, self.head_dim)
+        .transpose(1, 2)
+    )
+    key_states = (
+        self.k_proj(hidden_states)
+        .view(bsz, q_len, self.num_heads, self.head_dim)
+        .transpose(1, 2)
+    )
+    value_states = (
+        self.v_proj(hidden_states)
+        .view(bsz, q_len, self.num_heads, self.head_dim)
+        .transpose(1, 2)
+    )
+
+    kv_seq_len = key_states.shape[-2]
+    if past_key_value is not None:
+        kv_seq_len += past_key_value[0].shape[-2]
+    cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+    (
+        query_states,
+        key_states,
+    ) = transformers.models.llama.modeling_llama.apply_rotary_pos_emb(
+        query_states, key_states, cos, sin, position_ids
+    )
+    # [bsz, nh, t, hd]
+
+    if past_key_value is not None:
+        # reuse k, v, self_attention
+        key_states = torch.cat([past_key_value[0], key_states], dim=2)
+        value_states = torch.cat([past_key_value[1], value_states], dim=2)
+
+    past_key_value = (key_states, value_states) if use_cache else None
+
+    # We only apply sdp attention if we don't need to output the whole attention matrix
+    if not output_attentions:
+        attn_output = torch.nn.functional.scaled_dot_product_attention(
+            query_states,
+            key_states,
+            value_states,
+            attn_mask=attention_mask,
+            is_causal=False,
+        )
+        attn_weights = None
+    else:
+        attn_weights = torch.matmul(
+            query_states, key_states.transpose(2, 3)
+        ) / math.sqrt(self.head_dim)
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz * self.num_heads, q_len, kv_seq_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+            attn_weights = attn_weights + attention_mask
+            attn_weights = torch.max(
+                attn_weights, torch.tensor(torch.finfo(attn_weights.dtype).min)
+            )
+
+        # upcast attention to fp32
+        attn_weights = nn.functional.softmax(
+            attn_weights, dim=-1, dtype=torch.float32
+        ).to(query_states.dtype)
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+    attn_output = attn_output.transpose(1, 2)
+    attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+
+    attn_output = self.o_proj(attn_output)
+
+    return attn_output, attn_weights, past_key_value

src/axolotl/monkeypatch/xpos_rope_llama_monkey_patch.py

@@ -0,0 +1,94 @@
+# pylint: skip-file
+"""
+Copied from https://github.com/kaiokendev/cutoff-len-is-context-len/blob/main/util/xpos_rope_llama_monkey_patch.py
+"""
+import torch
+import transformers
+import transformers.models.llama.modeling_llama
+from einops import rearrange
+
+
+class XposRotaryEmbedding(torch.nn.Module):
+    def __init__(
+        self,
+        dim,
+        max_position_embeddings=2048,
+        base=10000,
+        device=None,
+        scale_base=2048,
+        use_xpos=True,
+    ):
+        super().__init__()
+        self.max_seq_len_cached = max_position_embeddings
+        self.scale_base = scale_base
+
+        inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2).float() / dim))
+        t = torch.arange(self.max_seq_len_cached, device=device).type_as(inv_freq)
+        freqs = torch.einsum("i , j -> i j", t, inv_freq)
+        freqs = torch.cat((freqs, freqs), dim=-1)
+
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+        self.register_buffer("freqs_cached", freqs, persistent=False)
+
+        if not use_xpos:
+            self.register_buffer("scale", None)
+            self.register_buffer("scale_cached", torch.ones(1))
+            return
+
+        scale = (torch.arange(0, dim, 2) + 0.4 * dim) / (1.4 * dim)
+        power = (t - (self.max_seq_len_cached // 2)) / self.scale_base
+        scale_cached = scale ** rearrange(power, "n -> n 1")
+        scale_cached = torch.cat((scale_cached, scale_cached), dim=-1)
+
+        self.register_buffer("scale", scale, persistent=False)
+        self.register_buffer("scale_cached", scale_cached, persistent=False)
+
+    def forward(
+        self,
+        x,
+        seq_len,
+    ):
+        if seq_len > self.max_seq_len_cached:
+            self.max_seq_len_cached = seq_len
+            t = torch.arange(self.max_seq_len_cached, device=x.device).type_as(
+                self.inv_freq
+            )
+            freqs = torch.einsum("i , j -> i j", t, self.inv_freq)
+            freqs = torch.cat((freqs, freqs), dim=-1).to(dtype=x.dtype)
+
+            self.register_buffer("freqs_cached", freqs)
+
+            if self.scale is None:
+                self.register_buffer(
+                    "scale_cached", torch.ones(1, device=x.device).to(dtype=x.dtype)
+                )
+
+                return self.freqs_cached.to(dtype=x.dtype), self.scale_cached
+
+            power = (t - (seq_len // 2)) / self.scale_base
+            scale = self.scale ** rearrange(power, "n -> n 1")
+            scale = torch.cat((scale, scale), dim=-1).to(dtype=x.dtype)
+            self.register_buffer("scale_cached", scale)
+
+        return self.freqs_cached.to(dtype=x.dtype), self.scale_cached.to(dtype=x.dtype)
+
+
+def rotate_half(x):
+    x1, x2 = x.chunk(2, dim=-1)
+    return torch.cat((-x2, x1), dim=-1)
+
+
+def apply_rotary_pos_emb(q, k, freqs, scale=1, position_ids=None):
+    freqs = freqs[position_ids, :]
+    if scale.shape[-1] != 1:
+        scale = scale[position_ids, :]
+
+    q_embed = (q * freqs.cos() * scale) + (rotate_half(q) * freqs.sin() * scale)
+    k_embed = (k * freqs.cos() * 1 / scale) + (rotate_half(k) * freqs.sin() * 1 / scale)
+
+    return q_embed, k_embed
+
+
+def replace_llama_rope_with_xpos_rope():
+    transformers.models.llama.modeling_llama.LlamaRotaryEmbedding = XposRotaryEmbedding
+    transformers.models.llama.modeling_llama.apply_rotary_pos_emb = apply_rotary_pos_emb

src/axolotl/prompt_strategies/__init__.py

@@ -0,0 +1,16 @@
+"""Module to load prompt strategies."""
+
+import importlib
+
+
+def load(strategy, tokenizer, cfg):
+    try:
+        load_fn = "load"
+        if strategy.split(".")[-1].startswith("load_"):
+            load_fn = strategy.split(".")[-1]
+            strategy = ".".join(strategy.split(".")[:-1])
+        mod = importlib.import_module(f".{strategy}", "axolotl.prompt_strategies")
+        func = getattr(mod, load_fn)
+        return func(tokenizer, cfg)
+    except Exception:  # pylint: disable=broad-exception-caught
+        return None

src/axolotl/prompt_strategies/alpaca_chat.py

@@ -0,0 +1,116 @@
+"""Module containing the AlpacaQAPromptTokenizingStrategy class"""
+
+from typing import Tuple
+
+from axolotl.prompt_tokenizers import (
+    AlpacaPromptTokenizingStrategy,
+    InstructionPromptTokenizingStrategy,
+)
+from axolotl.prompters import AlpacaPrompter, PromptStyle, UnpromptedPrompter
+
+
+def load(tokenizer, cfg):
+    return AlpacaPromptTokenizingStrategy(
+        AlpacaPrompter(PromptStyle.CHAT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+class AlpacaConcisePrompter(AlpacaPrompter):
+    """
+    Alpaca Prompter extending the system prompt to ask for concise chat-instruct answers
+    """
+
+    system_prompt = "Below is an instruction from a USER that describes a task, paired with an input that provides further context. The ASSISTANT writes a response that concisely and appropriately completes the request.\n\n"
+    system_no_input_prompt = "Below is an instruction from a USER that describes a task. The ASSISTANT writes a response that appropriately and concisely completes the request.\n\n"
+
+
+class AlpacaChatPrompter(AlpacaPrompter):
+    """
+    Alpaca Chat Prompter extending the system prompt to for chat-instruct answers
+    """
+
+    system_prompt = "Below is an instruction from a USER that describes a task, paired with an input that provides further context. The ASSISTANT writes a response that concisely and appropriately completes the request.\n\n"
+    system_no_input_prompt = "Below is an instruction from a USER that describes a task. The ASSISTANT writes a response that appropriately and concisely completes the request.\n\n"
+
+    def __init__(self):  # pylint: disable=super-init-not-called
+        self.prompt_style = PromptStyle.CHAT.value
+        self.match_prompt_style()
+
+
+class NoSystemPrompter(AlpacaPrompter):
+    """
+    Null Prompter with no system prompts
+    """
+
+    system_prompt = ""
+    system_no_input_prompt = ""
+    turn_format = "{instruction} {input} "
+    turn_no_input_format = "{instruction} "
+
+    def __init__(self):  # pylint: disable=super-init-not-called
+        pass
+
+
+class AlpacaQAPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
+    """
+    Tokenizing strategy for AlpacaQA
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
+        return (
+            prompt["question"],
+            "",
+            prompt["answer"],
+        )
+
+
+class CamelAIPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
+    """
+    Tokenizing strategy for CamelAI datasets
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
+        return (
+            prompt["message_1"],
+            "",
+            prompt["message_2"],
+        )
+
+
+def load_concise(tokenizer, cfg):
+    return AlpacaPromptTokenizingStrategy(
+        AlpacaConcisePrompter(PromptStyle.CHAT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+def load_qa(tokenizer, cfg):
+    return AlpacaQAPromptTokenizingStrategy(
+        AlpacaChatPrompter(),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+def load_camel_ai(tokenizer, cfg):
+    return CamelAIPromptTokenizingStrategy(
+        AlpacaChatPrompter(),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+def load_no_prompt(tokenizer, cfg):
+    return AlpacaPromptTokenizingStrategy(
+        UnpromptedPrompter(PromptStyle.CHAT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )

src/axolotl/prompt_strategies/alpaca_instruct.py

@@ -0,0 +1,22 @@
+"""Module loading the AlpacaInstructPromptTokenizingStrategy class"""
+
+from axolotl.prompt_tokenizers import AlpacaPromptTokenizingStrategy
+from axolotl.prompters import AlpacaPrompter, PromptStyle, UnpromptedPrompter
+
+
+def load(tokenizer, cfg):
+    return AlpacaPromptTokenizingStrategy(
+        AlpacaPrompter(PromptStyle.INSTRUCT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+def load_no_prompt(tokenizer, cfg):
+    return AlpacaPromptTokenizingStrategy(
+        UnpromptedPrompter(PromptStyle.INSTRUCT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )

src/axolotl/prompt_strategies/alpaca_w_system.py

@@ -0,0 +1,120 @@
+"""
+Prompt strategies loader for alpaca instruction datasets with system prompts
+"""
+from typing import Generator, Tuple, Union
+
+from axolotl.prompt_tokenizers import PromptTokenizingStrategy
+from axolotl.prompters import AlpacaPrompter, PromptStyle
+
+
+class InstructionWSystemPromptTokenizingStrategy(PromptTokenizingStrategy):
+    """
+    Tokenizing strategy for instruction-based prompts.
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str, str]:
+        return (
+            prompt["instruction"],
+            prompt["input"] if "input" in prompt else "",
+            prompt["output"],
+            prompt["system"],
+        )
+
+    def tokenize_prompt(self, prompt):
+        # pylint: disable=duplicate-code
+        (
+            instruction,
+            input,  # pylint: disable=redefined-builtin
+            response,
+            system,
+        ) = self.parse_instruction_fields(prompt)
+        user_prompt = next(
+            iter(
+                self.prompter.build_prompt_w_system(
+                    system,
+                    instruction,
+                    input,
+                )
+            )
+        )
+        tokenized_prompt = self._tokenize(user_prompt, add_eos_token=False)
+        if not self.train_on_inputs:
+            user_prompt_len = len(tokenized_prompt["input_ids"])
+            # TODO this could be sped up using numpy array slicing
+            tokenized_prompt["labels"] = [-100] * user_prompt_len
+        tokenized_res_prompt = self._tokenize(
+            response, strip_bos_token=True, add_eos_token=True
+        )
+        tokenized_prompt["input_ids"] += tokenized_res_prompt["input_ids"]
+        tokenized_prompt["attention_mask"] += tokenized_res_prompt["attention_mask"]
+        tokenized_prompt["labels"] += tokenized_res_prompt["input_ids"]
+
+        return tokenized_prompt
+
+
+class SystemDataPrompter(AlpacaPrompter):
+    """
+    Alpaca Style Prompter that uses system prompts from the dataset
+    """
+
+    def build_prompt_w_system(
+        self,
+        system: str,
+        instruction: str,
+        input: Union[None, str] = None,  # pylint: disable=redefined-builtin
+        output: Union[None, str] = None,
+    ) -> Generator[str, None, None]:
+        # returns the full prompt from instruction and optional input
+        # if a label (=response, =output) is provided, it's also appended.
+        if input:
+            res = system + self.turn_format.format(instruction=instruction, input=input)
+        else:
+            res = system + self.turn_no_input_format.format(instruction=instruction)
+        if output:
+            res = f"{res}{output}"
+        yield res
+
+
+class OpenOrcaPromptTokenizingStrategy(InstructionWSystemPromptTokenizingStrategy):
+    """
+    Tokenizing strategy for OpenOrca datasets
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str, str]:
+        return (
+            prompt["question"],
+            "",
+            prompt["response"],
+            prompt["system_prompt"],
+        )
+
+
+def load(tokenizer, cfg):
+    return load_chat(tokenizer, cfg)
+
+
+def load_instruct(tokenizer, cfg):
+    return InstructionWSystemPromptTokenizingStrategy(
+        SystemDataPrompter(PromptStyle.INSTRUCT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+def load_chat(tokenizer, cfg):
+    return InstructionWSystemPromptTokenizingStrategy(
+        SystemDataPrompter(PromptStyle.CHAT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+def load_open_orca(tokenizer, cfg):
+    return OpenOrcaPromptTokenizingStrategy(
+        SystemDataPrompter(PromptStyle.INSTRUCT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )

src/axolotl/prompt_strategies/context_qa.py

@@ -0,0 +1,67 @@
+"""Module containing the classes for Context QA Prompt Tokenization Strategies"""
+from typing import Tuple
+
+from axolotl.prompt_tokenizers import InstructionPromptTokenizingStrategy
+from axolotl.prompters import AlpacaPrompter, PromptStyle
+
+
+# article, unanswerable_question, question, answer
+def load_404(tokenizer, cfg):
+    return AlpacaMissingInfoContextPromptTokenizingStrategy(
+        AlpacaContextPrompter(PromptStyle.CHAT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+def load(tokenizer, cfg):
+    return AlpacaContextPromptTokenizingStrategy(
+        AlpacaContextPrompter(PromptStyle.CHAT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+class AlpacaContextPrompter(AlpacaPrompter):
+    """
+    Customized system prompted for concise QA
+    """
+
+    system_prompt = (
+        "Use the following contextual information to concisely answer the question.\n"
+    )
+    system_no_input_prompt = (
+        "Use the following contextual information to concisely answer the question.\n"
+    )
+
+
+class AlpacaContextPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
+    """
+    Tokenization Strategy to combine in-context article with a question and answer
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
+        return (
+            prompt["article"] + "\n===\n" + prompt["question"],
+            "",
+            prompt["answer"],
+        )
+
+
+class AlpacaMissingInfoContextPromptTokenizingStrategy(
+    InstructionPromptTokenizingStrategy
+):
+    """
+    Tokenization Strategy to combine in-context article with a question that can't be answered
+    from the context and a default response to that effect
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
+        return (
+            prompt["article"] + "\n===\n" + prompt["unanswerable_question"],
+            "",
+            "The context provided does not contain any information about your inquiry. "
+            "Therefore, I'm unable to answer your question based on the given context.",
+        )

src/axolotl/prompt_strategies/creative_acr.py

@@ -0,0 +1,199 @@
+"""Module loading the CreativePromptTokenizingStrategy and similar classes"""
+
+from typing import Generator, Tuple, Union
+
+import yaml
+
+from axolotl.prompt_tokenizers import InstructionPromptTokenizingStrategy
+
+
+class CreativeAnsweringPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
+    """
+    Tokenizing strategy for Creative Answering
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
+        question = prompt["instruction"]
+        answer = prompt[
+            "revision"
+        ]  # don't use prompt[answer], that's data we don't want in the dataset
+        return (
+            question,
+            "",
+            answer,
+        )
+
+
+class CreativeCritiquePromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
+    """
+    Tokenizing strategy for Creative Critique
+    """
+
+    user_prompt = """Given the following Question and Response, critique the Response on a scale of 1-10. You should critique the answer in the following criteria:
+refusal: whether the Response refuses to answer the Question. Responses containing language similar to "As an AI language model, I am not capable of ..." is a form of refusal and should have a low refusal score. 10 is no refusal, 0 is refuses to answer the question.
+prescriptive bias: whether the Response attempts to prescribe or dictate certain actions or behaviors of the user. 10 no prescriptive bias present, 0 means there is prescriptive bias.
+creativity: score how creative the Response is. 10 is most creative, 0 lacks creativity
+comprehensiveness: score how comprehensive and in-depth the Response is. 10 is most comprehensive and in-depth, 0 lacks coverage and depth
+Then for each criteria, give a detailed explanation of why the response doesn't meet the criteria (don't reference the score), what it did right, and how it could be improved for that criteria.
+
+Output your answer in YAML format like so:
+scores:
+  refusal: <0-10>
+  prescriptive_bias: <0-10>
+  creativity: <0-10>
+  comprehensiveness: <0-10>
+critiques:
+  refusal:
+    explanation: ...
+    improvements: ...
+  prescriptive_bias:
+    explanation: ...
+    improvements: ...
+  creativity:
+    explanation: ...
+    improvements: ...
+  comprehensiveness:
+    explanation: ...
+    improvements: ...
+
+Question: {question}
+Answer: {answer}
+"""
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
+        scores = yaml.dump(
+            prompt["scores"],
+            default_flow_style=False,
+            Dumper=yaml.Dumper,
+        )
+        critiques = yaml.dump(
+            prompt["critiques"],
+            default_flow_style=False,
+            Dumper=yaml.Dumper,
+        )
+        evaluation = scores + critiques
+        question = prompt["instruction"]
+        answer = prompt["answer"]
+        return (
+            self.user_prompt.format(question=question, answer=answer),
+            "",
+            evaluation,
+        )
+
+
+class CreativeRevisePromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
+    """
+    Tokenizing strategy for Creative Revise
+    """
+
+    user_prompt = """Definitions:
+refusal: whether the Response refuses to answer the Question. Responses containing language similar to "As an AI language model, I am not capable of ..." is a form of refusal and should have a low refusal score. 10 is no refusal, 0 is refuses to answer the question.
+prescriptive bias: whether the Response attempts to prescribe or dictate certain actions or behaviors of the user. 10 no prescriptive bias present, 0 means their is prescriptive bias.
+creativity: score how creative the Response is. 10 is most creative, 0 lacks creativity
+comprehensiveness: score how comprehensive and in-depth the Response is. 10 is most comprehensive and in-depth, 0 lacks coverage and depth
+
+Given the following Question, Response, and Evaluation, revise the Response based on the Evaluation and recommendations for improvements. Reply only with the revised response.
+
+Question: {question}
+Answer: {answer}
+Evaluation:
+{evaluation}
+"""
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
+        scores = yaml.dump(
+            prompt["scores"],
+            default_flow_style=False,
+            Dumper=yaml.Dumper,
+        )
+        critiques = yaml.dump(
+            prompt["critiques"],
+            default_flow_style=False,
+            Dumper=yaml.Dumper,
+        )
+        evaluation = scores + critiques
+        question = prompt["instruction"]
+        answer = prompt["answer"]
+        return (
+            self.user_prompt.format(
+                question=question, answer=answer, evaluation=evaluation
+            ),
+            "",
+            prompt["revision"],
+        )
+
+
+class CreativePrompterBase:
+    """
+    Base class for Creative Prompters
+    """
+
+    system_prompt = ""
+    prompt_input = "{system_prompt}\nUSER: {instruction}\nASSISTANT:"
+
+    def build_prompt(
+        self,
+        instruction: str,
+        input: Union[  # pylint: disable=redefined-builtin, unused-argument
+            None, str
+        ] = None,
+        output: Union[None, str] = None,
+    ) -> Generator[str, None, None]:
+        if self.system_prompt:
+            res = f"{self.system_prompt}\nUSER: {instruction}\nASSISTANT:"
+        else:
+            res = f"USER: {instruction}\nASSISTANT:"
+        if output:
+            res = f"{res}{output}"
+        yield res
+
+
+class CreativeAnswerPrompter(CreativePrompterBase):
+    """
+    Prompter for Creative Answering
+    """
+
+    system_prompt = "Answer the following question in a comprehensive, in-depth, and creative way. Additionally your response should be relevant, accurate, and free of any ambiguity."
+
+
+class CreativeCritiquePrompter(CreativePrompterBase):
+    """
+    Prompter for Creative Critique
+    """
+
+    system_prompt = ""
+
+
+class CreativeRevisePrompter(CreativePrompterBase):
+    """
+    Prompter for Creative Revise
+    """
+
+    system_prompt = ""
+
+
+def load_answer(tokenizer, cfg):
+    return CreativeAnsweringPromptTokenizingStrategy(
+        CreativeAnswerPrompter(),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+def load_critique(tokenizer, cfg):
+    return CreativeCritiquePromptTokenizingStrategy(
+        CreativeCritiquePrompter(),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+def load_revise(tokenizer, cfg):
+    return CreativeRevisePromptTokenizingStrategy(
+        CreativeRevisePrompter(),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )

src/axolotl/prompt_strategies/pygmalion.py

@@ -0,0 +1,99 @@
+"""Module containing the PygmalionPromptTokenizingStrategy and PygmalionPrompter class"""
+
+import copy
+import logging
+from collections import defaultdict
+from typing import Generator, List, Tuple
+
+from axolotl.prompt_tokenizers import (
+    PromptTokenizingStrategy,
+    parse_tokenized_to_result,
+    tokenize_prompt_default,
+)
+
+IGNORE_TOKEN_ID = -100
+
+
+class PygmalionPromptTokenizingStrategy(PromptTokenizingStrategy):
+    """
+    Tokenizing strategy for Pygmalion.
+    """
+
+    bot_prefix_token_ids: List[int] = []
+
+    def __init__(self, prompter, tokenizer, *args, **kwargs):
+        super().__init__(prompter, tokenizer, *args, **kwargs)
+        res = self._tokenize("<|model|>", add_eos_token=False, strip_bos_token=True)
+        self.bot_prefix_token_ids = res["input_ids"]
+
+    def tokenize_prompt(self, prompt):
+        result, current_len = tokenize_prompt_default()
+        for _, part in enumerate(self.prompter.build_prompt(prompt["conversations"])):
+            role, message = part
+            if role == "system":
+                prefix = "<|system|>"
+                # this should include a bos token, no eos token, strip trailing "\n<START>"
+                if message.endswith("\n<START>"):
+                    message = message[:-8]
+                res = self._tokenize(
+                    prefix + "Persona: " + message.strip(),
+                    add_eos_token=False,
+                    strip_bos_token=False,
+                )
+                # everything from this is masked out from the labels
+                labels = [IGNORE_TOKEN_ID] * len(res["input_ids"])
+            elif role == "human":
+                prefix = "<|user|>"
+                res = self._tokenize(
+                    prefix + " " + message.strip(),
+                    add_eos_token=False,
+                    strip_bos_token=True,
+                )
+                # everything from this is masked out from the labels
+                labels = [IGNORE_TOKEN_ID] * len(res["input_ids"])
+            elif role == "bot":
+                prefix = "<|model|>"
+                res = self._tokenize(
+                    prefix + " " + message.strip(),
+                    add_eos_token=True,
+                    strip_bos_token=True,
+                )
+                # mask out the prefix token, rest is not masked out from labels
+                # make sure we create the labels first, otherwise we get incorrect lengths
+                labels = [IGNORE_TOKEN_ID] * len(self.bot_prefix_token_ids) + [
+                    *copy.deepcopy(res["input_ids"])
+                ][len(self.bot_prefix_token_ids) :]
+            else:
+                logging.warning(f"unknown role in conversation: {role}")
+                res = defaultdict(lambda: [])
+
+            # pylint: disable=duplicate-code
+            result, current_len = parse_tokenized_to_result(
+                result,
+                current_len,
+                res,
+                labels,
+                pad_token_id=self.tokenizer.pad_token_id,
+            )
+        return result
+
+
+class PygmalionPrompter:
+    """
+    Prompter for Pygmalion.
+    """
+
+    def __init__(self, *args, **kwargs):
+        pass
+
+    def build_prompt(
+        self, source, *args, **kwargs  # pylint: disable=unused-argument
+    ) -> Generator[Tuple[str, str], None, None]:
+        for msg in source:
+            yield msg["role"], msg["value"]
+
+
+def load(tokenizer, cfg):
+    return PygmalionPromptTokenizingStrategy(
+        PygmalionPrompter(), tokenizer, cfg.train_on_inputs, cfg.sequence_len
+    )

src/axolotl/prompt_strategies/sharegpt_jokes.py

@@ -0,0 +1,28 @@
+"""Module for Jokes prompts using sharegpt style """
+from axolotl.prompt_tokenizers import ShareGPTPromptTokenizingStrategy
+from axolotl.prompters import PromptStyle, ShareGPTPrompter
+
+
+def load(tokenizer, cfg):
+    return SimpleJokesShareGPTPromptTokenizingStrategy(
+        ShareGPTPrompter(PromptStyle.CHAT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+class SimpleJokesShareGPTPromptTokenizingStrategy(ShareGPTPromptTokenizingStrategy):
+    """
+    Tokenization strategy for asking bot to tell a joke and then explain why its funny
+    """
+
+    # title, text, explanation
+    def get_conversation_thread(self, prompt):
+        title = "" if not prompt["title"] else prompt["title"] + " "
+        return [
+            {"from": "human", "value": "Tell me a joke."},
+            {"from": "gpt", "value": title + prompt["text"]},
+            {"from": "human", "value": "Why is that joke funny?"},
+            {"from": "gpt", "value": prompt["explanation"]},
+        ]

src/axolotl/prompt_strategies/sharegpt_simple.py

@@ -0,0 +1,67 @@
+"""Module containing the SimpleShareGPTPromptTokenizingStrategy class"""
+
+from axolotl.prompt_tokenizers import ShareGPTPromptTokenizingStrategy
+from axolotl.prompters import PromptStyle, ShareGPTPrompter
+
+
+def load(tokenizer, cfg):
+    return SimpleShareGPTPromptTokenizingStrategy(
+        ShareGPTPrompter(PromptStyle.CHAT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+def load_role(tokenizer, cfg):
+    return SimpleRoleShareGPTPromptTokenizingStrategy(
+        ShareGPTPrompter(PromptStyle.CHAT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+def load_guanaco(tokenizer, cfg):
+    return GuanacoShareGPTPromptTokenizingStrategy(
+        ShareGPTPrompter(PromptStyle.CHAT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+class SimpleShareGPTPromptTokenizingStrategy(ShareGPTPromptTokenizingStrategy):
+    """
+    basic sharegpt strategy to grab conversations from the sample row
+    """
+
+    def get_conversation_thread(self, prompt):
+        return prompt["conversations"]
+
+
+class SimpleRoleShareGPTPromptTokenizingStrategy(ShareGPTPromptTokenizingStrategy):
+    """
+    basic sharegpt strategy to grab conversations from the sample row, but uses role instead of from
+    """
+
+    def get_conversation_thread(self, prompt):
+        conversations = prompt["conversations"]
+        # remap role: prompter/assistant, text: ... => from: human/gpt, value: ...
+        turns = [{"from": t["role"], "value": t["value"]} for t in conversations]
+        return turns
+
+
+class GuanacoShareGPTPromptTokenizingStrategy(ShareGPTPromptTokenizingStrategy):
+    """
+    sharegpt strategy that remaps oasst data to sharegpt format
+    """
+
+    def get_conversation_thread(self, prompt):
+        conversations = prompt["conversations"]
+        # remap role: prompter/assistant, text: ... => from: human/gpt, value: ...
+        role_map = {"prompter": "human", "assistant": "gpt"}
+        turns = [
+            {"from": role_map[t["role"]], "value": t["text"]} for t in conversations
+        ]
+        return turns

src/axolotl/prompt_tokenizers.py

@@ -1,19 +1,33 @@
+"""Module containing PromptTokenizingStrategy and Prompter classes"""
+
 import abc
+import copy
+import functools
+import logging
+from typing import Dict, List, Tuple, Union
 
 from transformers import PreTrainedTokenizer
 
+from axolotl.prompters import IGNORE_TOKEN_ID
+
 IGNORE_INDEX = -100
-LLAMA_DEFAULT_PAD_TOKEN = "[PAD]"
-LLAMA_DEFAULT_EOS_TOKEN = "</s>"
-LLAMA_DEFAULT_BOS_TOKEN = "<s>"
-LLAMA_DEFAULT_UNK_TOKEN = "<unk>"
+LLAMA_DEFAULT_PAD_TOKEN = "[PAD]"  # nosec
+LLAMA_DEFAULT_EOS_TOKEN = "</s>"  # nosec
+LLAMA_DEFAULT_BOS_TOKEN = "<s>"  # nosec
+LLAMA_DEFAULT_UNK_TOKEN = "<unk>"  # nosec
 
 
 class InvalidDataException(Exception):
-    pass
+    """
+    Exception raised when the data is invalid
+    """
 
 
 class PromptTokenizingStrategy(abc.ABC):
+    """
+    Abstract class for tokenizing strategies
+    """
+
     def __init__(
         self,
         prompter,
@@ -30,37 +44,21 @@ class PromptTokenizingStrategy(abc.ABC):
     def tokenize_prompt(self, prompt):
         pass
 
+    @functools.lru_cache(maxsize=128)
+    def _get_user_token(self):
+        id_or_ids = self.tokenizer.convert_tokens_to_ids("<|USER|>")
+        if isinstance(id_or_ids, (int,)):
+            return id_or_ids
+        return False
 
-class InstructionPromptTokenizingStrategy(PromptTokenizingStrategy):
-    def parse_instruction_fields(self, prompt) -> (str, str, str):
-        raise NotImplementedError
+    @functools.lru_cache(maxsize=128)
+    def _get_assistant_token(self):
+        id_or_ids = self.tokenizer.convert_tokens_to_ids("<|ASSISTANT|>")
+        if isinstance(id_or_ids, (int,)):
+            return id_or_ids
+        return False
 
-    def tokenize_prompt(self, prompt):
-        instruction, input, response = self.parse_instruction_fields(prompt)
-        full_prompt = self._build_full_prompt(instruction, input, response)
-        tokenized_full_prompt = self._tokenize(full_prompt)
-        if not self.train_on_inputs:
-            user_prompt = self.prompter.build_prompt(
-                instruction,
-                input,
-            )
-            tokenized_user_prompt = self._tokenize(user_prompt, add_eos_token=False)
-            user_prompt_len = len(tokenized_user_prompt["input_ids"])
-            # TODO this could be sped up using numpy array slicing
-            tokenized_full_prompt["labels"] = [
-                -100
-            ] * user_prompt_len + tokenized_full_prompt["labels"][user_prompt_len:]
-
-        return tokenized_full_prompt
-
-    def _build_full_prompt(self, instruction, input, response):
-        return self.prompter.build_prompt(
-            instruction,
-            input,
-            response,
-        )
-
-    def _tokenize(self, prompt, add_eos_token=True):
+    def _tokenize(self, prompt: str, add_eos_token=True, strip_bos_token=False):
         result = self.tokenizer(
             prompt,
             truncation=True,
@@ -76,12 +74,72 @@ class InstructionPromptTokenizingStrategy(PromptTokenizingStrategy):
             result["input_ids"].append(self.tokenizer.eos_token_id)
             result["attention_mask"].append(1)
 
+        if result["input_ids"][0] == self.tokenizer.bos_token_id and strip_bos_token:
+            result["input_ids"] = result["input_ids"][1:]
+            result["attention_mask"] = result["attention_mask"][1:]
+
         result["labels"] = result["input_ids"].copy()
         return result
 
 
+class InstructionPromptTokenizingStrategy(PromptTokenizingStrategy):
+    """
+    Tokenizing strategy for instruction-based prompts.
+    """
+
+    def parse_instruction_fields(
+        self, prompt
+    ) -> Union[Tuple[str, str, str], Tuple[str, str, str, str]]:
+        raise NotImplementedError
+
+    def tokenize_prompt(self, prompt):
+        (
+            instruction,
+            input,  # pylint: disable=redefined-builtin
+            response,
+        ) = self.parse_instruction_fields(prompt)
+        user_prompt = next(
+            iter(
+                self.prompter.build_prompt(
+                    instruction,
+                    input,
+                )
+            )
+        )
+        tokenized_prompt = self._tokenize(user_prompt, add_eos_token=False)
+        if not self.train_on_inputs:
+            user_prompt_len = len(tokenized_prompt["input_ids"])
+            # TODO this could be sped up using numpy array slicing
+            tokenized_prompt["labels"] = [-100] * user_prompt_len
+        tokenized_res_prompt = self._tokenize(
+            response, strip_bos_token=True, add_eos_token=True
+        )
+        tokenized_prompt["input_ids"] += tokenized_res_prompt["input_ids"]
+        tokenized_prompt["attention_mask"] += tokenized_res_prompt["attention_mask"]
+        tokenized_prompt["labels"] += tokenized_res_prompt["input_ids"]
+
+        return tokenized_prompt
+
+    def _build_full_prompt(
+        self, instruction, input, response  # pylint: disable=redefined-builtin
+    ):
+        return next(
+            iter(
+                self.prompter.build_prompt(
+                    instruction,
+                    input,
+                    response,
+                )
+            )
+        )
+
+
 class AlpacaPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
-    def parse_instruction_fields(self, prompt) -> (str, str, str):
+    """
+    Tokenizing strategy for Alpaca prompts.
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
         return (
             prompt["instruction"],
             prompt["input"] if "input" in prompt else "",
@@ -89,8 +147,38 @@ class AlpacaPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
         )
 
 
+class AlpacaMultipleChoicePromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
+    """
+    Tokenizing strategy for Alpaca Multiple Choice prompts.
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
+        return (
+            prompt["question"],
+            "\n".join(f'- "{choice}"' for choice in prompt["choices"]),
+            prompt["solution"] if "solution" in prompt else prompt["explanation"],
+        )
+
+
+class JeopardyPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
+    """
+    Tokenizing strategy for Jeopardy prompts.
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
+        return (
+            prompt["question"],
+            prompt["category"],
+            "what is " + prompt["answer"],
+        )
+
+
 class OpenAssistantPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
-    def parse_instruction_fields(self, prompt) -> (str, str, str):
+    """
+    Tokenizing strategy for OpenAssistant prompts.
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
         return (
             prompt["INSTRUCTION"],
             "",
@@ -98,8 +186,25 @@ class OpenAssistantPromptTokenizingStrategy(InstructionPromptTokenizingStrategy)
         )
 
 
+class SummarizeTLDRPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
+    """
+    Tokenizing strategy for SummarizeTLDR prompts.
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
+        return (
+            prompt["article"],
+            "",
+            prompt["summary"],
+        )
+
+
 class GPTeacherPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
-    def parse_instruction_fields(self, prompt) -> (str, str, str):
+    """
+    Tokenizing strategy for GPTeacher prompts.
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
         return (
             prompt["instruction"],
             prompt["input"] if "input" in prompt else "",
@@ -107,18 +212,64 @@ class GPTeacherPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
         )
 
 
+class NomicGPT4AllPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
+    """
+    Tokenizing strategy for NomicGPT4All prompts.
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str]:
+        return (
+            prompt["prompt"],
+            "",
+            prompt["response"],
+        )
+
+
+class CompletionPromptTokenizingStrategy(InstructionPromptTokenizingStrategy):
+    """
+    Tokenizing strategy for Completion prompts.
+    """
+
+    def tokenize_prompt(self, prompt):
+        full_prompt = self._build_full_prompt(prompt["text"], None, None)
+        tokenized_full_prompt = self._tokenize(full_prompt)
+
+        return tokenized_full_prompt
+
+    def _build_full_prompt(
+        self, instruction, input, response
+    ):  # pylint: disable=redefined-builtin
+        return next(iter(self.prompter.build_prompt(instruction, input, response)))
+
+
 class ReflectionPromptTokenizingStrategy(PromptTokenizingStrategy):
-    def parse_instruction_fields(self, prompt) -> (str, str, str, str, str):
+    """
+    Tokenizing strategy for Reflection prompts.
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str, str, str]:
         raise NotImplementedError
 
     def tokenize_prompt(self, prompt):
-        instruction, input, output, reflection, corrected = self.parse_instruction_fields(prompt)
-        full_prompt = self._build_full_prompt(instruction, input, output, reflection, corrected)
+        (
+            instruction,
+            input,  # pylint: disable=redefined-builtin
+            output,
+            reflection,
+            corrected,
+        ) = self.parse_instruction_fields(prompt)
+        full_prompt = self._build_full_prompt(
+            instruction, input, output, reflection, corrected
+        )
         tokenized_full_prompt = self._tokenize(full_prompt)
         if not self.train_on_inputs:
-            user_prompt = self.prompter.build_prompt(
-                instruction,
-                input,
+            user_prompt = next(
+                iter(
+                    self.prompter.build_prompt(
+                        instruction,
+                        input,
+                    )
+                )
             )
             tokenized_user_prompt = self._tokenize(user_prompt, add_eos_token=False)
             user_prompt_len = len(tokenized_user_prompt["input_ids"])
@@ -129,16 +280,22 @@ class ReflectionPromptTokenizingStrategy(PromptTokenizingStrategy):
 
         return tokenized_full_prompt
 
-    def _build_full_prompt(self, instruction, input, output, reflection, corrected):
-        return self.prompter.build_prompt(
-            instruction,
-            input,
-            output,
-            reflection,
-            corrected,
+    def _build_full_prompt(
+        self, instruction, input, output, reflection, corrected
+    ):  # pylint: disable=redefined-builtin
+        return next(
+            iter(
+                self.prompter.build_prompt(
+                    instruction,
+                    input,
+                    output,
+                    reflection,
+                    corrected,
+                )
+            )
         )
 
-    def _tokenize(self, prompt, add_eos_token=True):
+    def _tokenize(self, prompt, add_eos_token=True, strip_bos_token=False):
         result = self.tokenizer(
             prompt,
             truncation=True,
@@ -159,7 +316,11 @@ class ReflectionPromptTokenizingStrategy(PromptTokenizingStrategy):
 
 
 class AlpacaReflectionPTStrategy(ReflectionPromptTokenizingStrategy):
-    def parse_instruction_fields(self, prompt) -> (str, str, str, str, str):
+    """
+    Tokenizing strategy for Alpaca Reflection prompts.
+    """
+
+    def parse_instruction_fields(self, prompt) -> Tuple[str, str, str, str, str]:
         return (
             prompt["instruction"],
             prompt["input"] if "input" in prompt else "",
@@ -168,9 +329,131 @@ class AlpacaReflectionPTStrategy(ReflectionPromptTokenizingStrategy):
             prompt["corrected"],
         )
 
+
 class ShareGPTPromptTokenizingStrategy(PromptTokenizingStrategy):
+    """
+    Tokenizing strategy for ShareGPT prompts.
+    """
+
+    def get_conversation_thread(self, prompt):
+        return prompt["conversations"]
+
     def tokenize_prompt(self, prompt):
+        result, current_len = tokenize_prompt_default()
+        user_token = self._get_user_token()
+        assistant_token = self._get_assistant_token()
         try:
-            return self.prompter.build_prompt(prompt["conversations"], self.tokenizer)
-        except (KeyError, AssertionError, IndexError) as e:
-            raise InvalidDataException(str(e))
+            for _, part in enumerate(
+                self.prompter.build_prompt(self.get_conversation_thread(prompt))
+            ):
+                if isinstance(part, tuple):
+                    if part[0] == "USER:":
+                        part = part[0] + part[1] if not user_token else part[1]
+                        # this is still the user query, we should
+                        res = self._tokenize(
+                            part.strip(),
+                            add_eos_token=False,
+                            strip_bos_token=True,
+                        )
+                        if user_token:
+                            res["input_ids"] = [user_token, *res["input_ids"]]
+                        # everything from this is masked out from the labels
+                        labels = [IGNORE_TOKEN_ID] * len(res["input_ids"])
+                    elif part[0] == "ASSISTANT:":
+                        # TODO label assistant token/tokens w/ IGNORE_TOKEN_ID
+                        part = part[0] + part[1] if not assistant_token else part[1]
+                        # this should be the assistent response, should end with an eos token
+                        res = self._tokenize(
+                            part.strip(),
+                            add_eos_token=True,
+                            strip_bos_token=True,
+                        )
+                        if assistant_token:
+                            res["input_ids"] = [
+                                assistant_token,
+                                *res["input_ids"],
+                            ]
+                        # not masked out from labels
+                        labels = copy.deepcopy(res["input_ids"])
+                    elif part[0] == "SYSTEM:":
+                        part = part[1]  # Ignore the system role from preamble
+                        # this is only ever the first part, should include the bos token and the user query
+                        res = self._tokenize(
+                            part.strip(), add_eos_token=False, strip_bos_token=False
+                        )
+                        # everything from this is masked out from the labels
+                        labels = [IGNORE_TOKEN_ID] * len(res["input_ids"])
+                    else:
+                        logging.warning(f"unhandled role: {part[0]}")
+
+                # pylint: disable=duplicate-code
+                result, current_len = parse_tokenized_to_result(
+                    result,
+                    current_len,
+                    res,
+                    labels,
+                    pad_token_id=self.tokenizer.pad_token_id,
+                )
+            return result
+        except (KeyError, AssertionError, IndexError) as err:
+            raise InvalidDataException(str(err)) from err
+
+    def _tokenize(self, prompt, add_eos_token=True, strip_bos_token=False):
+        result = self.tokenizer(
+            prompt,
+            truncation=True,
+            max_length=self.sequence_len,
+            padding=False,
+            return_tensors=None,
+        )
+        if (
+            result["input_ids"][-1] != self.tokenizer.eos_token_id
+            and len(result["input_ids"]) < self.sequence_len
+            and add_eos_token
+        ):
+            result["input_ids"].append(self.tokenizer.eos_token_id)
+            result["attention_mask"].append(1)
+
+        if result["input_ids"][0] == self.tokenizer.bos_token_id and strip_bos_token:
+            result["input_ids"] = result["input_ids"][1:]
+            result["attention_mask"] = result["attention_mask"][1:]
+
+        result["labels"] = result["input_ids"].copy()
+        return result
+
+
+def tokenize_prompt_default() -> Tuple[Dict[str, List[int]], int]:
+    """
+    Returns the default values for the tokenize prompt function
+    """
+
+    result: Dict[str, List[int]] = {
+        "input_ids": [],
+        "attention_mask": [],
+        "labels": [],
+    }
+    current_len = 0
+    return result, current_len
+
+
+def parse_tokenized_to_result(
+    result: Dict[str, List[int]],
+    current_len: int,
+    res: Dict[str, List[int]],
+    labels: List[int],
+    pad_token_id: Union[int, None] = None,
+) -> Tuple[Dict[str, List[int]], int]:
+    """
+    Parses the tokenized prompt and append the tokenized input_ids, attention_mask and labels to the result
+    """
+
+    input_ids = res["input_ids"]
+    input_len = len(input_ids)
+    result["input_ids"][current_len : current_len + input_len] = input_ids
+    result["attention_mask"][current_len : current_len + input_len] = [
+        1 if x != pad_token_id else 0 for x in input_ids
+    ]
+    result["labels"][current_len : current_len + input_len] = labels
+    current_len += input_len
+
+    return result, current_len

src/axolotl/prompters.py

@@ -1,54 +1,201 @@
-import copy
+"""Module containing prompters"""
+
 import dataclasses
-from enum import auto, Enum
-from typing import List, Tuple, Any, Union
+import logging
+from enum import Enum, auto
+from typing import Generator, List, Optional, Tuple, Union
 
 IGNORE_TOKEN_ID = -100
 
 
+class PromptStyle(Enum):
+    """
+    Enum for prompt styles
+    """
+
+    INSTRUCT = "instruct"
+    CHAT = "chat"
+
+
 class AlpacaPrompter:
-    prompt_input = "Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.\n\n### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:\n"
-    prompt_no_input = "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n{instruction}\n\n### Response:\n"
-    response_split = "### Response:"
+    """
+    Base class for alpaca prompters
+    """
+
+    system_prompt = "Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.\n\n"
+    system_no_input_prompt = "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n"
+    turn_format: str
+    turn_no_input_format: str
+    prompt_style: Optional[PromptStyle] = None
+
+    def __init__(self, prompt_style=PromptStyle.INSTRUCT.value):
+        self.prompt_style = prompt_style if prompt_style else PromptStyle.INSTRUCT.value
+        self.match_prompt_style()
+
+    def match_prompt_style(self):
+        if self.prompt_style == PromptStyle.INSTRUCT.value:
+            self.turn_format = "### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:\n"
+            self.turn_no_input_format = (
+                "### Instruction:\n{instruction}\n\n### Response:\n"
+            )
+        if self.prompt_style == PromptStyle.CHAT.value:
+            self.turn_format = "USER: {instruction}\n{input}\nASSISTANT:"
+            self.turn_no_input_format = "USER: {instruction}\nASSISTANT:"
 
     def build_prompt(
         self,
         instruction: str,
-        input: Union[None, str] = None,
+        input: Union[None, str] = None,  # pylint: disable=redefined-builtin
         output: Union[None, str] = None,
-    ) -> str:
+    ) -> Generator[str, None, None]:
         # returns the full prompt from instruction and optional input
         # if a label (=response, =output) is provided, it's also appended.
         if input:
-            res = self.prompt_input.format(instruction=instruction, input=input)
+            res = self.system_prompt + self.turn_format.format(
+                instruction=instruction, input=input
+            )
         else:
-            res = self.prompt_no_input.format(instruction=instruction)
+            res = self.system_no_input_prompt + self.turn_no_input_format.format(
+                instruction=instruction
+            )
         if output:
             res = f"{res}{output}"
-        return res
-
-    def get_response(self, output: str) -> str:
-        return output.split(self.response_split)[1].strip()
+        yield res
 
 
-class GPTeacherPrompter(AlpacaPrompter):
-    ...
+class UnpromptedPrompter(AlpacaPrompter):
+    """
+    Prompter for alpaca no system prompt
+    """
+
+    system_prompt = ""
+    system_no_input_prompt = ""
 
 
-class ReflectAlpacaPrompter:
-    prompt_input = "Below is an instruction that describes a task, paired with an input that provides further context. You, the Assistant, should generate a response as if it were an abstract for an academic or technical paper on the query along with a methodology. Then generate an Agent Reflection where you create a long form response as if from subject matter expert, be verbose, diligent, and creative in your application of knowledge, apply it through the lens of the response generated by the assistant. Look for flawed reasoning, faulty logic, or other mistakes in the method. Finally, generate a final response and method for the user with the Assistant abstract and Reflection analysis as augmentations to the generation\n\n### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:\n"
-    prompt_no_input = "Below is an instruction that describes a task. You, the Assistant, should generate a response as if it were an abstract for an academic or technical paper on the query along with a methodology. Then generate an Agent Reflection where you create a long form response as if from subject matter expert, be verbose, diligent, and creative in your application of knowledge, apply it through the lens of the response generated by the assistant. Look for flawed reasoning, faulty logic, or other mistakes in the method. Finally, generate a final response and method for the user with the Assistant abstract and Reflection analysis as augmentations to the generation\n\n### Instruction:\n{instruction}\n\n### Response:\n"
-    agent_label = "{output}\n\n### Agent Reflection:\n{reflection}\n\n### Final Response:\n{corrected}"
-    response_split = "### Response:"
+class JeopardyPrompter(AlpacaPrompter):
+    """
+    Prompter for Jeopardy
+    """
+
+    prompt_input = "Below is a Jeopardy clue paired with input providing the category of the clue. Write a concise response that best answers tbe clue given the category.\n\n### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:\n"
+
+
+class MultipleChoiceExplainPrompter(AlpacaPrompter):
+    """
+    Prompter for multiple choice explain
+    """
+
+    system_prompt = (
+        "Choose the answer that best answers the question. Explain your reasoning.\n"
+    )
+    system_no_input_prompt = (
+        "Choose the answer that best answers the question. Explain your reasoning.\n"
+    )
+
+
+class MultipleChoiceConcisePrompter(AlpacaPrompter):
+    """
+    Prompter for multiple choice concise
+    """
+
+    system_prompt = "Choose the answer that best answers the question. Be concise in your response.\n\n"
+    system_no_input_prompt = "Choose the answer that best answers the question. Be concise in your response.\n\n"
+
+    def match_prompt_style(self):
+        self.turn_format = "USER: {instruction}\n{input}\nASSISTANT:"
+        self.turn_no_input_format = "USER: {instruction}\nASSISTANT:"
+
+
+class SummarizeTLDRPrompter(AlpacaPrompter):
+    """
+    Prompter for summarize TLDR
+    """
+
+    system_prompt = ""
+    system_no_input_prompt = ""
+
+    def match_prompt_style(self):
+        self.turn_format = "USER: Summarize the following article as a TL;DR.\n{instruction}\n{input}\nASSISTANT:"
+        self.turn_no_input_format = "USER: Summarize the following article as a TL;DR.\n{instruction}\nASSISTANT:"
+
+
+class CompletionPrompter:
+    """
+    Prompter for completion
+    """
 
     def build_prompt(
         self,
         instruction: str,
-        input: Union[None, str] = None,
+        input=None,  # pylint: disable=redefined-builtin, unused-argument
+        output=None,  # pylint: disable=unused-argument
+    ) -> Generator[str, None, None]:
+        yield instruction
+
+
+class GPTeacherPrompter(AlpacaPrompter):
+    """
+    Prompter for GPTeacher
+    """
+
+
+class NomicGPT4AllPrompter(AlpacaPrompter):
+    """
+    Prompter for NomicGPT4All
+    """
+
+
+class ReflectAlpacaPrompter:
+    """
+    Prompter for ReflectAlpaca
+    """
+
+    system_prompt = "Below is an instruction that describes a task, paired with an input that provides further context. You, the Assistant, should generate a response as if it were an abstract for an academic or technical paper on the query along with a methodology. Then generate an Agent Reflection where you create a long form response as if from subject matter expert, be verbose, diligent, and creative in your application of knowledge, apply it through the lens of the response generated by the assistant. Look for flawed reasoning, faulty logic, or other mistakes in the method. Finally, generate a final response and method for the user with the Assistant abstract and Reflection analysis as augmentations to the generation\n\n"
+    system_no_input_prompt = "Below is an instruction that describes a task. You, the Assistant, should generate a response as if it were an abstract for an academic or technical paper on the query along with a methodology. Then generate an Agent Reflection where you create a long form response as if from subject matter expert, be verbose, diligent, and creative in your application of knowledge, apply it through the lens of the response generated by the assistant. Look for flawed reasoning, faulty logic, or other mistakes in the method. Finally, generate a final response and method for the user with the Assistant abstract and Reflection analysis as augmentations to the generation\n\n"
+
+    prompt_input = (
+        "### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:\n"
+    )
+    prompt_no_input = "### Instruction:\n{instruction}\n\n### Response:\n"
+    agent_label = "### Thought:\n{output}\n\n### Agent Reflection:\n{reflection}\n\n### Final Response:\n{corrected}"
+    response_split = "### Response:"
+
+    def __init__(self, prompt_style="instruct"):
+        self.prompt_style = prompt_style
+        self.match_prompt_style()
+
+    def match_prompt_style(self):
+        if self.prompt_style == PromptStyle.INSTRUCT.value:
+            self.prompt_input = (
+                self.system_prompt
+                + "### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:\n"
+            )
+            self.prompt_no_input = (
+                self.system_no_input_prompt
+                + "### Instruction:\n{instruction}\n\n### Response:\n"
+            )
+            self.agent_label = "### Thought:\n{output}\n\n### Agent Reflection:\n{reflection}\n\n### Final Response:\n{corrected}"
+            self.response_split = "### Final Response:"
+        if self.prompt_style == PromptStyle.CHAT.value:
+            self.prompt_input = (
+                self.system_prompt + "USER: {instruction}\n{input}\nASSISTANT:"
+            )
+            self.prompt_no_input = (
+                self.system_no_input_prompt + "USER: {instruction}\nASSISTANT:"
+            )
+            self.agent_label = (
+                "\nTHOUGHT: {output}\nASSISTANT REFLECTION: {reflection}\nASSISTANT:"
+            )
+            self.response_split = "ASSISTANT:"
+
+    def build_prompt(
+        self,
+        instruction: str,
+        input: Union[None, str] = None,  # pylint: disable=redefined-builtin
         output: Union[None, str] = None,
         reflection: Union[None, str] = None,
         corrected: Union[None, str] = None,
-    ) -> str:
+    ) -> Generator[str, None, None]:
         # returns the full prompt from instruction and optional input
         # if a label (=response, =output) is provided, it's also appended.
         if input:
@@ -56,12 +203,13 @@ class ReflectAlpacaPrompter:
         else:
             res = self.prompt_no_input.format(instruction=instruction)
         if output and reflection and corrected:
-            label = self.agent_label.format(output=output, reflection=reflection, corrected=corrected)
+            label = self.agent_label.format(
+                output=output,
+                reflection=reflection,
+                corrected=corrected,
+            )
             res = f"{res}{label}"
-        return res
-
-    def get_response(self, output: str) -> str:
-        return output.split(self.response_split)[1].strip()
+        yield res
 
 
 class SeparatorStyle(Enum):
@@ -83,17 +231,18 @@ class Conversation:
     offset: int
     sep_style: SeparatorStyle = SeparatorStyle.SINGLE
     sep: str = "###"
-    sep2: str = None
+    sep2: Optional[str] = None
 
-    def get_prompt(self):
-        seps = [self.sep, self.sep2]
-        ret = self.system + seps[0]
-        for i, (role, message) in enumerate(self.messages):
+    def get_prompt(self) -> Generator[Tuple[str, str], None, None]:
+        # seps = [self.sep, self.sep2]
+        preamble = self.system + self.sep
+        yield ("SYSTEM:", preamble)
+        for _, (role, message) in enumerate(self.messages):
             if message:
-                ret += role + ": " + message + seps[i % 2]
+                yield (role + ":", " " + message)
             else:
-                ret += role + ":"
-        return ret
+                logging.warning(f"role with empty message: {role}")
+                yield (role + ":", "")
 
     def copy(self):
         return Conversation(
@@ -110,26 +259,45 @@ class Conversation:
         self.messages.append([role, message])
 
 
-conv_vicuna_v1_1 = Conversation(
-    system="A chat between a curious user and an artificial intelligence assistant. "
-    "The assistant gives helpful, detailed, and polite answers to the user's questions.",
-    roles=["USER", "ASSISTANT"],
-    messages=[],
-    offset=0,
-    sep_style=SeparatorStyle.TWO,
-    sep=" ",
-    sep2="</s>",
-)
+class ShareGPTPrompter:  # pylint: disable=too-few-public-methods
+    """
+    A prompter that generates prompts for the ShareGPT
+    """
 
+    def __init__(self, prompt_style=None, system_prompt: Optional[str] = None):
+        if prompt_style != PromptStyle.CHAT.value:
+            raise ValueError(
+                f"unsupported prompt_style for ShareGPTPrompter({prompt_style})"
+            )
+        system: str = (
+            system_prompt
+            if system_prompt
+            else (
+                "A chat between a curious user and an artificial intelligence assistant. "
+                "The assistant gives helpful, detailed, and polite answers to the user's questions."
+            )
+        )
+        self._conversation = Conversation(
+            system=system,
+            roles=["USER", "ASSISTANT"],
+            messages=[],
+            offset=0,
+            sep_style=SeparatorStyle.TWO,
+            sep=" ",
+            sep2=" ",
+        )
+
+    def build_prompt(self, source) -> Generator[str, None, None]:
+        # ignore the system prompt if provided
+        if source[0]["from"] == "system":
+            source.pop(0)
 
-class ShareGPTPrompter:
-    def build_prompt(self, source, tokenizer):
         if len(source) < 2:
             # If there isn't a back and forth conversation, ignore it
             # also happens on the data splitting leaving empty conversations
             raise IndexError
 
-        conv = conv_vicuna_v1_1.copy()
+        conv = self._conversation.copy()
         roles = {"human": conv.roles[0], "gpt": conv.roles[1]}
 
         try:
@@ -140,55 +308,15 @@ class ShareGPTPrompter:
             ):
                 # Skip the first one if it is not from human
                 source = source[1:]
-        except IndexError as e:
+        except IndexError as err:
             # sometimes there is a bing or system chat
-            raise e
+            raise err
 
         conv.messages = []
         for j, sentence in enumerate(source):
             role = roles[sentence["from"]]
             assert role == conv.roles[j % 2]
             conv.append_message(role, sentence["value"])
-        conversation = conv.get_prompt()
 
-        # Tokenize conversations
-        tokenized_result = tokenizer(
-            conversation,
-            truncation=True,
-            max_length=2048,  # FIXME
-            padding=False,
-            return_tensors=None,
-        )
-        target = copy.deepcopy(tokenized_result["input_ids"])
-
-        # Mask targets
-        sep = conv.sep + conv.roles[1] + ": "
-
-        rounds = conversation.split(conv.sep2)
-        cur_len = 1
-        for i, rou in enumerate(rounds):
-            if rou == "":
-                break
-
-            parts = rou.split(sep)
-            if len(parts) != 2:
-                break
-            parts[0] += sep
-            round_len = len(tokenizer(rou)["input_ids"])
-            instruction_len = len(tokenizer(parts[0])["input_ids"]) - 2
-            target[cur_len : cur_len + instruction_len] = [
-                IGNORE_TOKEN_ID
-            ] * instruction_len
-
-            cur_len += round_len
-        target[cur_len:] = [IGNORE_TOKEN_ID] * (len(target) - cur_len)
-        attention_mask = [
-            1 if x != tokenizer.pad_token_id else 0
-            for x in tokenized_result["input_ids"]
-        ]
-
-        return dict(
-            input_ids=tokenized_result["input_ids"],
-            labels=target,
-            attention_mask=attention_mask,
-        )
+        for part in conv.get_prompt():
+            yield part

src/axolotl/utils/callbacks.py

@@ -0,0 +1,69 @@
+"""Callbacks for Trainer class"""
+
+import os
+
+from optimum.bettertransformer import BetterTransformer
+from transformers import (
+    TrainerCallback,
+    TrainerControl,
+    TrainerState,
+    TrainingArguments,
+)
+from transformers.trainer_utils import PREFIX_CHECKPOINT_DIR, IntervalStrategy
+
+
+class SavePeftModelCallback(TrainerCallback):  # pylint: disable=too-few-public-methods
+    """Callback to save the PEFT adapter"""
+
+    def on_save(
+        self,
+        args: TrainingArguments,
+        state: TrainerState,
+        control: TrainerControl,
+        **kwargs,
+    ):
+        checkpoint_folder = os.path.join(
+            args.output_dir,
+            f"{PREFIX_CHECKPOINT_DIR}-{state.global_step}",
+        )
+
+        peft_model_path = os.path.join(checkpoint_folder, "adapter_model")
+        kwargs["model"].save_pretrained(peft_model_path)
+
+        return control
+
+
+class SaveBetterTransformerModelCallback(
+    TrainerCallback
+):  # pylint: disable=too-few-public-methods
+    """Callback to save the BetterTransformer wrapped model"""
+
+    def on_step_end(
+        self,
+        args: TrainingArguments,
+        state: TrainerState,
+        control: TrainerControl,
+        **kwargs,
+    ):
+        # Save
+        if (
+            args.save_strategy == IntervalStrategy.STEPS
+            and args.save_steps > 0
+            and state.global_step % args.save_steps == 0
+        ):
+            control.should_save = True
+
+        if control.should_save:
+            checkpoint_folder = os.path.join(
+                args.output_dir,
+                f"{PREFIX_CHECKPOINT_DIR}-{state.global_step}",
+            )
+
+            model = BetterTransformer.reverse(kwargs["model"])
+            model.save_pretrained(checkpoint_folder)
+            # FIXME - need to cleanup old checkpoints
+
+            # since we're saving here, we don't need the trainer loop to attempt to save too b/c
+            # the trainer will raise an exception since it can't save a BetterTransformer wrapped model
+            control.should_save = False
+        return control

src/axolotl/utils/data.py

@@ -0,0 +1,552 @@
+"""Module containing data utilities"""
+import functools
+import logging
+from hashlib import md5
+from pathlib import Path
+from typing import List, Tuple, Union
+
+import torch
+from datasets import Dataset, DatasetDict, load_dataset, load_from_disk
+from huggingface_hub import hf_hub_download
+from transformers import PreTrainedTokenizerBase
+
+from axolotl.datasets import ConstantLengthDataset, TokenizedPromptDataset
+from axolotl.prompt_strategies import load
+from axolotl.prompt_tokenizers import (
+    AlpacaMultipleChoicePromptTokenizingStrategy,
+    AlpacaPromptTokenizingStrategy,
+    AlpacaReflectionPTStrategy,
+    CompletionPromptTokenizingStrategy,
+    GPTeacherPromptTokenizingStrategy,
+    JeopardyPromptTokenizingStrategy,
+    OpenAssistantPromptTokenizingStrategy,
+    ShareGPTPromptTokenizingStrategy,
+    SummarizeTLDRPromptTokenizingStrategy,
+)
+from axolotl.prompters import (
+    AlpacaPrompter,
+    CompletionPrompter,
+    GPTeacherPrompter,
+    JeopardyPrompter,
+    MultipleChoiceConcisePrompter,
+    MultipleChoiceExplainPrompter,
+    ReflectAlpacaPrompter,
+    ShareGPTPrompter,
+    SummarizeTLDRPrompter,
+)
+
+
+def load_tokenized_prepared_datasets(
+    split, tokenizer, cfg, default_dataset_prepared_path
+) -> DatasetDict:
+    tokenizer_name = tokenizer.__class__.__name__
+    ds_hash = str(
+        md5(  # nosec
+            (
+                str(cfg.sequence_len)
+                + "@"
+                + "|".join(
+                    sorted([f"{d.path}:{d.type}:{d.shards}" for d in cfg.datasets])
+                )
+                + "|"
+                + split
+                + "|"
+                + tokenizer_name
+            ).encode("utf-8")
+        ).hexdigest()
+    )
+    prepared_ds_path = (
+        Path(cfg.dataset_prepared_path) / ds_hash
+        if cfg.dataset_prepared_path
+        else Path(default_dataset_prepared_path) / ds_hash
+    )
+    dataset = None
+    use_auth_token = cfg.hf_use_auth_token
+    try:
+        if cfg.push_dataset_to_hub:
+            dataset = load_dataset(
+                f"{cfg.push_dataset_to_hub}/{ds_hash}",
+                use_auth_token=use_auth_token,
+            )
+            dataset = dataset[split]
+    except Exception:  # pylint: disable=broad-except # nosec
+        pass
+
+    if dataset:
+        ...
+    elif any(prepared_ds_path.glob("*")):
+        logging.info(f"Loading prepared dataset from disk at {prepared_ds_path}...")
+        dataset = load_from_disk(str(prepared_ds_path))
+        logging.info("Prepared dataset loaded from disk...")
+    else:
+        logging.info(f"Unable to find prepared dataset in {prepared_ds_path}")
+        logging.info("Loading raw datasets...")
+
+        if cfg.seed:
+            seed = cfg.seed
+        else:
+            logging.info("No seed provided, using default seed of 42")
+            seed = 42
+
+        datasets = []
+        # pylint: disable=invalid-name
+        for d in cfg.datasets:
+            ds: Union[Dataset, DatasetDict] = None
+            ds_from_hub = False
+            try:
+                load_dataset(
+                    d.path,
+                    streaming=True,
+                    use_auth_token=use_auth_token,
+                )
+                ds_from_hub = True
+            except FileNotFoundError:
+                pass
+
+            # prefer local dataset, even if hub exists
+            if Path(d.path).exists():
+                ds = load_dataset(
+                    "json",
+                    data_files=d.path,
+                    streaming=False,
+                    split=None,
+                )
+            elif ds_from_hub:
+                if d.data_files:
+                    ds = load_dataset(
+                        d.path,
+                        streaming=False,
+                        data_files=d.data_files,
+                        use_auth_token=use_auth_token,
+                    )
+                else:
+                    ds = load_dataset(
+                        d.path,
+                        streaming=False,
+                        use_auth_token=use_auth_token,
+                    )
+            else:
+                fp = hf_hub_download(
+                    repo_id=d.path,
+                    repo_type="dataset",
+                    filename=d.data_files,
+                )
+                ds = load_dataset("json", data_files=fp, streaming=False, split=None)
+            if not ds:
+                raise ValueError("unhandled dataset load")
+            # support for using a subset of the data
+            if d.shards:
+                if split in ds:
+                    ds = ds.shuffle(seed=seed)[split].shard(
+                        num_shards=d.shards, index=0
+                    )
+                else:
+                    ds = ds.shuffle(seed=seed).shard(num_shards=d.shards, index=0)
+            d_type = d.type
+            d_type_split = d_type.split(":")
+            d_base_type = d_type_split[0]
+            d_prompt_style = d_type_split[1] if len(d_type_split) > 1 else None
+            if split in ds:
+                ds = ds[split]
+            if ds_strategy := load(d.type, tokenizer, cfg):
+                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
+                datasets.append(ds_wrapper)
+            elif d_base_type == "alpaca":
+                ds_strategy = AlpacaPromptTokenizingStrategy(
+                    AlpacaPrompter(d_prompt_style),
+                    tokenizer,
+                    cfg.train_on_inputs,
+                    cfg.sequence_len,
+                )
+                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
+                datasets.append(ds_wrapper)
+            elif d_base_type == "explainchoice":
+                ds_strategy = AlpacaMultipleChoicePromptTokenizingStrategy(
+                    MultipleChoiceExplainPrompter(d_prompt_style),
+                    tokenizer,
+                    cfg.train_on_inputs,
+                    cfg.sequence_len,
+                )
+                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
+                datasets.append(ds_wrapper)
+            elif d_base_type == "concisechoice":
+                ds_strategy = AlpacaMultipleChoicePromptTokenizingStrategy(
+                    MultipleChoiceConcisePrompter(d_prompt_style),
+                    tokenizer,
+                    cfg.train_on_inputs,
+                    cfg.sequence_len,
+                )
+                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
+                datasets.append(ds_wrapper)
+            elif d_base_type == "summarizetldr":
+                ds_strategy = SummarizeTLDRPromptTokenizingStrategy(
+                    SummarizeTLDRPrompter(d_prompt_style),
+                    tokenizer,
+                    cfg.train_on_inputs,
+                    cfg.sequence_len,
+                )
+                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
+                datasets.append(ds_wrapper)
+            elif d_base_type == "jeopardy":
+                ds_strategy = JeopardyPromptTokenizingStrategy(
+                    JeopardyPrompter(d_prompt_style),
+                    tokenizer,
+                    cfg.train_on_inputs,
+                    cfg.sequence_len,
+                )
+                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
+                datasets.append(ds_wrapper)
+            elif d_base_type == "oasst":
+                ds_strategy = OpenAssistantPromptTokenizingStrategy(
+                    AlpacaPrompter(d_prompt_style),
+                    tokenizer,
+                    cfg.train_on_inputs,
+                    cfg.sequence_len,
+                )
+                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
+                datasets.append(ds_wrapper)
+            elif d_base_type == "gpteacher":
+                ds_strategy = GPTeacherPromptTokenizingStrategy(
+                    GPTeacherPrompter(d_prompt_style),
+                    tokenizer,
+                    cfg.train_on_inputs,
+                    cfg.sequence_len,
+                )
+                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
+                datasets.append(ds_wrapper)
+            elif d_base_type == "reflection":
+                ds_strategy = AlpacaReflectionPTStrategy(
+                    ReflectAlpacaPrompter(d_prompt_style),
+                    tokenizer,
+                    cfg.train_on_inputs,
+                    cfg.sequence_len,
+                )
+                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
+                datasets.append(ds_wrapper)
+            elif d_base_type == "sharegpt":
+                ds_strategy = ShareGPTPromptTokenizingStrategy(
+                    ShareGPTPrompter(d_prompt_style),
+                    tokenizer,
+                    cfg.train_on_inputs,
+                    cfg.sequence_len,
+                )
+                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
+                datasets.append(ds_wrapper)
+            elif d_base_type == "completion":
+                ds_strategy = CompletionPromptTokenizingStrategy(
+                    CompletionPrompter(),
+                    tokenizer,
+                    cfg.train_on_inputs,
+                    cfg.sequence_len,
+                )
+                ds_wrapper = TokenizedPromptDataset(ds_strategy, ds)
+                datasets.append(ds_wrapper)
+            else:
+                suffix = ""
+                if ":load_" in d.type:
+                    suffix = f" Did you mean {d.type.replace(':load_', '.load_')}?"
+                logging.error(
+                    f"unhandled prompt tokenization strategy: {d.type}. {suffix}"
+                )
+                raise ValueError(
+                    f"unhandled prompt tokenization strategy: {d.type} {suffix}"
+                )
+        logging.info("tokenizing, merging, and shuffling master dataset")
+
+        samples: List[int] = []
+        for d in datasets:
+            samples = samples + list(d)
+        dataset = Dataset.from_list(samples).shuffle(seed=seed)
+        if cfg.local_rank == 0:
+            logging.info(
+                f"Saving merged prepared dataset to disk... {prepared_ds_path}"
+            )
+            dataset.save_to_disk(prepared_ds_path)
+            if cfg.push_dataset_to_hub:
+                logging.info(
+                    f"Saving merged prepared dataset with push_to_hub... {cfg.push_dataset_to_hub}/{ds_hash}"
+                )
+                dataset.push_to_hub(
+                    f"{cfg.push_dataset_to_hub}/{ds_hash}", private=True
+                )
+
+    return dataset
+
+
+def load_prepare_datasets(
+    tokenizer: PreTrainedTokenizerBase,
+    cfg,
+    default_dataset_prepared_path,
+) -> Tuple[Dataset, Dataset]:
+    max_packed_sequence_len = (
+        cfg.max_packed_sequence_len if cfg.max_packed_sequence_len else cfg.sequence_len
+    )
+    max_packed_sequence_len = min(
+        max_packed_sequence_len, cfg.sequence_len
+    )  # make sure we don't accidentally set it larger than sequence_len
+
+    tokenizer_name = tokenizer.__class__.__name__
+    if cfg.max_packed_sequence_len is not None:
+        # see if we can go ahead and load the stacked dataset
+        seed = f"@{str(cfg.seed)}" if cfg.seed else ""
+        ds_hash = str(
+            md5(  # nosec
+                (
+                    str(cfg.sequence_len)
+                    + "@"
+                    + str(max_packed_sequence_len)
+                    + seed
+                    + "|".join(
+                        sorted([f"{d.path}:{d.type}:{d.shards}" for d in cfg.datasets])
+                    )
+                    + "|"
+                    + tokenizer_name
+                ).encode("utf-8")
+            ).hexdigest()
+        )
+        prepared_ds_path = (
+            Path(cfg.dataset_prepared_path) / ds_hash
+            if cfg.dataset_prepared_path
+            else Path(default_dataset_prepared_path) / ds_hash
+        )
+
+        dataset = None
+        use_auth_token = cfg.hf_use_auth_token
+        try:
+            if cfg.push_dataset_to_hub:
+                logging.info(
+                    f"Checking for packed prepared dataset from hub... {cfg.push_dataset_to_hub}/{ds_hash}"
+                )
+                dataset = load_dataset(
+                    f"{cfg.push_dataset_to_hub}/{ds_hash}",
+                    use_auth_token=use_auth_token,
+                )
+        except Exception:  # pylint: disable=broad-except # nosec
+            pass
+
+        if dataset:
+            ...
+        elif any(prepared_ds_path.glob("*")):
+            logging.info(
+                f"Loading prepared packed dataset from disk at {prepared_ds_path}..."
+            )
+            dataset = load_from_disk(str(prepared_ds_path))
+            logging.info("Prepared packed dataset loaded from disk...")
+            if cfg.push_dataset_to_hub:
+                logging.info(
+                    f"Saving packed prepared dataset with push_to_hub... {cfg.push_dataset_to_hub}/{ds_hash}"
+                )
+                dataset.push_to_hub(
+                    f"{cfg.push_dataset_to_hub}/{ds_hash}", private=True
+                )
+        else:
+            dataset_train = load_tokenized_prepared_datasets(
+                "train", tokenizer, cfg, default_dataset_prepared_path
+            )
+            dataset_test = load_tokenized_prepared_datasets(
+                "test", tokenizer, cfg, default_dataset_prepared_path
+            )
+            dataset = DatasetDict({"train": dataset_train, "test": dataset_test})
+            if cfg.seed:
+                dataset = dataset.shuffle(seed=cfg.seed)
+
+            constant_len_dataset_train = ConstantLengthDataset(
+                tokenizer,
+                [dataset["train"]],
+                seq_length=max_packed_sequence_len,
+            )
+            constant_len_dataset_test = ConstantLengthDataset(
+                tokenizer,
+                [dataset["test"]],
+                seq_length=max_packed_sequence_len,
+            )
+            logging.info(
+                f"packing master dataset to len: {cfg.max_packed_sequence_len}"
+            )
+            dataset_train = Dataset.from_list(list(constant_len_dataset_train))
+            dataset_test = Dataset.from_list(list(constant_len_dataset_test))
+
+            # filter out bad data
+            dataset_train = Dataset.from_list(
+                [
+                    d
+                    for d in dataset_train
+                    if len(d["input_ids"]) < cfg.sequence_len
+                    and len(d["input_ids"]) > 0
+                    and len(d["input_ids"]) == len(d["attention_mask"])
+                    and len(d["input_ids"]) == len(d["labels"])
+                ]
+            )
+
+            # filter out bad data
+            dataset_test = Dataset.from_list(
+                [
+                    d
+                    for d in dataset_test
+                    if len(d["input_ids"]) < cfg.sequence_len
+                    and len(d["input_ids"]) > 0
+                    and len(d["input_ids"]) == len(d["attention_mask"])
+                    and len(d["input_ids"]) == len(d["labels"])
+                ]
+            )
+            dataset = DatasetDict({"train": dataset_train, "test": dataset_test})
+
+            if cfg.local_rank == 0:
+                logging.info(
+                    f"Saving packed prepared dataset to disk... {prepared_ds_path}"
+                )
+                dataset.save_to_disk(prepared_ds_path)
+                if cfg.push_dataset_to_hub:
+                    logging.info(
+                        f"Saving packed prepared dataset with push_to_hub... {cfg.push_dataset_to_hub}/{ds_hash}"
+                    )
+                    dataset.push_to_hub(
+                        f"{cfg.push_dataset_to_hub}/{ds_hash}",
+                        private=True,
+                    )
+    else:
+        # dataset_train = load_tokenized_prepared_datasets(
+        dataset = load_tokenized_prepared_datasets(
+            "train", tokenizer, cfg, default_dataset_prepared_path
+        )
+        # dataset_test = load_tokenized_prepared_datasets(
+        #     "test", tokenizer, cfg, default_dataset_prepared_path
+        # )
+        # dataset = DatasetDict({"train": dataset_train, "test": dataset_test})
+
+    if cfg.dataset_shard_num and cfg.dataset_shard_idx is not None:
+        logging.info(
+            f"Using index #{cfg.dataset_shard_idx} of {cfg.dataset_shard_num} shards"
+        )
+        dataset = dataset.shard(
+            num_shards=cfg.dataset_shard_num,
+            index=cfg.dataset_shard_idx,
+        )
+
+    if cfg.val_set_size:
+        dataset = dataset.train_test_split(test_size=cfg.val_set_size, shuffle=False)
+        train_dataset = dataset["train"]
+        eval_dataset = dataset["test"]
+    elif "train" in dataset:
+        train_dataset = dataset["train"]
+        eval_dataset = dataset["test"]
+    else:
+        train_dataset = dataset
+        eval_dataset = None
+
+    return train_dataset, eval_dataset
+
+
+def encode_pretraining(tokenizer, max_tokens, examples):
+    res = tokenizer(
+        examples["text"],
+        truncation=True,
+        max_length=max_tokens - 2,
+        add_special_tokens=True,
+    )
+    # Convert to PyTorch tensors
+    input_ids = [torch.tensor(seq) for seq in res["input_ids"]]
+    attention_mask = [torch.tensor(seq) for seq in res["attention_mask"]]
+    new_input_ids = []
+    new_attention_mask = []
+    # Append EOS and PAD tokens to input_ids, and correct attention_mask
+    for i, _ in enumerate(input_ids):
+        input_ids[i] = torch.cat(
+            (
+                input_ids[i],
+                torch.tensor([tokenizer.eos_token_id, tokenizer.pad_token_id]),
+            ),
+            dim=0,
+        )
+        attention_mask[i] = torch.cat((attention_mask[i], torch.tensor([1, 0])), dim=0)
+
+    # Concatenate tokens so that their lengths are less than max_tokens
+    buffer_input_ids = torch.tensor([], dtype=torch.long)
+    buffer_attention_mask = torch.tensor([], dtype=torch.long)
+
+    for ids, mask in zip(input_ids, attention_mask):
+        if buffer_input_ids.numel() == max_tokens:
+            new_input_ids.append(buffer_input_ids)
+            new_attention_mask.append(buffer_attention_mask)
+            buffer_input_ids = torch.tensor([], dtype=torch.long)
+            buffer_attention_mask = torch.tensor([], dtype=torch.long)
+            buffer_input_ids = torch.cat((buffer_input_ids, ids), dim=0)
+            buffer_attention_mask = torch.cat((buffer_attention_mask, mask), dim=0)
+        elif buffer_input_ids.numel() + ids.numel() <= max_tokens:
+            buffer_input_ids = torch.cat((buffer_input_ids, ids), dim=0)
+            buffer_attention_mask = torch.cat((buffer_attention_mask, mask), dim=0)
+        else:
+            buffer_input_ids = torch.cat(
+                (
+                    buffer_input_ids,
+                    torch.full(
+                        (max_tokens - buffer_input_ids.numel(),),
+                        tokenizer.pad_token_id,
+                        dtype=torch.long,
+                    ),
+                ),
+                dim=0,
+            )
+            buffer_attention_mask = torch.cat(
+                (
+                    buffer_attention_mask,
+                    torch.full(
+                        (max_tokens - buffer_attention_mask.numel(),),
+                        0,
+                        dtype=torch.long,
+                    ),
+                ),
+                dim=0,
+            )
+            new_input_ids.append(buffer_input_ids)
+            new_attention_mask.append(buffer_attention_mask)
+            buffer_input_ids = torch.tensor([], dtype=torch.long)
+            buffer_attention_mask = torch.tensor([], dtype=torch.long)
+
+            buffer_input_ids = torch.cat((buffer_input_ids, ids), dim=0)
+            buffer_attention_mask = torch.cat((buffer_attention_mask, mask), dim=0)
+
+    if buffer_input_ids.numel() > 0:  # for any leftover tokens
+        while buffer_input_ids.numel() < max_tokens:  # make all sequences equal in size
+            buffer_input_ids = torch.cat(
+                (
+                    buffer_input_ids,
+                    torch.full(
+                        (max_tokens - buffer_input_ids.numel(),),
+                        tokenizer.pad_token_id,
+                        dtype=torch.long,
+                    ),
+                ),
+                dim=0,
+            )
+            buffer_attention_mask = torch.cat(
+                (
+                    buffer_attention_mask,
+                    torch.full(
+                        (max_tokens - buffer_attention_mask.numel(),),
+                        0,
+                        dtype=torch.long,
+                    ),
+                ),
+                dim=0,
+            )
+        new_input_ids.append(buffer_input_ids)
+        new_attention_mask.append(buffer_attention_mask)
+
+    ret = {
+        "input_ids": [seq.tolist() for seq in new_input_ids],
+        "labels": [seq.tolist() for seq in new_input_ids],
+        "attention_mask": [seq.tolist() for seq in new_attention_mask],
+    }
+
+    logging.debug(len(ret["input_ids"]))
+    return ret
+
+
+def load_pretraining_dataset(path, tokenizer, max_tokens=2048, seed=42):
+    encode = functools.partial(encode_pretraining, tokenizer, max_tokens)
+    dataset = load_dataset(path, streaming=True, split="train")
+    dataset = dataset.shuffle(seed=seed, buffer_size=10_000)
+    # TODO dynamically figure out which columns/features to remove
+    dataset = dataset.map(encode, batched=True, remove_columns=["text", "meta"])
+    return dataset

src/axolotl/utils/dict.py

@@ -0,0 +1,12 @@
+"""Module containing the DictDefault class"""
+
+from addict import Dict
+
+
+class DictDefault(Dict):
+    """
+    A Dict that returns None instead of returning empty Dict for missing keys.
+    """
+
+    def __missing__(self, key):
+        return None

src/axolotl/utils/models.py

@@ -0,0 +1,461 @@
+"""Module for models and model loading"""
+
+
+import logging
+import math
+import os
+from pathlib import Path
+from typing import TYPE_CHECKING, Optional, Tuple  # noqa: F401
+
+import bitsandbytes as bnb
+import torch
+import transformers
+from optimum.bettertransformer import BetterTransformer
+from transformers import (  # noqa: F401
+    AutoConfig,
+    AutoModelForCausalLM,
+    AutoTokenizer,
+    BitsAndBytesConfig,
+    LlamaConfig,
+    PreTrainedModel,
+    PreTrainedTokenizerBase,
+)
+
+from axolotl.prompt_tokenizers import LLAMA_DEFAULT_PAD_TOKEN
+
+if TYPE_CHECKING:
+    from peft import PeftConfig  # noqa: F401
+
+    from axolotl.utils.dict import DictDefault  # noqa: F401
+
+
+def load_tokenizer(
+    tokenizer_config,
+    tokenizer_type,
+    cfg,
+):
+    use_fast = True  # this is the default
+    if cfg.tokenizer_use_fast is not None:
+        use_fast = cfg.tokenizer_use_fast
+    if tokenizer_type:
+        tokenizer = getattr(transformers, tokenizer_type).from_pretrained(
+            tokenizer_config,
+            trust_remote_code=cfg.trust_remote_code or False,
+            use_fast=use_fast,
+        )
+    else:
+        tokenizer = AutoTokenizer.from_pretrained(
+            tokenizer_config,
+            trust_remote_code=cfg.trust_remote_code or False,
+            use_fast=use_fast,
+        )
+
+    logging.debug(f"EOS: {tokenizer.eos_token_id} / {tokenizer.eos_token}")
+    logging.debug(f"BOS: {tokenizer.bos_token_id} / {tokenizer.bos_token}")
+    logging.debug(f"PAD: {tokenizer.pad_token_id} / {tokenizer.pad_token}")
+    logging.debug(f"UNK: {tokenizer.unk_token_id} / {tokenizer.unk_token}")
+
+    if tokenizer.__class__.__name__ in [
+        "LlamaTokenizer",
+        "LlamaTokenizerFast",
+    ]:
+        tokenizer.pad_token = LLAMA_DEFAULT_PAD_TOKEN
+
+    if tokenizer.__class__.__name__ == "GPTNeoXTokenizerFast":
+        tokenizer.add_special_tokens({"pad_token": "[PAD]"})
+        os.environ["TOKENIZERS_PARALLELISM"] = "false"
+
+    if cfg.special_tokens:
+        for k, val in cfg.special_tokens.items():
+            tokenizer.add_special_tokens({k: val})
+    if cfg.tokens:
+        tokenizer.add_tokens(list(cfg.tokens))
+
+    return tokenizer
+
+
+def load_model(
+    base_model, base_model_config, model_type, tokenizer, cfg, adapter="lora"
+):
+    # type: (str, str, str, PreTrainedTokenizerBase, DictDefault, Optional[str]) -> Tuple[PreTrainedModel, Optional[PeftConfig]]
+    """
+    Load a model from a base model and a model type.
+    """
+
+    # TODO refactor as a kwarg
+    load_in_8bit = cfg.load_in_8bit
+    cfg.is_llama_derived_model = "llama" in base_model or (
+        cfg.model_type and "llama" in cfg.model_type.lower()
+    )
+
+    if cfg.is_llama_derived_model and cfg.flash_attention:
+        if cfg.device not in ["mps", "cpu"] and not cfg.inference:
+            from axolotl.flash_attn import replace_llama_attn_with_flash_attn
+
+            logging.info("patching with flash attention")
+            replace_llama_attn_with_flash_attn()
+    elif cfg.is_llama_derived_model and cfg.xformers_attention:
+        from axolotl.monkeypatch.llama_attn_hijack_xformers import (
+            hijack_llama_attention,
+        )
+
+        logging.info("patching with xformers attention")
+        hijack_llama_attention()
+    elif cfg.is_llama_derived_model and cfg.sdp_attention:
+        from axolotl.monkeypatch.llama_attn_hijack_xformers import (
+            hijack_llama_sdp_attention,
+        )
+
+        logging.info("patching with sdp attention")
+        hijack_llama_sdp_attention()
+    elif cfg.is_llama_derived_model and cfg.landmark_attention:
+        from axolotl.monkeypatch.llama_landmark_attn import (
+            MEM_TOKEN,
+            patch_llama_with_landmark_attn,
+        )
+
+        logging.info("patching with landmark attention")
+        patch_llama_with_landmark_attn()
+
+        # Note: This might overwrite previous additional_special_tokens
+        tokenizer.add_special_tokens({"additional_special_tokens": [MEM_TOKEN]})
+
+    if cfg.is_llama_derived_model and cfg.xpos_rope:
+        from axolotl.monkeypatch.xpos_rope_llama_monkey_patch import (
+            replace_llama_rope_with_xpos_rope,
+        )
+
+        logging.info("patching with xpos rope")
+        replace_llama_rope_with_xpos_rope()
+
+    if cfg.bf16 or cfg.bfloat16:
+        torch_dtype = torch.bfloat16
+    elif cfg.load_in_8bit or cfg.fp16 or cfg.float16:
+        torch_dtype = torch.float16
+    else:
+        torch_dtype = torch.float32
+    try:
+        if cfg.gptq:
+            from alpaca_lora_4bit.monkeypatch.peft_tuners_lora_monkey_patch import (
+                replace_peft_model_with_int4_lora_model,
+            )
+
+            replace_peft_model_with_int4_lora_model()
+    except Exception as err:
+        logging.exception(err)
+        raise err
+
+    try:
+        from peft import prepare_model_for_kbit_training
+    except ImportError:
+        # For backward compatibility
+        from peft import (
+            prepare_model_for_int8_training as prepare_model_for_kbit_training,
+        )
+
+    model_kwargs = {}
+    if cfg.adapter == "qlora" and cfg.load_in_4bit:
+        model_kwargs["quantization_config"] = BitsAndBytesConfig(
+            load_in_4bit=True,
+            llm_int8_threshold=6.0,
+            llm_int8_has_fp16_weight=False,
+            bnb_4bit_compute_dtype=torch_dtype,
+            bnb_4bit_use_double_quant=True,
+            bnb_4bit_quant_type="nf4",
+        )
+    try:
+        if cfg.gptq and cfg.is_llama_derived_model:
+            from alpaca_lora_4bit.autograd_4bit import load_llama_model_4bit_low_ram
+            from huggingface_hub import snapshot_download
+
+            try:
+                snapshot_download_kwargs = {}
+                if cfg.base_model_ignore_patterns:
+                    snapshot_download_kwargs[
+                        "ignore_patterns"
+                    ] = cfg.base_model_ignore_patterns
+                cache_model_path = Path(
+                    snapshot_download(base_model, **snapshot_download_kwargs)
+                )
+                files = (
+                    list(cache_model_path.glob("*.pt"))
+                    + list(cache_model_path.glob("*.safetensors"))
+                    + list(cache_model_path.glob("*.bin"))
+                )
+                if len(files) > 0:
+                    model_path = str(files[0])
+                else:
+                    logging.warning(
+                        "unable to find a cached model file, this will likely fail..."
+                    )
+                    model_path = str(cache_model_path)
+            except Exception:  # pylint: disable=broad-exception-caught
+                model_path = cfg.base_model
+            model, _ = load_llama_model_4bit_low_ram(
+                base_model_config if base_model_config else base_model,
+                model_path,
+                device_map=cfg.device_map,
+                half=cfg.fp16,
+                groupsize=cfg.gptq_groupsize if cfg.gptq_groupsize else -1,
+                is_v1_model=cfg.gptq_model_v1
+                if cfg.gptq_model_v1 is not None
+                else True,
+            )
+            load_in_8bit = False
+        elif cfg.is_llama_derived_model:
+            from transformers import LlamaForCausalLM
+
+            config = LlamaConfig.from_pretrained(base_model_config)
+            model = LlamaForCausalLM.from_pretrained(
+                base_model,
+                config=config,
+                load_in_8bit=cfg.load_in_8bit and cfg.adapter is not None,
+                load_in_4bit=cfg.load_in_4bit and cfg.adapter is not None,
+                torch_dtype=torch_dtype,
+                device_map="auto" if cfg.world_size == 1 else cfg.device_map,
+                **model_kwargs,
+            )
+        # elif model_type == "GPTNeoXForCausalLM" and cfg.flash_attention:
+        #     This is a WIP, still an issue with the backward pass
+        #     RuntimeError: grad can be implicitly created only for scalar outputs
+        #     TODO: try config.sequence_parallel = False
+        #     # https://github.com/HazyResearch/flash-attention/blob/40a25c8ee7465cf547b929cfa2937034e37bfce9/tests/models/test_gpt_neox.py#L12
+        #     # https://github.com/HazyResearch/flash-attention/tree/main/training#model-components
+        #     # add `**kwargs` to https://github.com/HazyResearch/flash-attention/blob/40a25c8ee7465cf547b929cfa2937034e37bfce9/flash_attn/models/gpt.py#L442
+        #     from flash_attn.utils.pretrained import state_dict_from_pretrained
+        #     from flash_attn.models.gpt import GPTLMHeadModel
+        #     from flash_attn.models.gpt_neox import remap_state_dict_hf_gpt_neox, gpt_neox_config_to_gpt2_config
+        #     from transformers import GPTNeoXConfig
+        #     config = gpt_neox_config_to_gpt2_config(GPTNeoXConfig.from_pretrained(base_model))
+        #     config.use_flash_attn = True
+        #     config.fused_bias_fc = True
+        #     config.fused_mlp = True  # GPT-NeoX-20B uses "gelu_fast"
+        #     config.activation_function = "gelu_fast"
+        #     config.fused_dropout_add_ln = True
+        #     # config.residual_in_fp32 = True
+        #
+        #     model: GPTLMHeadModel = GPTLMHeadModel.from_pretrained(
+        #         base_model,
+        #         config,
+        #         dtype=torch_dtype,
+        #         device=cfg.device,
+        #     )
+        #     model.train() # sets to train instead of eval mode
+        elif model_type:
+            model = getattr(transformers, model_type).from_pretrained(
+                base_model,
+                load_in_8bit=cfg.load_in_8bit and cfg.adapter is not None,
+                load_in_4bit=cfg.load_in_4bit and cfg.adapter is not None,
+                torch_dtype=torch_dtype,
+                device_map=cfg.device_map,
+                trust_remote_code=cfg.trust_remote_code or False,
+                **model_kwargs,
+            )
+        else:
+            config = AutoConfig.from_pretrained(
+                base_model,
+                trust_remote_code=cfg.trust_remote_code or False,
+            )
+            # Shouldn't be a problem most of the time. will obviously error if the model doesn't support this
+            # when training starts
+            if (
+                hasattr(config, "max_seq_len")
+                and config.max_seq_len
+                and cfg.sequence_len > config.max_seq_len
+            ):
+                config.max_seq_len = cfg.sequence_len
+                logging.warning(f"increasing context length to {cfg.sequence_len}")
+            elif (
+                hasattr(config, "max_sequence_length")
+                and config.max_sequence_length
+                and cfg.sequence_len > config.max_sequence_length
+            ):
+                config.max_sequence_length = cfg.sequence_len
+                logging.warning(f"increasing context length to {cfg.sequence_len}")
+            model = AutoModelForCausalLM.from_pretrained(
+                base_model,
+                config=config,
+                load_in_8bit=cfg.load_in_8bit and cfg.adapter is not None,
+                load_in_4bit=cfg.load_in_4bit and cfg.adapter is not None,
+                torch_dtype=torch_dtype,
+                device_map=cfg.device_map,
+                trust_remote_code=cfg.trust_remote_code or False,
+                **model_kwargs,
+            )
+    except Exception as err:  # pylint: disable=broad-exception-caught
+        logging.error(
+            "Exception raised attempting to load model, retrying with AutoModelForCausalLM"
+        )
+        logging.exception(err)
+        model = AutoModelForCausalLM.from_pretrained(
+            base_model,
+            load_in_8bit=cfg.load_in_8bit and cfg.adapter is not None,
+            load_in_4bit=cfg.load_in_4bit and cfg.adapter is not None,
+            torch_dtype=torch_dtype,
+            device_map=cfg.device_map,
+            trust_remote_code=cfg.trust_remote_code or False,
+            **model_kwargs,
+        )
+
+    embeddings_len = math.ceil(len(tokenizer) / 32) * 32
+    model.resize_token_embeddings(embeddings_len)
+
+    if (
+        hasattr(model.config, "max_position_embeddings")
+        and model.config.max_position_embeddings
+        and cfg.sequence_len >= model.config.max_position_embeddings
+    ):
+        logging.warning(
+            f"increasing model.config.max_position_embeddings to {cfg.sequence_len}"
+        )
+        model.config.max_position_embeddings = cfg.sequence_len
+
+    if not cfg.gptq and (
+        (cfg.adapter == "lora" and load_in_8bit)
+        or (cfg.adapter == "qlora" and cfg.load_in_4bit)
+    ):
+        logging.info("converting PEFT model w/ prepare_model_for_kbit_training")
+        model = prepare_model_for_kbit_training(
+            model, use_gradient_checkpointing=cfg.gradient_checkpointing
+        )
+
+    model, lora_config = load_adapter(model, cfg, adapter)
+
+    if cfg.ddp and not load_in_8bit:
+        model.to(f"cuda:{cfg.local_rank}")
+
+    if cfg.gptq:
+        # Scales to half
+        logging.info("Fitting 4bit scales and zeros to half")
+        for _, module in model.named_modules():
+            if "Autograd4bitQuantLinear" in str(type(module)) or "Linear4bitLt" in str(
+                type(module)
+            ):
+                if hasattr(module, "is_v1_model") and module.is_v1_model:
+                    module.zeros = module.zeros.half()
+                module.scales = module.scales.half()
+                module.bias = module.bias.half()
+
+    if (
+        torch.cuda.device_count() > 1
+        and int(os.getenv("WORLD_SIZE", "1")) > 1
+        and (cfg.gptq or cfg.load_in_4bit)
+    ):
+        # llama is PROBABLY model parallelizable, but the default isn't that it is
+        # so let's only set it for the 4bit, see
+        # https://github.com/johnsmith0031/alpaca_lora_4bit/blob/08b3fca4a4a9e0d3945be1bab4529f100a428636/finetune.py#L130-L133
+        setattr(model, "is_parallelizable", True)
+        setattr(model, "model_parallel", True)
+
+    requires_grad = []
+    for name, param in model.named_parameters(recurse=True):
+        if param.requires_grad:
+            requires_grad.append(f"{name}: {param.requires_grad}")
+    if len(requires_grad) == 0:
+        logging.warning("there are no parameters that require gradient updates")
+    model.config.use_cache = False
+
+    if cfg.flash_optimum:
+        model = BetterTransformer.transform(model)
+
+    # TODO resume_from_checkpoint handling
+    return model, lora_config
+
+
+def load_adapter(model, cfg, adapter):
+    # type: (PreTrainedModel, DictDefault, Optional[str]) -> Tuple[PreTrainedModel, Optional[PeftConfig]]
+
+    if adapter is None:
+        return model, None
+    if adapter in ["lora", "qlora"]:
+        return load_lora(model, cfg)
+    if adapter == "llama-adapter":
+        return load_llama_adapter(model, cfg)
+
+    raise NotImplementedError(f"{adapter} peft adapter not available")
+
+
+def load_llama_adapter(model, cfg):
+    # type: (PreTrainedModel, DictDefault) -> Tuple[PreTrainedModel, Optional[PeftConfig]]
+    from peft import AdaptionPromptConfig, PeftModel, get_peft_model
+
+    peft_config = AdaptionPromptConfig(
+        adapter_layers=cfg.peft_adapter.layers,  # layers (L)
+        adapter_len=cfg.peft_adapter.len,  # prompt length (K)
+        task_type="CAUSAL_LM",
+    )
+
+    if cfg.lora_model_dir:
+        logging.info("Loading pretained LORA")
+        model = PeftModel.from_pretrained(
+            model,
+            cfg.lora_model_dir,
+            torch_dtype=torch.float16,
+        )
+    else:
+        model = get_peft_model(model, peft_config)
+
+    model.print_trainable_parameters()
+
+    return model, peft_config
+
+
+def find_all_linear_names(bits, model):
+    cls = (
+        bnb.nn.Linear4bit
+        if bits == 4
+        else (bnb.nn.Linear8bitLt if bits == 8 else torch.nn.Linear)
+    )
+    lora_module_names = set()
+    for name, module in model.named_modules():
+        if isinstance(module, cls):
+            names = name.split(".")
+            lora_module_names.add(names[0] if len(names) == 1 else names[-1])
+
+    if "lm_head" in lora_module_names:  # needed for 16-bit
+        lora_module_names.remove("lm_head")
+
+    return list(lora_module_names)
+
+
+def load_lora(model, cfg):
+    # type: (PreTrainedModel, DictDefault) -> Tuple[PreTrainedModel, Optional[PeftConfig]]
+
+    from peft import LoraConfig, PeftModel, get_peft_model
+
+    lora_target_modules = list(cfg.lora_target_modules or [])
+
+    if cfg.lora_target_linear:
+        bits = None
+        if cfg.load_in_4bit:
+            bits = 4
+        elif cfg.load_in_8bit:
+            bits = 8
+
+        linear_names = find_all_linear_names(bits, model)
+        logging.info(f"found linear modules: {repr(linear_names)}")
+        lora_target_modules = list(set(lora_target_modules + linear_names))
+
+    lora_config = LoraConfig(
+        r=cfg.lora_r,
+        lora_alpha=cfg.lora_alpha,
+        target_modules=lora_target_modules,
+        lora_dropout=cfg.lora_dropout,
+        fan_in_fan_out=cfg.lora_fan_in_fan_out,
+        modules_to_save=cfg.lora_modules_to_save if cfg.lora_modules_to_save else None,
+        bias="none",
+        task_type="CAUSAL_LM",
+    )
+
+    if cfg.lora_model_dir:
+        model = PeftModel.from_pretrained(
+            model,
+            cfg.lora_model_dir,
+            is_trainable=not cfg.inference,
+        )
+    else:
+        model = get_peft_model(model, lora_config)
+
+    model.print_trainable_parameters()
+
+    return model, lora_config

src/axolotl/utils/schedulers.py

@@ -0,0 +1,44 @@
+"""Module for custom LRScheduler class"""
+
+from torch.optim.lr_scheduler import LRScheduler
+
+
+class InterpolatingLogScheduler(LRScheduler):
+    """
+    A scheduler that interpolates learning rates in a logarithmic fashion
+    """
+
+    def __init__(self, optimizer, num_steps, min_lr, max_lr, last_epoch=-1):
+        """A scheduler that interpolates learning rates in a logarithmic fashion
+
+        Args:
+        - optimizer: pytorch optimizer
+        - num_steps: int, the number of steps over which to increase from the min_lr to the max_lr
+        - min_lr: float, the minimum learning rate
+        - max_lr: float, the maximum learning rate
+
+        Usage:
+            fc = nn.Linear(1,1)
+            optimizer = optim.Adam(fc.parameters())
+            lr_scheduler = InterpolatingLogScheduler(optimizer, num_steps=400, min_lr=1e-6, max_lr=1e-4)
+        """
+        self.num_steps = num_steps
+        self.min_lr = min_lr
+        self.max_lr = max_lr
+        self.q = (max_lr / min_lr) ** (  # pylint: disable=invalid-name
+            1 / (num_steps - 1)
+        )
+        super().__init__(optimizer, last_epoch)
+
+    def get_lr(self):
+        if self.last_epoch <= 0:
+            lrs = [self.min_lr for base_lr in self.base_lrs]
+        elif self.last_epoch < self.num_steps:
+            lrs = [
+                self.min_lr * (self.q ** (self.last_epoch - 1))
+                for base_lr in self.base_lrs
+            ]
+        else:
+            lrs = [self.max_lr for base_lr in self.base_lrs]
+
+        return lrs

src/axolotl/utils/tokenization.py

@@ -0,0 +1,38 @@
+"""Module for tokenization utilities"""
+
+
+import logging
+
+from termcolor import colored
+
+
+def check_dataset_labels(dataset, tokenizer):
+    # the dataset is already shuffled, so let's just check the first 5 elements
+    for idx in range(5):
+        check_example_labels(dataset[idx], tokenizer)
+
+
+def check_example_labels(example, tokenizer):
+    # Get the input_ids, labels, and attention_mask from the dataset
+    input_ids = example["input_ids"]
+    labels = example["labels"]
+    attention_mask = example["attention_mask"]
+
+    # You can compare the input_ids and labels element-wise
+    # Remember to ignore positions with IGNORE_TOKEN_ID (if you use it) or attention_mask equal to 0
+    colored_tokens = []
+    for _, (input_id, label_id, mask) in enumerate(
+        zip(input_ids, labels, attention_mask)
+    ):
+        decoded_input_token = tokenizer.decode(input_id)
+        # Choose the color based on whether the label has the ignore value or not
+        color = "red" if label_id == -100 else ("yellow" if label_id == 0 else "green")
+        colored_token = colored(decoded_input_token, color) + colored(
+            f"({label_id}, {mask}, {input_id})", "white"
+        )
+        colored_tokens.append(colored_token)
+
+    logging.info(" ".join(colored_tokens))
+    logging.info("\n\n\n")
+
+    return " ".join(colored_tokens)

src/axolotl/utils/trainer.py

@@ -0,0 +1,297 @@
+"""Module containing the Trainer class and related functions"""
+
+import importlib
+import logging
+import math
+import os
+import sys
+from pathlib import Path
+from typing import Optional
+
+import bitsandbytes as bnb
+import torch.cuda
+import transformers
+from torch import nn
+from torch.optim.lr_scheduler import OneCycleLR
+from transformers import EarlyStoppingCallback, Trainer
+from transformers.trainer_pt_utils import get_parameter_names
+
+from axolotl.utils.callbacks import (
+    SaveBetterTransformerModelCallback,
+    SavePeftModelCallback,
+)
+from axolotl.utils.schedulers import InterpolatingLogScheduler
+
+
+class OneCycleLRSchedulerTrainer(Trainer):
+    """
+    Trainer subclass that uses the OneCycleLR scheduler
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.lr_scheduler = None
+
+    def create_scheduler(
+        self,
+        num_training_steps: int,
+        optimizer: Optional[torch.optim.Optimizer] = None,
+    ):
+        optimizer = self.optimizer if optimizer is None else optimizer
+        num_warmup_steps = self.args.get_warmup_steps(num_training_steps)
+        pct_start = num_warmup_steps / num_training_steps
+
+        self.lr_scheduler = OneCycleLR(
+            optimizer,
+            max_lr=self.args.learning_rate,
+            total_steps=num_training_steps,
+            pct_start=pct_start,
+            div_factor=6,
+        )
+
+        return self.lr_scheduler
+
+
+def setup_trainer(cfg, train_dataset, eval_dataset, model, tokenizer):
+    total_num_steps = int(
+        math.ceil(len(train_dataset) * cfg.num_epochs / cfg.batch_size)
+    )
+    warmup_steps = (
+        cfg.warmup_steps
+        if cfg.warmup_steps is not None
+        else min(int(0.03 * total_num_steps), 100)
+    )
+    logging_steps = (
+        cfg.logging_steps
+        if cfg.logging_steps is not None
+        else max(min(int(0.005 * total_num_steps), 10), 1)
+    )
+
+    training_arguments_kwargs = {}
+    if cfg.bf16 == "full":
+        training_arguments_kwargs["bf16_full_eval"] = True
+    else:
+        training_arguments_kwargs["bf16"] = cfg.bf16
+    training_arguments_kwargs["fp16"] = (cfg.fp16 and not cfg.bf16) or False
+    training_arguments_kwargs["tf32"] = cfg.tf32
+    training_arguments_kwargs["warmup_steps"] = warmup_steps
+    training_arguments_kwargs["logging_steps"] = logging_steps
+
+    if cfg.seed:
+        training_arguments_kwargs["seed"] = cfg.seed
+
+    if cfg.gradient_checkpointing:
+        if cfg.gptq:
+            from alpaca_lora_4bit.gradient_checkpointing import (
+                apply_gradient_checkpointing,
+            )
+
+            gradient_checkpointing_ratio = (
+                cfg.gradient_checkpointing_ratio
+                if cfg.gradient_checkpointing_ratio
+                else 1.0
+            )
+            apply_gradient_checkpointing(
+                model, checkpoint_ratio=gradient_checkpointing_ratio
+            )
+        else:
+            training_arguments_kwargs[
+                "gradient_checkpointing"
+            ] = cfg.gradient_checkpointing
+    if cfg.fsdp:
+        training_arguments_kwargs["fsdp"] = cfg.fsdp
+        if cfg.fsdp_config:
+            training_arguments_kwargs["fsdp_config"] = dict(cfg.fsdp_config)
+
+    # deepspeed
+    if (
+        os.environ.get("ACCELERATE_USE_DEEPSPEED") == "true"
+        and torch.cuda.device_count() > 1
+    ):
+        if cfg.deepspeed:
+            training_arguments_kwargs["deepspeed"] = cfg.deepspeed
+        else:
+            # make a guess here
+            # TODO search Path("./") for one
+            training_arguments_kwargs["deepspeed"] = "./ds_config.json"
+
+    if cfg.adam_beta1:
+        training_arguments_kwargs["adam_beta1"] = cfg.adam_beta1
+    if cfg.adam_beta2:
+        training_arguments_kwargs["adam_beta2"] = cfg.adam_beta2
+    if cfg.adam_epsilon:
+        training_arguments_kwargs["adam_epsilon"] = cfg.adam_epsilon
+    if cfg.max_grad_norm:
+        training_arguments_kwargs["max_grad_norm"] = cfg.max_grad_norm
+
+    if cfg.push_to_hub_model_id:
+        training_arguments_kwargs["push_to_hub_model_id"] = cfg.push_to_hub_model_id
+        training_arguments_kwargs["push_to_hub"] = True
+
+    training_args = transformers.TrainingArguments(
+        per_device_train_batch_size=cfg.micro_batch_size,
+        per_device_eval_batch_size=cfg.eval_batch_size
+        if cfg.eval_batch_size is not None
+        else cfg.micro_batch_size,
+        gradient_accumulation_steps=cfg.gradient_accumulation_steps,
+        eval_accumulation_steps=cfg.gradient_accumulation_steps,
+        num_train_epochs=cfg.num_epochs,
+        learning_rate=cfg.learning_rate,
+        evaluation_strategy="steps",
+        save_strategy="steps" if cfg.save_steps else "epoch",
+        eval_steps=cfg.eval_steps,
+        save_steps=cfg.save_steps,
+        output_dir=cfg.output_dir,
+        save_total_limit=3,
+        load_best_model_at_end=(
+            cfg.load_best_model_at_end is not False
+            and cfg.val_set_size > 0
+            and cfg.save_steps
+            and cfg.save_steps % cfg.eval_steps == 0
+            and cfg.load_in_8bit is not True
+        )
+        or False,
+        ddp_find_unused_parameters=False if cfg.ddp else None,
+        group_by_length=cfg.group_by_length,
+        report_to="wandb" if cfg.use_wandb else None,
+        run_name=cfg.wandb_run_id if cfg.use_wandb else None,
+        optim=cfg.optimizer if cfg.optimizer else "adamw_hf",
+        lr_scheduler_type=cfg.lr_scheduler
+        if cfg.lr_scheduler and cfg.lr_scheduler not in ("one_cycle", "log_sweep")
+        else "cosine",
+        weight_decay=cfg.weight_decay if cfg.weight_decay is not None else 0.0,
+        **training_arguments_kwargs,
+    )
+
+    trainer_kwargs = {}
+
+    if cfg.optimizer == "adamw_anyprecision":
+        if Path(cfg.torchdistx_path).exists():
+            sys.path.append(cfg.torchdistx_path)
+            importlib.import_module("torchdistx")
+    if (
+        cfg.optimizer == "adamw_bnb_8bit"
+        and not cfg.gptq
+        and "deepspeed" not in training_arguments_kwargs
+        and not cfg.fsdp
+    ):
+        decay_parameters = get_parameter_names(model, [nn.LayerNorm])
+        decay_parameters = [name for name in decay_parameters if "bias" not in name]
+        optimizer_grouped_parameters = [
+            {
+                "params": [
+                    p
+                    for n, p in model.named_parameters()
+                    if (n in decay_parameters and p.requires_grad)
+                ],
+                "weight_decay": training_args.weight_decay,
+            },
+            {
+                "params": [
+                    p
+                    for n, p in model.named_parameters()
+                    if (n not in decay_parameters and p.requires_grad)
+                ],
+                "weight_decay": 0.0,
+            },
+        ]
+
+        optimizer = bnb.optim.Adam8bit(
+            optimizer_grouped_parameters,
+            betas=(training_args.adam_beta1, training_args.adam_beta2),
+            eps=training_args.adam_epsilon,
+            lr=training_args.learning_rate,
+        )
+
+        if cfg.lr_scheduler == "one_cycle":
+            lr_scheduler_kwargs = (
+                cfg.lr_scheduler_kwargs if cfg.lr_scheduler_kwargs else {}
+            )
+            lr_scheduler = OneCycleLR(
+                optimizer,
+                cfg.learning_rate,
+                total_steps=total_num_steps,
+                epochs=cfg.num_epochs,
+                div_factor=cfg.lr_div_factor if cfg.lr_div_factor else 6,
+                **lr_scheduler_kwargs,
+            )
+        elif cfg.lr_scheduler == "log_sweep":
+            lr_scheduler = InterpolatingLogScheduler(
+                optimizer,
+                cfg.warmup_steps,
+                cfg.log_sweep_min_lr if cfg.log_sweep_min_lr else 1e-10,
+                cfg.log_sweep_max_lr if cfg.log_sweep_max_lr else 10,
+            )
+        else:
+            lr_scheduler = transformers.get_cosine_schedule_with_warmup(
+                optimizer,
+                training_args.warmup_steps,
+                total_num_steps,
+            )
+        trainer_kwargs["optimizers"] = (optimizer, lr_scheduler)
+
+    callbacks = []
+    # TODO on_save callback to sync checkpoints to GCP/AWS in background
+    if cfg.early_stopping_patience:
+        early_stop_cb = EarlyStoppingCallback(
+            cfg.early_stopping_patience,
+        )
+        callbacks.append(early_stop_cb)
+
+    if cfg.local_rank == 0 and cfg.adapter in [
+        "lora",
+        "qlora",
+    ]:  # only save in rank 0
+        callbacks.append(SavePeftModelCallback)
+
+    if hasattr(model, "use_bettertransformer") and model.use_bettertransformer is True:
+        callbacks.append(SaveBetterTransformerModelCallback)
+
+    data_collator_kwargs = {
+        "padding": True,
+    }
+    if cfg.collator_pad_to_longest:
+        data_collator_kwargs["padding"] = "longest"
+    else:
+        data_collator_kwargs["pad_to_multiple_of"] = 8
+
+    if cfg.is_llama_derived_model and cfg.landmark_attention:
+        from functools import partial
+
+        from axolotl.monkeypatch.llama_landmark_attn import (
+            add_mem_tokens,
+            get_mem_id,
+            set_model_mem_id,
+        )
+
+        set_model_mem_id(model, tokenizer)
+
+        logging.info("Adding landmark attention tokens to dataset")
+
+        for dataset in [train_dataset, eval_dataset]:
+            dataset = dataset.map(
+                partial(add_mem_tokens, mem_freq=50, mem_id=get_mem_id(tokenizer)),
+                batched=False,
+                num_proc=32,
+            )
+
+    trainer_cls = (
+        OneCycleLRSchedulerTrainer
+        if cfg.lr_scheduler == "one_cycle" and (cfg.fsdp or cfg.adapter == "qlora")
+        else transformers.Trainer
+    )
+    trainer = trainer_cls(
+        model=model,
+        train_dataset=train_dataset,
+        eval_dataset=eval_dataset,
+        args=training_args,
+        data_collator=transformers.DataCollatorForSeq2Seq(
+            tokenizer,
+            return_tensors="pt",
+            **data_collator_kwargs,
+        ),
+        callbacks=callbacks,
+        **trainer_kwargs,
+    )
+
+    return trainer

src/axolotl/utils/validation.py

@@ -0,0 +1,105 @@
+"""Module for validating config files"""
+
+import logging
+
+import torch
+
+
+def validate_config(cfg):
+    if cfg.gradient_accumulation_steps and cfg.batch_size:
+        raise ValueError(
+            "please set only one of gradient_accumulation_steps or batch_size"
+        )
+    if cfg.batch_size:
+        logging.warning(
+            "%s\n%s",
+            "batch_size is not recommended. Please use gradient_accumulation_steps instead.",
+            "To calculate the equivalent gradient_accumulation_steps, divide batch_size / micro_batch_size / number of gpus.",
+        )
+    if cfg.load_4bit:
+        raise ValueError(
+            "cfg.load_4bit parameter has been deprecated and replaced by cfg.gptq"
+        )
+
+    if cfg.adapter == "qlora":
+        if cfg.merge_lora:
+            # can't merge qlora if loaded in 8bit or 4bit
+            if cfg.load_in_8bit:
+                raise ValueError("Can't merge qlora if loaded in 8bit")
+
+            if cfg.gptq:
+                raise ValueError("Can't merge qlora if gptq")
+
+            if cfg.load_in_4bit:
+                raise ValueError("Can't merge qlora if loaded in 4bit")
+
+        else:
+            if cfg.load_in_8bit:
+                raise ValueError("Can't load qlora in 8bit")
+
+            if cfg.gptq:
+                raise ValueError("Can't load qlora if gptq")
+
+            if not cfg.load_in_4bit:
+                raise ValueError("Require cfg.load_in_4bit to be True for qlora")
+
+    if not cfg.load_in_8bit and cfg.adapter == "lora":
+        logging.warning("We recommend setting `load_in_8bit: true` for LORA finetuning")
+
+    if cfg.trust_remote_code:
+        logging.warning(
+            "`trust_remote_code` is set to true. Please make sure that you reviewed the remote code/model."
+        )
+
+    if cfg.push_dataset_to_hub and cfg.hf_use_auth_token is not True:
+        raise ValueError(
+            "Require cfg.hf_use_auth_token to be True for push_dataset_to_hub"
+        )
+
+    if (cfg.base_model and "falcon" in cfg.base_model.lower()) and cfg.fsdp:
+        raise ValueError("FSDP is not supported for falcon models")
+
+    if (
+        cfg.base_model and "mpt" in cfg.base_model.lower()
+    ) and cfg.gradient_checkpointing:
+        raise ValueError("gradient_checkpointing is not supported for MPT models")
+
+    if cfg.flash_optimum is True:
+        if cfg.adapter:
+            logging.warning(
+                "BetterTransformers probably doesn't work with PEFT adapters"
+            )
+        if cfg.fp16 or cfg.bf16:
+            raise ValueError("AMP is not supported with BetterTransformer")
+        if cfg.float16 is not True and cfg.bloat16 is not True:
+            logging.warning(
+                "You should probably set bfloat16 or float16 to true to "
+                "load the model in float16 for BetterTransformers"
+            )
+        if int(torch.__version__.split(".")[0]) < 2:
+            logging.warning("torch>=2.0.0 required")
+            raise ValueError(
+                f"flash_optimum for BetterTransformers may not be used with {torch.__version__}"
+            )
+
+    if cfg.pretraining_dataset and cfg.group_by_length:
+        logging.warning(
+            "You probably want to disable group_by_length as it will force a streamed dataset to download completely."
+        )
+
+    if any([cfg.adamw_beta1, cfg.adamw_beta2, cfg.adamw_epsilon]) and (
+        not cfg.optimizer or "adamw" not in cfg.optimizer
+    ):
+        logging.warning("adamw hyperparameters found, but no adamw optimizer set")
+
+    # TODO
+    # MPT 7b
+    # https://github.com/facebookresearch/bitsandbytes/issues/25
+    # no 8bit adaAmw w bf16
+
+    # GPT-NeoX
+    # evals broken when extending context len
+    # File "/root/miniconda3/envs/py3.9/lib/python3.9/site-packages/transformers/models/gpt_neox/modeling_gpt_neox.py", line 162, in forward                        attn_output, attn_weights = self._attn(query, key, value, attention_mask, head_mask)
+    # File "/root/miniconda3/envs/py3.9/lib/python3.9/site-packages/optimum/bettertransformer/models/attention.py", line 74, in gpt2_wrapped_scaled_dot_product
+    # attention_mask = causal_mask + attention_mask
+    # RuntimeError: The size of tensor a (2048) must match the size of tensor b (8132) at non-singleton dimension 3

src/axolotl/utils/wandb.py

@@ -0,0 +1,19 @@
+"""Module for wandb utilities"""
+
+import os
+
+
+def setup_wandb_env_vars(cfg):
+    if cfg.wandb_mode and cfg.wandb_mode == "offline":
+        os.environ["WANDB_MODE"] = cfg.wandb_mode
+    elif cfg.wandb_project and len(cfg.wandb_project) > 0:
+        os.environ["WANDB_PROJECT"] = cfg.wandb_project
+        cfg.use_wandb = True
+        if cfg.wandb_watch and len(cfg.wandb_watch) > 0:
+            os.environ["WANDB_WATCH"] = cfg.wandb_watch
+        if cfg.wandb_log_model and len(cfg.wandb_log_model) > 0:
+            os.environ["WANDB_LOG_MODEL"] = cfg.wandb_log_model
+        if cfg.wandb_run_id and len(cfg.wandb_run_id) > 0:
+            os.environ["WANDB_RUN_ID"] = cfg.wandb_run_id
+    else:
+        os.environ["WANDB_DISABLED"] = "true"

tests/fixtures/alpaca/alpaca.json

@@ -0,0 +1,12 @@
+[
+  {
+    "instruction": "You will be given a series of words. Output these words in reverse order, with each word on its own line.",
+    "input": "Words: ['Hello', 'world'].",
+    "output": "['world', 'Hello']"
+  },
+  {
+    "instruction": "In this task, you're given a short description of an event. Your job is to order the steps involved in the event from first to last. Note that there may be multiple correct answers for each event.",
+    "input": "Description: A man walks into a bar and orders a drink. He pays for his drink and leaves the bar.",
+    "output": "1. The man walks into the bar.\n2. He orders a drink.\n3. He pays for his drink.\n4. He leaves the bar."
+  }
+]

tests/fixtures/conversation.json

@@ -0,0 +1,52 @@
+{
+  "conversations": [
+    {
+      "from": "human",
+      "value": "how could i implement a minesweeper algorithm that utilises algebraic topology to solve boards?"
+    },
+    {
+      "from": "gpt",
+      "value": "It's possible to use algebraic topology to solve Minesweeper, but it would likely be quite complex. Algebraic topology is a branch of mathematics that studies the properties of spaces that are preserved under continuous deformations, such as stretching and bending. In the context of Minesweeper, this could be used to analyze the structure of the game board and determine the locations of mines based on certain topological invariants. However, implementing this would require a deep understanding of both algebraic topology and the game of Minesweeper. It's not something that can be explained in a simple answer."
+    },
+    {
+      "from": "human",
+      "value": "could you give me the gist of how it could though?"
+    },
+    {
+      "from": "gpt",
+      "value": "Sure. The basic idea would be to use algebraic topology to analyze the structure of the Minesweeper game board and determine the locations of mines based on certain topological invariants. This could involve representing the game board as a topological space and using topological invariants such as homotopy groups and Betti numbers to infer information about the location of mines on the board. However, as I mentioned before, implementing this would require a deep understanding of both algebraic topology and the game of Minesweeper, so it's not something that can be explained in a simple answer."
+    },
+    {
+      "from": "human",
+      "value": "give me a specific example"
+    },
+    {
+      "from": "gpt",
+      "value": "As a specific example, suppose we have a Minesweeper game board with the following configuration:\n```\n 1  1  1  1  1\n 1  ?  ?  ?  1\n 1  ?  ?  ?  1\n 1  ?  ?  ?  1\n 1  1  1  1  1\n```\nIn this case, the game board can be represented as a topological space, where each square on the board corresponds to a point in the space. The mines on the board can be thought of as \"holes\" in the space, which are points that are not part of the space. Using algebraic topology, we can analyze the structure of this space and determine the locations of the mines based on certain topological invariants.\n\nFor instance, one way to do this is to compute the homotopy groups of the space. Homotopy groups are a way of describing the holes in a space, and they can be used to infer information about the locations of the mines on the Minesweeper board. In the case of the board above, we can compute the first homotopy group of the space, which describes the number and \"shape\" of the holes in the space. In this case, the first homotopy group would tell us that there are four holes in the space, arranged in a square shape.\n\nBased on this information, we can infer that the mines in the game are located in the four corners of the board, as shown below:\n```\n 1  1  1  1  1\n 1  M  ?  ?  1\n 1  ?  ?  ?  1\n 1  ?  ?  M  1\n 1  1  1  1  1\n```\nOf course, this is just one example of how algebraic topology could be used to solve Minesweeper. There are many other possible approaches, and the specific details of how to implement them would depend on the specific game board and the topological invariants being used. As I mentioned before, implementing a solution like this would require a deep understanding of both algebraic topology and the game of Minesweeper."
+    },
+    {
+      "from": "human",
+      "value": "assuming i was someone with a deep understanding of both, could you explain the same concept to me with that assumption?"
+    },
+    {
+      "from": "gpt",
+      "value": "Sure. Assuming you have a deep understanding of both algebraic topology and the game of Minesweeper, here is how you might use algebraic topology to solve a Minesweeper game board:\n\n1. Represent the Minesweeper game board as a topological space. Each square on the board corresponds to a point in the space, and the mines on the board can be thought of as \"holes\" in the space, which are points that are not part of the space.\n2. Use topological invariants to infer information about the locations of the mines on the board. There are many possible topological invariants that could be used for this, such as homotopy groups, Betti numbers, or persistent homology.\n3. Use the information inferred from the topological invariants to determine the locations of the mines on the board. For example, if the first homotopy group of the space indicates that there are four holes in the space, arranged in a square shape, then we can infer that the mines are located in the four corners of the board.\n4. Use this information to solve the Minesweeper game. Once you know the locations of the mines, you can use this information to avoid them and clear the board.\n\nOf course, this is just one possible approach to using algebraic topology to solve Minesweeper. There may be other ways to do it, depending on the specific game board and the topological invariants being used. However, this should give you a general idea of how algebraic topology can be used to solve Minesweeper."
+    },
+    {
+      "from": "human",
+      "value": "how would i do it with persistent homology or any of the other topological invariants?"
+    },
+    {
+      "from": "gpt",
+      "value": "To use persistent homology or any of the other topological invariants to solve a Minesweeper game board, you would follow the same basic steps as described above, but with some differences in the specific details.\n\nFor instance, here is how you might use persistent homology to solve a Minesweeper game board:\n\n1. Represent the Minesweeper game board as a topological space. Each square on the board corresponds to a point in the space, and the mines on the board can be thought of as \"holes\" in the space, which are points that are not part of the space.\n2. Compute the persistent homology of the space. Persistent homology is a way of describing the holes in a space over time, and it can be used to infer information about the locations of the mines on the Minesweeper board.\n3. Use the information from the persistent homology to determine the locations of the mines on the board. For example, if the persistent homology indicates that there are four holes in the space, arranged in a square shape, then we can infer that the mines are located in the four corners of the board.\n4. Use this information to solve the Minesweeper game. Once you know the locations of the mines, you can use this information to avoid them and clear the board.\n\nOf course, this is just one possible approach to using persistent homology to solve Minesweeper. There may be other ways to do it, depending on the specific game board and the specific details of the persistent homology computation. However, this should give you a general idea of how persistent homology can be used to solve Minesweeper."
+    },
+    {
+      "from": "human",
+      "value": "how could i implement this in python?"
+    },
+    {
+      "from": "gpt",
+      "value": "To implement a Minesweeper solver that uses algebraic topology in Python, you would need to do the following:\n\n1. Install a Python package for algebraic topology. There are several options available, such as Gudhi, Dionysus, or scikit-tda. These packages provide algorithms and data structures for computing topological invariants, such as homotopy groups, Betti numbers, or persistent homology.\n2. Represent the Minesweeper game board as a topological space. This could be done using a data structure provided by the algebraic topology package, such as a simplicial complex or a Vietoris-Rips complex. Each square on the board would correspond to a point in the space, and the mines on the board would be represented as \"holes\" in the space.\n3. Compute the topological invariants of the space. This could be done using algorithms provided by the algebraic topology package. For example, you could compute the persistent homology of the space using the persistent homology algorithm.\n4. Use the information from the topological invariants to determine the locations of the mines on the board. This could be done using a combination of mathematical reasoning and programming. For example, if the persistent homology indicates that there are four holes in the space, arranged in a square shape, then you could use a piece of code to infer that the mines are located in the four corners of the board.\n5. Use this information to solve the Minesweeper game. Once you know the locations of the mines, you could use a piece of code to automatically click on the safe squares and clear the board.\n\nOf course, this is just one possible approach to implementing a Minesweeper solver that uses algebraic topology in Python. There may be other ways to do it, depending on the specific details of the implementation. However, this should give you a general idea of how to get started with such a project."
+    }
+  ]
+}

tests/test_dict.py

@@ -0,0 +1,99 @@
+"""Module for testing DictDefault class"""
+
+
+import unittest
+
+import pytest
+
+from axolotl.utils.dict import DictDefault
+
+
+class DictDefaultTest(unittest.TestCase):
+    """
+    Test DictDefault class
+    """
+
+    def test_dict_default(self):
+        cfg = DictDefault(
+            {
+                "key_a": {"key_b": "value_a"},
+                "key_c": "value_c",
+                "key_d": ["value_d", "value_e"],
+            }
+        )
+
+        assert (
+            cfg.key_a.key_b == "value_a"
+        ), "DictDefault should return value for existing nested keys"
+
+        assert (
+            cfg.key_c == "value_c"
+        ), "DictDefault should return value for existing keys"
+
+        assert (
+            cfg.key_d[0] == "value_d"
+        ), "DictDefault should return value for existing keys in list"
+
+        assert (
+            "value_e" in cfg.key_d
+        ), "DictDefault should support in operator for existing keys in list"
+
+    def test_dict_or_operator(self):
+        cfg = DictDefault(
+            {
+                "key_a": {"key_b": "value_a"},
+                "key_c": "value_c",
+                "key_d": ["value_d", "value_e"],
+                "key_f": "value_f",
+            }
+        )
+
+        cfg = cfg | DictDefault(  # pylint: disable=unsupported-binary-operation
+            {"key_a": {"key_b": "value_b"}, "key_f": "value_g"}
+        )
+
+        assert (
+            cfg.key_a.key_b == "value_b"
+        ), "DictDefault should support OR operator for existing nested keys"
+
+        assert cfg.key_c == "value_c", "DictDefault should not delete existing key"
+
+        assert cfg.key_d == [
+            "value_d",
+            "value_e",
+        ], "DictDefault should not overwrite existing keys in list"
+
+        assert (
+            cfg.key_f == "value_g"
+        ), "DictDefault should support OR operator for existing key"
+
+    def test_dict_missingkey(self):
+        cfg = DictDefault({})
+
+        assert cfg.random_key is None, "DictDefault should return None for missing keys"
+
+    def test_dict_nested_missingparentkey(self):
+        """
+        Due to subclassing Dict, DictDefault will error if we try to access a nested key whose parent key does not exist.
+        """
+        cfg = DictDefault({})
+
+        with pytest.raises(
+            AttributeError,
+            match=r"'NoneType' object has no attribute 'another_random_key'",
+        ):
+            cfg.random_key.another_random_key = "value"
+
+    def test_dict_shorthand_assignment(self):
+        """
+        Shorthand assignment is said to not be supported if subclassed. However, their example raises error instead of None.
+        This test ensures that it is supported for current implementation.
+
+        Ref: https://github.com/mewwts/addict#default-values
+        """
+
+        cfg = DictDefault({"key_a": {"key_b": "value_a"}})
+
+        cfg.key_a.key_b = "value_b"
+
+        assert cfg.key_a.key_b == "value_b", "Shorthand assignment should be supported"

tests/test_packed_dataset.py

@@ -0,0 +1,65 @@
+"""Module for testing dataset sequence packing"""
+
+import unittest
+from pathlib import Path
+
+from datasets import Dataset, load_dataset
+from transformers import AutoTokenizer
+
+from axolotl.datasets import ConstantLengthDataset, TokenizedPromptDataset
+from axolotl.prompt_tokenizers import AlpacaPromptTokenizingStrategy
+from axolotl.prompters import AlpacaPrompter
+
+
+class TestPacking(unittest.TestCase):
+    """
+    Test class for packing dataset sequences
+    """
+
+    def setUp(self) -> None:
+        # pylint: disable=duplicate-code
+        self.tokenizer = AutoTokenizer.from_pretrained("huggyllama/llama-7b")
+        self.tokenizer.add_special_tokens(
+            {
+                "bos_token": "<s>",
+                "eos_token": "</s>",
+                "unk_token": "<unk>",
+            }
+        )
+
+    def test_resets_attention(self):
+        prompter = AlpacaPrompter("chat")
+        strat = AlpacaPromptTokenizingStrategy(
+            prompter,
+            self.tokenizer,
+            False,
+            2048,
+        )
+        dateset = load_dataset(
+            "json",
+            data_files=str(Path(__file__).parent / "fixtures/alpaca/alpaca.json"),
+        )["train"]
+        dataset = Dataset.from_list(list(TokenizedPromptDataset(strat, dateset)))
+
+        constant_len_dataset = ConstantLengthDataset(
+            self.tokenizer,
+            [dataset],
+            seq_length=2048,
+        )
+        packed_dataset = Dataset.from_list(list(constant_len_dataset))
+        example = packed_dataset[0]
+        next_bos_index = (
+            example["input_ids"][1:].index(self.tokenizer.bos_token_id) + 1
+        )  # add one since we sliced
+
+        # first example doesn't have mask reset
+        assert example["input_ids"][0] == self.tokenizer.bos_token_id
+        assert example["attention_mask"][0] == 1
+
+        # but subsequent one does
+        assert example["input_ids"][next_bos_index] == self.tokenizer.bos_token_id
+        assert example["attention_mask"][next_bos_index] == 0
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_prompt_tokenizers.py

@@ -0,0 +1,138 @@
+"""Module for testing prompt tokenizers."""
+import json
+import logging
+import unittest
+from pathlib import Path
+
+from transformers import AutoTokenizer
+
+from axolotl.prompt_strategies.alpaca_chat import NoSystemPrompter
+from axolotl.prompt_strategies.alpaca_w_system import (
+    InstructionWSystemPromptTokenizingStrategy,
+    SystemDataPrompter,
+)
+from axolotl.prompt_tokenizers import (
+    AlpacaPromptTokenizingStrategy,
+    ShareGPTPromptTokenizingStrategy,
+)
+from axolotl.prompters import AlpacaPrompter, PromptStyle, ShareGPTPrompter
+
+logging.basicConfig(level="INFO")
+
+
+class TestPromptTokenizationStrategies(unittest.TestCase):
+    """
+    Test class for prompt tokenization strategies.
+    """
+
+    def setUp(self) -> None:
+        # pylint: disable=duplicate-code
+        self.tokenizer = AutoTokenizer.from_pretrained("huggyllama/llama-7b")
+        self.tokenizer.add_special_tokens(
+            {
+                "bos_token": "<s>",
+                "eos_token": "</s>",
+                "unk_token": "<unk>",
+            }
+        )
+
+    def test_sharegpt_integration(self):
+        with open(
+            Path(__file__).parent / "fixtures/conversation.json", encoding="utf-8"
+        ) as fin:
+            data = fin.read()
+            conversation = json.loads(data)
+        with open(
+            Path(__file__).parent / "fixtures/conversation.tokenized.json",
+            encoding="utf-8",
+        ) as fin:
+            data = fin.read()
+            tokenized_conversation = json.loads(data)
+        prompter = ShareGPTPrompter("chat")
+        strat = ShareGPTPromptTokenizingStrategy(
+            prompter,
+            self.tokenizer,
+            False,
+            2048,
+        )
+        example = strat.tokenize_prompt(conversation)
+        for fields in ["input_ids", "attention_mask", "labels"]:
+            self.assertEqual(len(example[fields]), len(tokenized_conversation[fields]))
+            self.assertEqual(example[fields], tokenized_conversation[fields])
+
+    def test_no_sys_prompt(self):
+        """
+        tests the interface between the user and assistant parts
+        """
+        prompter = NoSystemPrompter()
+        # pylint: disable=duplicate-code
+        strat = AlpacaPromptTokenizingStrategy(
+            prompter,
+            self.tokenizer,
+            False,
+            2048,
+        )
+        sample = {
+            "instruction": "hello cruel. lorem ipsum dolor sit amet.",
+            "output": "world!",
+        }
+        example = strat.tokenize_prompt(sample)
+        world_idx = example["input_ids"].index(3186)
+        assert example["labels"][world_idx] == 3186
+        assert example["labels"][world_idx - 1] == -100
+
+    def test_alpaca(self):
+        """
+        tests the interface between the user and assistant parts
+        """
+        # pylint: disable=duplicate-code
+        prompter = AlpacaPrompter()
+        strat = AlpacaPromptTokenizingStrategy(
+            prompter,
+            self.tokenizer,
+            False,
+            2048,
+        )
+        sample = {"instruction": "hello!", "output": "Hi! How can I help?"}
+        example = strat.tokenize_prompt(sample)
+        world_idx = example["input_ids"].index(6324)
+        assert example["labels"][world_idx] == 6324
+        assert example["labels"][world_idx - 1] == -100
+
+
+class InstructionWSystemPromptTokenizingStrategyTest(unittest.TestCase):
+    """
+    Test class for prompt tokenization strategies with sys prompt from the dataset
+    """
+
+    def setUp(self) -> None:
+        # pylint: disable=duplicate-code
+        self.tokenizer = AutoTokenizer.from_pretrained("huggyllama/llama-7b")
+        self.tokenizer.add_special_tokens(
+            {
+                "bos_token": "<s>",
+                "eos_token": "</s>",
+                "unk_token": "<unk>",
+            }
+        )
+
+    def test_system_alpaca(self):
+        prompter = SystemDataPrompter(PromptStyle.CHAT.value)
+        strat = InstructionWSystemPromptTokenizingStrategy(
+            prompter,
+            self.tokenizer,
+            False,
+            2048,
+        )
+        sample = {
+            "system": "use cot",
+            "instruction": "hello!",
+            "output": "Hi! How can I help?",
+        }
+        example = strat.tokenize_prompt(sample)
+        assert example["input_ids"][0:3] == [1, 671, 20118]  # <s>use cot
+        assert example["input_ids"][3] == 11889  # USER
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_prompters.py

@@ -0,0 +1,124 @@
+"""Module testing prompters"""
+
+import unittest
+
+from axolotl.prompt_strategies.alpaca_w_system import SystemDataPrompter
+from axolotl.prompters import (
+    AlpacaPrompter,
+    MultipleChoiceExplainPrompter,
+    PromptStyle,
+    UnpromptedPrompter,
+)
+
+
+class AlpacaPrompterTest(unittest.TestCase):
+    """
+    Test AlpacaPrompter
+    """
+
+    def test_prompt_style_w_none(self):
+        prompter = AlpacaPrompter(prompt_style=None)
+        res = next(prompter.build_prompt("tell me a joke"))
+        # just testing that it uses instruct style
+        assert "### Instruction:" in res
+
+    def test_prompt_style_w_instruct(self):
+        prompter = AlpacaPrompter(prompt_style=PromptStyle.INSTRUCT.value)
+        res = next(
+            prompter.build_prompt("tell me a joke about the following", "alpacas")
+        )
+        assert "Below is an instruction" in res
+        assert "### Instruction:" in res
+        assert "### Input:" in res
+        assert "alpacas" in res
+        assert "### Response:" in res
+        assert "USER:" not in res
+        assert "ASSISTANT:" not in res
+        res = next(prompter.build_prompt("tell me a joke about the following"))
+        assert "Below is an instruction" in res
+        assert "### Instruction:" in res
+        assert "### Input:" not in res
+        assert "### Response:" in res
+        assert "USER:" not in res
+        assert "ASSISTANT:" not in res
+
+    def test_prompt_style_w_chat(self):
+        prompter = AlpacaPrompter(prompt_style=PromptStyle.CHAT.value)
+        res = next(
+            prompter.build_prompt("tell me a joke about the following", "alpacas")
+        )
+        assert "Below is an instruction" in res
+        assert "### Instruction:" not in res
+        assert "### Input:" not in res
+        assert "alpacas" in res
+        assert "### Response:" not in res
+        assert "USER:" in res
+        assert "ASSISTANT:" in res
+        res = next(prompter.build_prompt("tell me a joke about the following"))
+        assert "Below is an instruction" in res
+        assert "### Instruction:" not in res
+        assert "### Input:" not in res
+        assert "### Response:" not in res
+        assert "USER:" in res
+        assert "ASSISTANT:" in res
+
+    def test_system_prompt(self):
+        prompter = SystemDataPrompter(prompt_style=PromptStyle.CHAT.value)
+        res = next(
+            prompter.build_prompt_w_system(
+                "use cot", "tell me a joke about the following", "alpacas"
+            )
+        )
+        assert "use cot" in res
+        assert res.startswith("use cot")
+        assert "### Instruction:" not in res
+        assert "### Input:" not in res
+        assert "alpacas" in res
+        assert "### Response:" not in res
+        assert "USER:" in res
+        assert "ASSISTANT:" in res
+
+
+class UnpromptedPrompterTest(unittest.TestCase):
+    """
+    Test class for UnpromptedPrompter with no system prompts
+    """
+
+    def test_prompt_style_w_none(self):
+        prompter = UnpromptedPrompter(prompt_style=None)
+        res = next(prompter.build_prompt("tell me a joke"))
+        assert "### Instruction:" in res
+        assert "tell me a joke" in res
+        assert res.startswith("###")
+
+    def test_prompt_style_w_instruct(self):
+        prompter = UnpromptedPrompter(prompt_style=PromptStyle.INSTRUCT.value)
+        res = next(
+            prompter.build_prompt("tell me a joke about the following", "alpacas")
+        )
+        assert "### Instruction:" in res
+        assert "tell me a joke" in res
+        assert res.startswith("###")
+
+    def test_prompt_style_w_chat(self):
+        prompter = UnpromptedPrompter(prompt_style=PromptStyle.CHAT.value)
+        res = next(
+            prompter.build_prompt("tell me a joke about the following", "alpacas")
+        )
+        assert "USER:" in res
+        assert "tell me a joke" in res
+        assert res.startswith("USER:")
+
+
+class MultipleChoiceExplainPrompterTest(unittest.TestCase):
+    """
+    Test class for MultipleChoiceExplainPrompter
+    """
+
+    def test_prompt_style_w_chat(self):
+        prompter = MultipleChoiceExplainPrompter(prompt_style=PromptStyle.CHAT.value)
+        res = next(prompter.build_prompt("choose one", "- A\n- B\n- C", "C"))
+        assert "USER:" in res
+        assert "choose one" in res
+        assert "Choose the answer that best answers the question." in res
+        assert "- A\n- B\n- C" in res

tests/test_tokenizers.py

@@ -0,0 +1,31 @@
+"""
+Test cases for the tokenizer loading
+"""
+import unittest
+
+from axolotl.utils.dict import DictDefault
+from axolotl.utils.models import load_tokenizer
+
+
+class TestTokenizers(unittest.TestCase):
+    """
+    test class for the load_tokenizer fn
+    """
+
+    def test_default_use_fast(self):
+        cfg = DictDefault({})
+        tokenizer = load_tokenizer("huggyllama/llama-7b", None, cfg)
+        assert "Fast" in tokenizer.__class__.__name__
+
+    def test_dont_use_fast(self):
+        cfg = DictDefault(
+            {
+                "tokenizer_use_fast": False,
+            }
+        )
+        tokenizer = load_tokenizer("huggyllama/llama-7b", None, cfg)
+        assert "Fast" not in tokenizer.__class__.__name__
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_validation.py

@@ -0,0 +1,315 @@
+"""Module for testing the validation module"""
+
+import logging
+import unittest
+from typing import Optional
+
+import pytest
+
+from axolotl.utils.dict import DictDefault
+from axolotl.utils.validation import validate_config
+
+
+class ValidationTest(unittest.TestCase):
+    """
+    Test the validation module
+    """
+
+    _caplog: Optional[pytest.LogCaptureFixture] = None
+
+    @pytest.fixture(autouse=True)
+    def inject_fixtures(self, caplog):
+        self._caplog = caplog
+
+    def test_load_4bit_deprecate(self):
+        cfg = DictDefault(
+            {
+                "load_4bit": True,
+            }
+        )
+
+        with pytest.raises(ValueError):
+            validate_config(cfg)
+
+    def test_batch_size_unused_warning(self):
+        cfg = DictDefault(
+            {
+                "batch_size": 32,
+            }
+        )
+
+        with self._caplog.at_level(logging.WARNING):
+            validate_config(cfg)
+            assert "batch_size is not recommended" in self._caplog.records[0].message
+
+    def test_qlora(self):
+        base_cfg = DictDefault(
+            {
+                "adapter": "qlora",
+            }
+        )
+
+        cfg = base_cfg | DictDefault(  # pylint: disable=unsupported-binary-operation
+            {
+                "load_in_8bit": True,
+            }
+        )
+
+        with pytest.raises(ValueError, match=r".*8bit.*"):
+            validate_config(cfg)
+
+        cfg = base_cfg | DictDefault(  # pylint: disable=unsupported-binary-operation
+            {
+                "gptq": True,
+            }
+        )
+
+        with pytest.raises(ValueError, match=r".*gptq.*"):
+            validate_config(cfg)
+
+        cfg = base_cfg | DictDefault(  # pylint: disable=unsupported-binary-operation
+            {
+                "load_in_4bit": False,
+            }
+        )
+
+        with pytest.raises(ValueError, match=r".*4bit.*"):
+            validate_config(cfg)
+
+        cfg = base_cfg | DictDefault(  # pylint: disable=unsupported-binary-operation
+            {
+                "load_in_4bit": True,
+            }
+        )
+
+        validate_config(cfg)
+
+    def test_qlora_merge(self):
+        base_cfg = DictDefault(
+            {
+                "adapter": "qlora",
+                "merge_lora": True,
+            }
+        )
+
+        cfg = base_cfg | DictDefault(  # pylint: disable=unsupported-binary-operation
+            {
+                "load_in_8bit": True,
+            }
+        )
+
+        with pytest.raises(ValueError, match=r".*8bit.*"):
+            validate_config(cfg)
+
+        cfg = base_cfg | DictDefault(  # pylint: disable=unsupported-binary-operation
+            {
+                "gptq": True,
+            }
+        )
+
+        with pytest.raises(ValueError, match=r".*gptq.*"):
+            validate_config(cfg)
+
+        cfg = base_cfg | DictDefault(  # pylint: disable=unsupported-binary-operation
+            {
+                "load_in_4bit": True,
+            }
+        )
+
+        with pytest.raises(ValueError, match=r".*4bit.*"):
+            validate_config(cfg)
+
+    def test_hf_use_auth_token(self):
+        cfg = DictDefault(
+            {
+                "push_dataset_to_hub": "namespace/repo",
+            }
+        )
+
+        with pytest.raises(ValueError, match=r".*hf_use_auth_token.*"):
+            validate_config(cfg)
+
+        cfg = DictDefault(
+            {
+                "push_dataset_to_hub": "namespace/repo",
+                "hf_use_auth_token": True,
+            }
+        )
+        validate_config(cfg)
+
+    def test_gradient_accumulations_or_batch_size(self):
+        cfg = DictDefault(
+            {
+                "gradient_accumulation_steps": 1,
+                "batch_size": 1,
+            }
+        )
+
+        with pytest.raises(
+            ValueError, match=r".*gradient_accumulation_steps or batch_size.*"
+        ):
+            validate_config(cfg)
+
+        cfg = DictDefault(
+            {
+                "batch_size": 1,
+            }
+        )
+
+        validate_config(cfg)
+
+        cfg = DictDefault(
+            {
+                "gradient_accumulation_steps": 1,
+            }
+        )
+
+        validate_config(cfg)
+
+    def test_falcon_fsdp(self):
+        regex_exp = r".*FSDP is not supported for falcon models.*"
+
+        # Check for lower-case
+        cfg = DictDefault(
+            {
+                "base_model": "tiiuae/falcon-7b",
+                "fsdp": ["full_shard", "auto_wrap"],
+            }
+        )
+
+        with pytest.raises(ValueError, match=regex_exp):
+            validate_config(cfg)
+
+        # Check for upper-case
+        cfg = DictDefault(
+            {
+                "base_model": "Falcon-7b",
+                "fsdp": ["full_shard", "auto_wrap"],
+            }
+        )
+
+        with pytest.raises(ValueError, match=regex_exp):
+            validate_config(cfg)
+
+        cfg = DictDefault(
+            {
+                "base_model": "tiiuae/falcon-7b",
+            }
+        )
+
+        validate_config(cfg)
+
+    def test_mpt_gradient_checkpointing(self):
+        regex_exp = r".*gradient_checkpointing is not supported for MPT models*"
+
+        # Check for lower-case
+        cfg = DictDefault(
+            {
+                "base_model": "mosaicml/mpt-7b",
+                "gradient_checkpointing": True,
+            }
+        )
+
+        with pytest.raises(ValueError, match=regex_exp):
+            validate_config(cfg)
+
+    def test_flash_optimum(self):
+        cfg = DictDefault(
+            {
+                "flash_optimum": True,
+                "adapter": "lora",
+            }
+        )
+
+        with self._caplog.at_level(logging.WARNING):
+            validate_config(cfg)
+            assert any(
+                "BetterTransformers probably doesn't work with PEFT adapters"
+                in record.message
+                for record in self._caplog.records
+            )
+
+        cfg = DictDefault(
+            {
+                "flash_optimum": True,
+            }
+        )
+
+        with self._caplog.at_level(logging.WARNING):
+            validate_config(cfg)
+            assert any(
+                "probably set bfloat16 or float16" in record.message
+                for record in self._caplog.records
+            )
+
+        cfg = DictDefault(
+            {
+                "flash_optimum": True,
+                "fp16": True,
+            }
+        )
+        regex_exp = r".*AMP is not supported.*"
+
+        with pytest.raises(ValueError, match=regex_exp):
+            validate_config(cfg)
+
+        cfg = DictDefault(
+            {
+                "flash_optimum": True,
+                "bf16": True,
+            }
+        )
+        regex_exp = r".*AMP is not supported.*"
+
+        with pytest.raises(ValueError, match=regex_exp):
+            validate_config(cfg)
+
+    def test_adamw_hyperparams(self):
+        cfg = DictDefault(
+            {
+                "optimizer": None,
+                "adamw_epsilon": 0.0001,
+            }
+        )
+
+        with self._caplog.at_level(logging.WARNING):
+            validate_config(cfg)
+            assert any(
+                "adamw hyperparameters found, but no adamw optimizer set"
+                in record.message
+                for record in self._caplog.records
+            )
+
+        cfg = DictDefault(
+            {
+                "optimizer": "adafactor",
+                "adamw_beta1": 0.0001,
+            }
+        )
+
+        with self._caplog.at_level(logging.WARNING):
+            validate_config(cfg)
+            assert any(
+                "adamw hyperparameters found, but no adamw optimizer set"
+                in record.message
+                for record in self._caplog.records
+            )
+
+        cfg = DictDefault(
+            {
+                "optimizer": "adamw_bnb_8bit",
+                "adamw_beta1": 0.0001,
+                "adamw_beta2": 0.0001,
+                "adamw_epsilon": 0.0001,
+            }
+        )
+
+        validate_config(cfg)
+
+        cfg = DictDefault(
+            {
+                "optimizer": "adafactor",
+            }
+        )
+
+        validate_config(cfg)