experiment w latent space

* Fix(template): Inform to place stack trace to Issue

* Update following suggestions

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

---------

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

.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/CODE_OF_CONDUCT.md

@@ -0,0 +1,129 @@
+# Contributor Covenant Code of Conduct
+
+## Our Pledge
+
+We as members, contributors, and leaders pledge to make participation in our
+community a harassment-free experience for everyone, regardless of age, body
+size, visible or invisible disability, ethnicity, sex characteristics, gender
+identity and expression, level of experience, education, socio-economic status,
+nationality, personal appearance, race, religion, or sexual identity
+and orientation.
+
+We pledge to act and interact in ways that contribute to an open, welcoming,
+diverse, inclusive, and healthy community.
+
+## Our Standards
+
+Examples of behavior that contributes to a positive environment for our
+community include:
+
+* Demonstrating empathy and kindness toward other people
+* Being respectful of differing opinions, viewpoints, and experiences
+* Giving and gracefully accepting constructive feedback
+* Accepting responsibility and apologizing to those affected by our mistakes,
+  and learning from the experience
+* Focusing on what is best not just for us as individuals, but for the
+  overall community
+
+Examples of unacceptable behavior include:
+
+* The use of sexualized language or imagery, and sexual attention or
+  advances of any kind
+* Trolling, insulting or derogatory comments, and personal or political attacks
+* Public or private harassment
+* Publishing others' private information, such as a physical or email
+  address, without their explicit permission
+* Other conduct which could reasonably be considered inappropriate in a
+  professional setting
+
+## Enforcement Responsibilities
+
+Community leaders are responsible for clarifying and enforcing our standards of
+acceptable behavior and will take appropriate and fair corrective action in
+response to any behavior that they deem inappropriate, threatening, offensive,
+or harmful.
+
+Community leaders have the right and responsibility to remove, edit, or reject
+comments, commits, code, wiki edits, issues, and other contributions that are
+not aligned to this Code of Conduct, and will communicate reasons for moderation
+decisions when appropriate.
+
+## Scope
+
+This Code of Conduct applies within all community spaces, and also applies when
+an individual is officially representing the community in public spaces.
+Examples of representing our community include using an official e-mail address,
+posting via an official social media account, or acting as an appointed
+representative at an online or offline event.
+
+## Enforcement
+
+Instances of abusive, harassing, or otherwise unacceptable behavior may be
+reported to the community leaders responsible for enforcement on Discord
+at https://discord.gg/QYF8QrtEUm
+
+All complaints will be reviewed and investigated promptly and fairly.
+
+All community leaders are obligated to respect the privacy and security of the
+reporter of any incident.
+
+## Enforcement Guidelines
+
+Community leaders will follow these Community Impact Guidelines in determining
+the consequences for any action they deem in violation of this Code of Conduct:
+
+### 1. Correction
+
+**Community Impact**: Use of inappropriate language or other behavior deemed
+unprofessional or unwelcome in the community.
+
+**Consequence**: A private, written warning from community leaders, providing
+clarity around the nature of the violation and an explanation of why the
+behavior was inappropriate. A public apology may be requested.
+
+### 2. Warning
+
+**Community Impact**: A violation through a single incident or series
+of actions.
+
+**Consequence**: A warning with consequences for continued behavior. No
+interaction with the people involved, including unsolicited interaction with
+those enforcing the Code of Conduct, for a specified period of time. This
+includes avoiding interactions in community spaces as well as external channels
+like social media. Violating these terms may lead to a temporary or
+permanent ban.
+
+### 3. Temporary Ban
+
+**Community Impact**: A serious violation of community standards, including
+sustained inappropriate behavior.
+
+**Consequence**: A temporary ban from any sort of interaction or public
+communication with the community for a specified period of time. No public or
+private interaction with the people involved, including unsolicited interaction
+with those enforcing the Code of Conduct, is allowed during this period.
+Violating these terms may lead to a permanent ban.
+
+### 4. Permanent Ban
+
+**Community Impact**: Demonstrating a pattern of violation of community
+standards, including sustained inappropriate behavior,  harassment of an
+individual, or aggression toward or disparagement of classes of individuals.
+
+**Consequence**: A permanent ban from any sort of public interaction within
+the community.
+
+## Attribution
+
+This Code of Conduct is adapted from the [Contributor Covenant][homepage],
+version 2.0, available at
+https://www.contributor-covenant.org/version/2/0/code_of_conduct.html.
+
+Community Impact Guidelines were inspired by [Mozilla's code of conduct
+enforcement ladder](https://github.com/mozilla/diversity).
+
+[homepage]: https://www.contributor-covenant.org
+
+For answers to common questions about this code of conduct, see the FAQ at
+https://www.contributor-covenant.org/faq. Translations are available at
+https://www.contributor-covenant.org/translations.

.github/CONTRIBUTING.md

@@ -0,0 +1,76 @@
+# Contributing to axolotl
+
+First of all, thank you for your interest in contributing to axolotl! We appreciate the time and effort you're willing to invest in making our project better. This document provides guidelines and information to make the contribution process as smooth as possible.
+
+## Table of Contents
+
+- [Code of Conduct](#code-of-conduct)
+- [Getting Started](#getting-started)
+- [How to Contribute](#how-to-contribute)
+  - [Reporting Bugs](#reporting-bugs)
+  - [Suggesting Enhancements](#suggesting-enhancements)
+  - [Submitting Pull Requests](#submitting-pull-requests)
+- [Style Guidelines](#style-guidelines)
+  - [Code Style](#code-style)
+  - [Commit Messages](#commit-messages)
+- [Additional Resources](#additional-resources)
+
+## Code of Conductcode
+
+All contributors are expected to adhere to our [Code of Conduct](CODE_OF_CONDUCT.md). Please read it before participating in the axolotl community.
+
+## Getting Started
+
+Bugs? Please check for open issue else create a new [Issue](https://github.com/OpenAccess-AI-Collective/axolotl/issues/new).
+
+PRs are **greatly welcome**!
+
+1. Fork the repository and clone it to your local machine.
+2. Set up the development environment by following the instructions in the [README.md](https://github.com/OpenAccess-AI-Collective/axolotl/tree/main/README.md) file.
+3. Explore the codebase, run tests, and verify that everything works as expected.
+
+Please run below to setup env
+```bash
+pip3 install -r requirements-dev.txt -r requirements-tests.txt
+pre-commit install
+
+# test
+pytest tests/
+```
+
+## How to Contribute
+
+### Reporting Bugs
+
+If you encounter a bug or issue while using axolotl, please open a new issue on the [GitHub Issues](https://github.com/OpenAccess-AI-Collective/axolotl/issues) page. Provide a clear and concise description of the problem, steps to reproduce it, and any relevant error messages or logs.
+
+### Suggesting Enhancements
+
+We welcome ideas for improvements and new features. To suggest an enhancement, open a new issue on the [GitHub Issues](https://github.com/OpenAccess-AI-Collective/axolotl/issues) page. Describe the enhancement in detail, explain the use case, and outline the benefits it would bring to the project.
+
+### Submitting Pull Requests
+
+1. Create a new branch for your feature or bugfix. Use a descriptive name like `feature/your-feature-name` or `fix/your-bugfix-name`.
+2. Make your changes, following the [Style Guidelines](#style-guidelines) below.
+3. Test your changes and ensure that they don't introduce new issues or break existing functionality.
+4. Commit your changes, following the [commit message guidelines](#commit-messages).
+5. Push your branch to your fork on GitHub.
+6. Open a new pull request against the `main` branch of the axolotl repository. Include a clear and concise description of your changes, referencing any related issues.
+
+## Style Guidelines
+
+### Code Style
+
+axolotl uses [{codestyle}]({URLofCodestyle}) as its code style guide. Please ensure that your code follows these guidelines.
+
+### Commit Messages
+
+Write clear and concise commit messages that briefly describe the changes made in each commit. Use the imperative mood and start with a capitalized verb, e.g., "Add new feature" or "Fix bug in function".
+
+## Additional Resources
+
+- [GitHub Help](https://help.github.com/)
+- [GitHub Pull Request Documentation](https://docs.github.com/en/github/collaborating-with-issues-and-pull-requests)
+- [{codestyle}]({URLofCodestyle})
+
+Thank you once again for your interest in contributing to axolotl. We look forward to collaborating with you and creating an even better project together!

.github/FUNDING.yml

@@ -0,0 +1,13 @@
+# These are supported funding model platforms
+
+github: OpenAccess-AI-Collective # Replace with up to 4 GitHub Sponsors-enabled usernames e.g., [user1, user2]
+patreon: # Replace with a single Patreon username
+open_collective: # Replace with a single Open Collective username
+ko_fi: # Replace with a single Ko-fi username
+tidelift: # Replace with a single Tidelift platform-name/package-name e.g., npm/babel
+community_bridge: # Replace with a single Community Bridge project-name e.g., cloud-foundry
+liberapay: # Replace with a single Liberapay username
+issuehunt: # Replace with a single IssueHunt username
+otechie: # Replace with a single Otechie username
+lfx_crowdfunding: # Replace with a single LFX Crowdfunding project-name e.g., cloud-foundry
+custom: # Replace with up to 4 custom sponsorship URLs e.g., ['link1', 'link2']

.github/ISSUE_TEMPLATE/bug-report.yaml

@@ -0,0 +1,105 @@
+name: Bug Report
+description: File a bug report
+labels: ["bug", "needs triage"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        ## Before you start
+        Please **make sure you are on the latest version.**
+        If you encountered the issue after you installed, updated, or reloaded, **please try restarting before reporting the bug**.
+
+  - type: checkboxes
+    id: no-duplicate-issues
+    attributes:
+      label: "Please check that this issue hasn't been reported before."
+      description: "The **Label filters** may help make your search more focussed."
+      options:
+        - label: "I searched previous [Bug Reports](https://github.com/OpenAccess-AI-Collective/axolotl/labels/bug) didn't find any similar reports."
+          required: true
+
+  - type: textarea
+    id: expected
+    attributes:
+      label: Expected Behavior
+      description: Tell us what **should** happen.
+    validations:
+      required: true
+
+  - type: textarea
+    id: what-happened
+    attributes:
+      label: Current behaviour
+      description: |
+        Tell us what happens instead of the expected behavior.
+        Provide stacktrace and/or screenshots.
+    validations:
+      required: true
+
+  - type: textarea
+    id: reproduce
+    attributes:
+      label: Steps to reproduce
+      description: |
+        Which exact steps can a developer take to reproduce the issue?
+        The more detail you provide, the easier it will be to narrow down and fix the bug.
+        Please paste in tasks and/or queries **as text, not screenshots**.
+      placeholder: |
+        Example of the level of detail needed to reproduce any bugs efficiently and reliably.
+        1. Go to the '...' page.
+        2. Click on the '...' button.
+        3. Scroll down to '...'.
+        4. Observe the error.
+    validations:
+      required: true
+
+  - type: textarea
+    id: possible-solution
+    attributes:
+      label: Possible solution
+      description: |
+        Not obligatory, but please suggest a fix or reason for the bug, if you have an idea.
+
+
+  - type: checkboxes
+    id: operating-systems
+    attributes:
+      label: Which Operating Systems are you using?
+      description: You may select more than one.
+      options:
+        - label: Linux
+        - label: macOS
+        - label: Windows
+
+  - type: input
+    id: Python-version
+    attributes:
+      label: Python Version
+      description: Which {Programming} version are you using?
+      placeholder: 3.10 / please change accordingly
+    validations:
+      required: true
+
+  - type: input
+    id: axolotl-branch-commit
+    attributes:
+      label: axolotl branch-commit
+      description: On which branch/commit are you?
+      placeholder: main/4d6490b
+    validations:
+      required: true
+
+  - type: checkboxes
+    id: acknowledgements
+    attributes:
+      label: 'Acknowledgements'
+      description: 'Please confirm the following:'
+      options:
+        - label: 'My issue title is concise, descriptive, and in title casing.'
+          required: true
+        - label: 'I have searched the existing issues to make sure this bug has not been reported yet.'
+          required: true
+        - label: 'I am using the latest version of axolotl.'
+          required: true
+        - label: 'I have provided enough information for the maintainers to reproduce and diagnose the issue.'
+          required: true

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,7 @@
+blank_issues_enabled: false
+contact_links:
+  - name: Ask a question
+    url: https://github.com/OpenAccess-AI-Collective/axolotl/discussions/categories/q-a
+    about: Ask questions and discuss with other community members
+  - name: Discuss the Project in Discord
+    url: https://discord.gg/HhrNrHJPRb

.github/ISSUE_TEMPLATE/docs.yml

@@ -0,0 +1,46 @@
+name: Documentation Improvement / Clarity
+description: Make a suggestion to improve the project documentation.
+labels: ['needs triage', 'docs']
+body:
+  - type: markdown
+    attributes:
+      value: '## :book: Documentation :book:'
+  - type: markdown
+    attributes:
+      value: |
+        * Ask questions in [Discord](https://discord.gg/HhrNrHJPRb).
+        * Before you file an issue read the [Contributing guide](./CONTRIBUTING.md).
+        * Check to make sure someone hasn't already opened a [similar issue](https://github.com/OpenAccess-AI-Collective/axolotl/issues).
+  - type: textarea
+    attributes:
+      label: What piece of documentation is affected?
+      description: Please link to the article you'd like to see updated.
+    validations:
+      required: true
+  - type: textarea
+    attributes:
+      label: What part(s) of the article would you like to see updated?
+      description: |
+        - Give as much detail as you can to help us understand the change you want to see.
+        - Why should the docs be changed? What use cases does it support?
+        - What is the expected outcome?
+    validations:
+      required: true
+  - type: textarea
+    attributes:
+      label: Additional Information
+      description: Add any other context or screenshots about the feature request here.
+    validations:
+      required: false
+  - type: checkboxes
+    id: acknowledgements
+    attributes:
+      label: 'Acknowledgements'
+      description: 'Please confirm the following:'
+      options:
+        - label: 'My issue title is concise, descriptive, and in title casing.'
+          required: true
+        - label: 'I have searched the existing issues to make sure this feature has not been requested yet.'
+          required: true
+        - label: 'I have provided enough information for the maintainers to understand and evaluate this request.'
+          required: true

.github/ISSUE_TEMPLATE/feature-request.yaml

@@ -0,0 +1,63 @@
+name: Feature Request / Enhancement
+description: Suggest a new feature or feature enhancement for the project
+labels: ["enhancement", "needs triage"]
+body:
+  - type: checkboxes
+    id: no-duplicate-issues
+    attributes:
+      label: "⚠️ Please check that this feature request hasn't been suggested before."
+      description: "There are two locations for previous feature requests. Please search in both. Thank you. The **Label filters** may help make your search more focussed."
+      options:
+        - label: "I searched previous [Ideas in Discussions](https://github.com/OpenAccess-AI-Collective/axolotl/discussions/categories/ideas) didn't find any similar feature requests."
+          required: true
+        - label: "I searched previous [Issues](https://github.com/OpenAccess-AI-Collective/axolotl/labels/enhancement) didn't find any similar feature requests."
+          required: true
+
+  - type: textarea
+    id: feature-description
+    validations:
+      required: true
+    attributes:
+      label: "🔖 Feature description"
+      description: "A clear and concise description of what the feature request is."
+      placeholder: "You should add ..."
+
+  - type: textarea
+    id: solution
+    validations:
+      required: true
+    attributes:
+      label: "✔️ Solution"
+      description: "A clear and concise description of what you want to happen, and why."
+      placeholder: "In my use-case, ..."
+
+  - type: textarea
+    id: alternatives
+    validations:
+      required: false
+    attributes:
+      label: "❓ Alternatives"
+      description: "A clear and concise description of any alternative solutions or features you've considered."
+      placeholder: "I have considered ..."
+
+  - type: textarea
+    id: additional-context
+    validations:
+      required: false
+    attributes:
+      label: "📝 Additional Context"
+      description: "Add any other context or screenshots about the feature request here."
+      placeholder: "..."
+
+  - type: checkboxes
+    id: acknowledgements
+    attributes:
+      label: 'Acknowledgements'
+      description: 'Please confirm the following:'
+      options:
+        - label: 'My issue title is concise, descriptive, and in title casing.'
+          required: true
+        - label: 'I have searched the existing issues to make sure this feature has not been requested yet.'
+          required: true
+        - label: 'I have provided enough information for the maintainers to understand and evaluate this request.'
+          required: true

.github/PULL_REQUEST_TEMPLATE/pull_request_template_simple.md

@@ -0,0 +1,22 @@
+<!--- Provide a general summary of your changes in the Title above -->
+
+# Description
+
+<!--- Describe your changes in detail -->
+
+## Motivation and Context
+
+<!--- Why is this change required? What problem does it solve? -->
+<!--- If it fixes an open issue, please link to the issue here. -->
+
+## How has this been tested?
+
+<!--- Please describe in detail how you tested your changes. -->
+<!--- Include details of your testing environment, tests ran to see how -->
+<!--- your change affects other areas of the code, etc. -->
+
+## Screenshots (if appropriate)
+
+## Types of changes
+
+<!--- What types of changes does your code introduce? Put an `x` in all the boxes that apply: -->

.github/SECURITY.md

@@ -0,0 +1,9 @@
+# Security Policy
+
+## Supported Versions
+
+Due to the nature of the fast development that is happening in this project, only the latest released version can be supported.
+
+## Reporting a Vulnerability
+
+If you find a vulnerability, please contact us on  [Discord](https://discord.gg/xcu3ECkH9a) rather than creating a GitHub issue to allow us some time to fix it before it is a known vulnerability to others.

.github/SUPPORT.md

@@ -0,0 +1,10 @@
+# Support
+
+If you need help with this project or have questions, please:
+
+1. Check the documentation.
+2. Search the existing issues and pull requests.
+3. Create a new issue if your question is not answered or your problem is not solved.
+4. Have a look in the [Discord server](https://discord.gg/HhrNrHJPRb)
+
+Please note that this project is maintained by volunteers who have limited availability. We'll do our best to address your questions and concerns in a timely manner.

.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 +1,56 @@
+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: 2.0.1
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 9.0+PTX"
+          - cuda: "118"
+            cuda_version: 11.8.0
+            python_version: "3.10"
+            pytorch: 2.0.1
+            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 9.0+PTX"
+    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: ${{ steps.metadata.outputs.labels }}
+          build-args: |
+            CUDA_VERSION=${{ matrix.cuda_version }}
+            CUDA=${{ matrix.cuda }}
+            PYTHON_VERSION=${{ matrix.python_version }}
+            PYTORCH_VERSION=${{ matrix.pytorch }}
+            TORCH_CUDA_ARCH_LIST=${{ matrix.torch_cuda_arch_list }}

.github/workflows/main.yml

@@ -0,0 +1,108 @@
+name: ci-cd
+
+on:
+  push:
+    branches:
+      - "main"
+
+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: 2.0.1
+            axolotl_extras:
+          - cuda: cu118
+            cuda_version: 11.8.0
+            python_version: "3.10"
+            pytorch: 2.0.1
+            axolotl_extras:
+          - cuda: cu118
+            cuda_version: 11.8.0
+            python_version: "3.9"
+            pytorch: 2.0.1
+            axolotl_extras: gptq
+    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 }}
+          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 }}
+  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: 118
+            cuda_version: 11.8.0
+            python_version: "3.9"
+            pytorch: 2.0.1
+            axolotl_extras:
+          - cuda: 118
+            cuda_version: 11.8.0
+            python_version: "3.10"
+            pytorch: 2.0.1
+            axolotl_extras:
+            is_latest: true
+          - cuda: 118
+            cuda_version: 11.8.0
+            python_version: "3.9"
+            pytorch: 2.0.1
+            axolotl_extras: gptq
+    runs-on: self-hosted
+    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${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+            CUDA=${{ matrix.cuda }}
+          file: ./docker/Dockerfile-runpod
+          push: ${{ github.event_name != 'pull_request' }}
+          tags: |
+             ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
+             ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
+          labels: ${{ steps.metadata.outputs.labels }}

.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.

LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

README.md

@@ -1,47 +1,726 @@
-~~# Axolotl
+# Axolotl
 
-#### You know you're going to axolotl questions
+Axolotl is a tool designed to streamline the fine-tuning of various AI models, offering support for multiple configurations and architectures.
 
-## Getting Started
+<table>
+<tr>
+<td>
 
-- 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))
+## Table of Contents
+- [Introduction](#axolotl)
+- [Supported Features](#axolotl-supports)
+- [Quickstart](#quickstart-)
+- [Installation](#installation)
+  - [Docker Installation](#environment)
+  - [Conda/Pip venv Installation](#condapip-venv)
+  - [LambdaLabs Installation](#lambdalabs)
+- [Dataset](#dataset)
+  - [How to Add Custom Prompts](#how-to-add-custom-prompts)
+- [Config](#config)
+  - [Train](#train)
+  - [Inference](#inference)
+  - [Merge LORA to Base](#merge-lora-to-base)
+- [Common Errors](#common-errors-)
+- [Need Help?](#need-help-)
+- [Badge](#badge-)
+- [Community Showcase](#community-showcase)
+- [Contributing](#contributing-)
+
+</td>
+<td>
+
+<div align="center">
+  <img src="image/axolotl.png" alt="axolotl" width="160">
+  <div>
+    <p>
+      <b>Axolotl provides a unified repository for fine-tuning <br />a variety of AI models with ease</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>
+
+</td>
+</tr>
+</table>
+
+## Axolotl supports
+
+|          | fp16/fp32 | lora | qlora | gptq | gptq w/flash attn | flash attn | xformers attn |
+|----------|:----------|:-----|-------|------|-------------------|------------|---------------|
+| llama    | ✅         | ✅    | ✅     | ✅             | ✅                 | ✅          | ✅             |
+| Pythia   | ✅         | ✅    | ✅     | ❌             | ❌                 | ❌          | ❓             |
+| cerebras | ✅         | ✅    | ✅     | ❌             | ❌                 | ❌          | ❓             |
+| mpt      | ✅         | ❌    | ❓     | ❌             | ❌                 | ❌          | ❓             |
+| falcon   | ✅         | ✅    | ✅     | ❌             | ❌                 | ❌          | ❓             |
+| gpt-j    | ✅         | ✅    | ✅     | ❌             | ❌                 | ❓          | ❓             |
+| XGen     | ✅         | ❓    | ✅     | ❓             | ❓                 | ❓          | ✅             |
+
+
+## Quickstart ⚡
+
+Get started with Axolotl in just a few steps! This quickstart guide will walk you through setting up and running a basic fine-tuning task.
+
+**Requirements**: Python >=3.9 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
+
+# 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-py3.10-cu118-2.0.1
+  ```
+  - `winglian/axolotl-runpod:main-py3.10-cu118-2.0.1`: for runpod
+  - `winglian/axolotl-runpod:main-py3.9-cu118-2.0.1-gptq`: for gptq
+
+  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 -U torch --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
+
+Axolotl supports a variety of dataset formats. Below are some of the formats you can use.
+Have dataset(s) in one of the following format (JSONL recommended):
+
+- `alpaca`: instruction; input(optional)
+  ```json
+  {"instruction": "...", "input": "...", "output": "..."}
+  ```
+- `sharegpt:chat`: conversations where `from` is `human`/`gpt`
+  ```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_simple.load_guanaco`: conversations where `from` is `prompter`/`assistant` instead of default sharegpt
+  ```json
+  {"conversations": [{"from": "...", "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 `<prompt_strategies_file>.load_<load_fn>`.
+
+Optionally, download some datasets, see [data/README.md](data/README.md)
+
+
+
+### Config
+
+See [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
+  sequence_len: 2048 # max token length for prompt
+
+  # huggingface repo
+  datasets:
+    - path: vicgalle/alpaca-gpt4
+      type: alpaca # format from earlier
+
+  # huggingface repo with specific configuration/subset
+  datasets:
+    - path: EleutherAI/pile
+      name: enron_emails
+      type: completion # format from earlier
+
+  # local
+  datasets:
+    - path: json
+      data_files: data.jsonl # or json
+      type: alpaca # format from earlier
+  ```
+
+- 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
+# you can specify to choose a specific model revision from huggingface hub
+model_revision:
+# Optional tokenizer configuration override in case you want to use a different tokenizer
+# than the one defined in the base model
+tokenizer_config:
+# If you want to specify the type of model to load, AutoModelForCausalLM is a good choice too
+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 to use the legacy tokenizer setting, defaults to True
+tokenizer_legacy:
+# resize the model embeddings when new tokens are added to multiples of 32
+# this is reported to improve training speed on some models
+resize_token_embeddings_to_32x:
+
+# 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:
+  # hf dataset repo | "json" for local dataset, make sure to fill data_files
   - path: vicgalle/alpaca-gpt4
-    type: alpaca
+  # The type of prompt to use for training. [alpaca, sharegpt, gpteacher, oasst, reflection]
+    type: alpaca # format | format:<prompt_style> (chat/instruct) | <prompt_strategies>.load_<load_fn>
+    data_files: # path to source data files
+    shards: # number of shards to split data into
+    name: # name of dataset configuration to load
+
+# 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
+hub_model_id: # repo path to push finetuned model
+# how to push checkpoints to hub
+# https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy
+hub_strategy:
+# 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. 0 for no eval.
+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
+# FutureWarning: This will soon be DEPRECATED
+max_packed_sequence_len: 1024
+# use efficient multi-packing with block diagonal attention and per sequence position_ids. Recommend set to 'true'
+sample_packing:
+# you can set these packing optimizations AFTER starting a training at least once.
+# The trainer will provide recommended values for these values.
+sample_packing_eff_est:
+total_num_tokens:
+
+# 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: # "offline" to save run metadata locally and not sync to the server, "disabled" to turn off wandb
+wandb_project: # your wandb project name
+wandb_entity: # a wandb Team name if using a Team
+wandb_watch:
+wandb_run_id: # set the name of your wandb run
+wandb_log_model: # "checkpoint" to log model to wandb Artifacts every `save_steps` or "end" to log only at the end of training
+
+# 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
+lr_quadratic_warmup:
+logging_steps:
+save_steps: # leave empty to save at each epoch
+eval_steps:
+save_total_limit: # checkpoints saved at a time
+max_steps:
+
+# save model as safetensors (require safetensors package)
+save_safetensors:
+
+# whether to mask out or include the human's prompt from the training labels
+train_on_inputs: false
+# group similarly sized data to minimize padding
+# may be slower to start, as it must download and sort the entire dataset
+# note that training loss may have an oscillating pattern with this enabled
+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/Dao-AILab/flash-attention:
+flash_attention:
+# 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:
+# RoPE Scaling https://github.com/huggingface/transformers/pull/24653
+rope_scaling:
+  type: # linear | dynamic
+  factor: # float
+
+# 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 config path
+deepspeed:
+
+# Path to torch distx for optim 'adamw_anyprecision'
+torchdistx_path:
+
+# Set padding for data collator to 'longest'
+collator_pad_to_longest:
+
+# Set to HF dataset for type: 'completion' for streaming instead of pre-tokenize
+pretraining_dataset:
+
+# 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>
 
+### Train
 
-- 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:
+Run
+```bash
+accelerate launch scripts/finetune.py configs/your_config.yml
+```
 
-    - `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`
+#### Multi-GPU
 
+You can optionally pre-tokenize dataset with the following before finetuning:
+```bash
+CUDA_VISIBLE_DEVICES="" accelerate ... --prepare_ds_only
+```
+
+##### Config
+
+- llama FSDP
 ```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
+fsdp:
+  - full_shard
+  - auto_wrap
+fsdp_config:
+  fsdp_offload_params: true
+  fsdp_state_dict_type: FULL_STATE_DICT
+  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
 ```
 
-- 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`~~
+- llama Deepspeed
+```yaml
+deepspeed: deepspeed/zero3.json
+```
+
+##### Weights & Biases Logging
+
+- wandb options
+```yaml
+wandb_mode:
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+```
+
+### 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 🧰
+
+> If you encounter a 'Cuda out of memory' error, it means your GPU ran out of memory during the training process. Here's how to resolve it:
+
+Please reduce any below
+  - `micro_batch_size`
+  - `eval_batch_size`
+  - `gradient_accumulation_steps`
+  - `sequence_len`
+
+> `failed (exitcode: -9)` usually means your system has run out of system memory.
+Similarly, you should consider reducing the same settings as when you run out of VRAM.
+Additionally, look into upgrading your system RAM which should be simpler than GPU upgrades.
+
+> 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.
+
+> accelerate config missing
+
+It's safe to ignore it.
+
+## 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
+
+Check out some of the projects and models that have been built using Axolotl! Have a model you'd like to add to our Community Showcase? Open a PR with your model.
+
+Open Access AI Collective
+- [Minotaur 13b](https://huggingface.co/openaccess-ai-collective/minotaur-13b)
+- [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 🤝
+
+Please read the [contributing guide](./.github/CONTRIBUTING.md)
+
+Bugs? Please check the [open issues](https://github.com/OpenAccess-AI-Collective/axolotl/issues/bug) else create a new Issue.
+
+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

deepspeed/zero3.json

@@ -0,0 +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,
+    "hysteresis": 2,
+    "min_loss_scale": 1
+  },
+  "optimizer": {
+    "type": "AdamW",
+    "params": {
+      "lr": "auto",
+      "betas": [
+        0.9,
+        0.95
+      ],
+      "eps": 1e-8,
+      "weight_decay": "auto"
+    }
+  },
+  "scheduler": {
+    "type": "WarmupLR",
+    "params": {
+      "warmup_min_lr": "auto",
+      "warmup_max_lr": "auto",
+      "warmup_num_steps": "auto"
+    }
+  },
+  "train_batch_size": "auto",
+  "train_micro_batch_size_per_gpu": "auto",
+  "wall_clock_breakdown": false
+}

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 +1,30 @@
+ARG BASE_TAG=main-base
+FROM winglian/axolotl-base:$BASE_TAG
+
+ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
+ARG AXOLOTL_EXTRAS=""
+ARG CUDA="118"
+ENV BNB_CUDA_VERSION=$CUDA
+
+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"
+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
+
+# fix so that git fetch/pull from remote works
+RUN cd axolotl && \
+    git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
+    git config --get remote.origin.fetch
+
+# helper for huggingface-login cli
+RUN git config --global credential.helper store

docker/Dockerfile-base

@@ -0,0 +1,104 @@
+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_VERSION="2.0.1"
+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_VERSION}+cu${CUDA} --extra-index-url https://download.pytorch.org/whl/cu$CUDA
+
+
+FROM base-builder AS flash-attn-builder
+
+WORKDIR /workspace
+
+ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
+
+RUN git clone https://github.com/Dao-AILab/flash-attention.git && \
+    cd flash-attention && \
+    git checkout v2.0.4  && \
+    python3 setup.py bdist_wheel && \
+    cd csrc/fused_dense_lib && \
+    python3 setup.py bdist_wheel && \
+    cd ../xentropy && \
+    python3 setup.py bdist_wheel && \
+    cd ../rotary && \
+    python3 setup.py bdist_wheel && \
+    cd ../layer_norm && \
+    python3 setup.py bdist_wheel
+
+FROM base-builder AS deepspeed-builder
+
+ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
+
+WORKDIR /workspace
+
+RUN git clone https://github.com/microsoft/DeepSpeed.git && \
+    cd DeepSpeed && \
+    MAX_CONCURRENCY=8 DS_BUILD_SPARSE_ATTN=0 DS_BUILD_OPS=1 python3 setup.py bdist_wheel
+
+FROM base-builder AS bnb-builder
+
+WORKDIR /workspace
+ARG CUDA="118"
+ENV CUDA=$CUDA
+
+RUN git clone https://github.com/TimDettmers/bitsandbytes.git && \
+    cd bitsandbytes && \
+    CUDA_VERSION=$CUDA make cuda11x && \
+    python setup.py bdist_wheel
+
+FROM base-builder
+
+ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6 9.0+PTX"
+ENV TORCH_CUDA_ARCH_LIST=$TORCH_CUDA_ARCH_LIST
+
+# recompile apex
+RUN python3 -m pip uninstall -y apex
+RUN git clone https://github.com/NVIDIA/apex
+#  `MAX_JOBS=1` disables parallel building to avoid cpu memory OOM when building image on GitHub Action (standard) runners
+RUN cd apex && MAX_JOBS=1 python3 -m pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" ./
+
+RUN mkdir -p /workspace/builds
+COPY --from=bnb-builder /workspace/bitsandbytes /workspace/builds/bitsandbytes
+
+RUN mkdir -p /workspace/wheels/bitsandbytes
+COPY --from=deepspeed-builder /workspace/DeepSpeed/dist/deepspeed-*.whl wheels
+COPY --from=bnb-builder /workspace/bitsandbytes/dist/bitsandbytes-*.whl wheels
+COPY --from=bnb-builder /workspace/bitsandbytes/bitsandbytes/libbitsandbytes*.so wheels/bitsandbytes
+COPY --from=flash-attn-builder /workspace/flash-attention/dist/flash_attn-*.whl wheels
+COPY --from=flash-attn-builder /workspace/flash-attention/csrc/fused_dense_lib/dist/fused_dense_lib-*.whl wheels
+COPY --from=flash-attn-builder /workspace/flash-attention/csrc/xentropy/dist/xentropy_cuda_lib-*.whl wheels
+COPY --from=flash-attn-builder /workspace/flash-attention/csrc/rotary/dist/rotary_emb-*.whl wheels
+COPY --from=flash-attn-builder /workspace/flash-attention/csrc/layer_norm/dist/dropout_layer_norm-*.whl wheels
+
+RUN pip3 install wheels/deepspeed-*.whl wheels/flash_attn-*.whl wheels/fused_dense_lib-*.whl wheels/xentropy_cuda_lib-*.whl wheels/rotary_emb-*.whl wheels/dropout_layer_norm-*.whl
+RUN cd /workspace/builds/bitsandbytes && python3 setup.py install
+RUN git lfs install --skip-repo
+RUN pip3 install awscli && \
+    # The base image ships with `pydantic==1.8.2` which is not working
+    pip3 install -U --no-cache-dir pydantic==1.10.10

docker/Dockerfile-runpod

@@ -0,0 +1,18 @@
+ARG BASE_TAG=main
+FROM winglian/axolotl:$BASE_TAG
+
+ENV HF_DATASETS_CACHE="/workspace/data/huggingface-cache/datasets"
+ENV HUGGINGFACE_HUB_CACHE="/workspace/data/huggingface-cache/hub"
+ENV TRANSFORMERS_CACHE="/workspace/data/huggingface-cache/hub"
+
+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 +0,0 @@
-{
-  "bf16": {
-    "enabled": "auto"
-  },
-  "fp16": {
-    "enabled": "auto",
-    "loss_scale": 0,
-    "loss_scale_window": 1000,
-    "initial_scale_power": 16,
-    "hysteresis": 2,
-    "min_loss_scale": 1
-  },
-  "scheduler": {
-    "type": "WarmupLR",
-    "params": {
-      "warmup_min_lr": "auto",
-      "warmup_max_lr": "auto",
-      "warmup_num_steps": "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
-  },
-  "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
-}

examples/cerebras/qlora.yml

@@ -0,0 +1,61 @@
+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_entity:
+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: 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: 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 +1,64 @@
+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_entity:
+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 +1,93 @@
+# 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_entity:
+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 +1,64 @@
+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_entity:
+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 +1,58 @@
+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_entity:
+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: 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: 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 +1,63 @@
+base_model: Neko-Institute-of-Science/LLaMA-7B-4bit-128g
+base_model_config: Neko-Institute-of-Science/LLaMA-7B-4bit-128g
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+trust_remote_code:
+load_in_8bit: true
+gptq: true
+datasets:
+  - path: vicgalle/alpaca-gpt4
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.02
+adapter:
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len:
+lora_r: 8
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_modules:
+  - q_proj
+  - v_proj
+lora_fan_in_fan_out: false
+wandb_project: llama-7b-lora-int4
+wandb_entity:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./llama-7b-lora-int4
+gradient_accumulation_steps: 1
+micro_batch_size: 1
+num_epochs: 3
+optimizer: adamw_bnb_8bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.0000002
+train_on_inputs: false
+group_by_length: false
+fp16: true
+bf16: false
+tf32: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 5
+xformers_attention:
+flash_attention:
+gradient_checkpointing: true
+gptq_groupsize: 128
+gptq_model_v1: false
+warmup_steps: 20
+eval_steps: 110
+save_steps: 660
+debug:
+deepspeed:
+weight_decay: 0.0001
+fsdp:
+fsdp_config:
+tokens:
+  pad_token: "[PAD]"
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"

examples/jeopardy-bot/config.yml

@@ -0,0 +1,56 @@
+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_entity:
+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/llama-2/README.md

@@ -0,0 +1,20 @@
+# Overview
+
+This is an example of a llama-2 configuration for 7b and 13b. The yaml file contains configuration for the 7b variant, but you can just aswell use the same settings for 13b.
+
+The 7b variant fits on any 24GB VRAM GPU and will take up about 17 GB of VRAM during training if using qlora and 20 GB if using lora. On a RTX 4090 it trains 3 epochs of the default dataset in about 15 minutes.
+
+The 13b variant will fit if you change these settings to these values:
+gradient_accumulation_steps: 2
+micro_batch_size: 1
+
+```shell
+accelerate launch scripts/finetune.py examples/llama-2/qlora.yml
+
+```
+or
+
+```shell
+accelerate launch scripts/finetune.py examples/llama-2/lora.yml
+
+```

examples/llama-2/lora.yml

@@ -0,0 +1,67 @@
+base_model: meta-llama/Llama-2-7b-hf
+base_model_config: meta-llama/Llama-2-7b-hf
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: true
+load_in_4bit: false
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
+output_dir: ./lora-out
+
+sequence_len: 4096
+sample_packing: true
+
+adapter: lora
+lora_model_dir:
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 3
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+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/llama-2/qlora.yml

@@ -0,0 +1,69 @@
+base_model: meta-llama/Llama-2-7b-hf
+base_model_config: meta-llama/Llama-2-7b-hf
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+is_llama_derived_model: true
+
+load_in_8bit: false
+load_in_4bit: true
+strict: false
+
+datasets:
+  - path: mhenrichsen/alpaca_2k_test
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
+output_dir: ./qlora-out
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 4096
+sample_packing: true
+
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_modules:
+lora_target_linear: true
+lora_fan_in_fan_out:
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 3
+optimizer: paged_adamw_32bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_steps: 10
+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/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 +1,59 @@
+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_entity:
+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 +1,63 @@
+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_entity:
+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 +1,68 @@
+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_entity:
+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 +1,62 @@
+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: 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_entity:
+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: 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: 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 +1,50 @@
+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_entity:
+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 +1,30 @@
 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_entity:
 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 +33,6 @@ 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 +1,60 @@
+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_entity:
+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 +1,56 @@
+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_entity:
+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:

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

@@ -0,0 +1,91 @@
+# An example finetuning Saleforce's XGen-7b model with 8k context using qlora
+# on Tim Dettmer's Guanaco dataset.
+base_model: Salesforce/xgen-7b-8k-base
+base_model_config: Salesforce/xgen-7b-8k-base
+trust_remote_code: true
+model_type: AutoModelForCausalLM
+tokenizer_type: AutoTokenizer
+load_in_8bit: false
+# enable 4bit for QLoRA
+load_in_4bit: true
+gptq: false
+strict: false
+push_dataset_to_hub:
+datasets:
+  - path: timdettmers/openassistant-guanaco
+    data_files:
+      - openassistant_best_replies_train.jsonl
+    type: "completion"
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
+# enable QLoRA
+adapter: qlora
+lora_model_dir:
+sequence_len: 8192
+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_entity:
+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: 1
+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: 0.00002
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: false
+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: 50
+save_steps: 50
+debug:
+deepspeed:
+weight_decay: 0.0
+special_tokens:
+  eos_token: "<|endoftext|>"
+  bos_token: "<|endoftext|>"
+  unk_token: "<|endoftext|>"
+  pad_token: "<|endoftext|>"

requirements-dev.txt

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

requirements-tests.txt

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

requirements.txt

@@ -1,14 +1,24 @@
 peft @ git+https://github.com/huggingface/peft.git
 transformers @ git+https://github.com/huggingface/transformers.git
-attrdict
+bitsandbytes>=0.41.1
+accelerate @ git+https://github.com/huggingface/accelerate@2a289f6108e77a77a4efffb3f6316bc98538413b
+addict
 fire
 PyYAML==6.0
-black
-bitsandbytes
 datasets
-accelerate
+accelerate>=0.19.0
 sentencepiece
 wandb
-flash-attn
-deepspeed
 einops
+xformers
+optimum
+hf_transfer
+numba
+numpy==1.24.4
+# qlora things
+bert-score==0.3.13
+evaluate==0.4.0
+rouge-score==0.1.2
+scipy
+scikit-learn==1.2.2
+pynvml

scripts/alpaca_json_to_jsonl.py

@@ -1,24 +1,41 @@
+"""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,
+)
+from axolotl.logging_config import configure_logging
+
+configure_logging()
 
 # add src to the pythonpath so we don't need to pip install this
 project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
 src_dir = os.path.join(project_root, "src")
 sys.path.insert(0, src_dir)
 
-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 +45,7 @@ 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 +1,132 @@
+"""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.logging_config import configure_logging
+from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.data import prepare_dataset
+from axolotl.utils.dict import DictDefault
+from axolotl.utils.distributed import is_main_process
+from axolotl.utils.models import load_model, load_tokenizer
+from axolotl.utils.tokenization import check_dataset_labels
+from axolotl.utils.trainer import setup_trainer
+from axolotl.utils.wandb import setup_wandb_env_vars
 
 project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
 src_dir = os.path.join(project_root, "src")
 sys.path.insert(0, src_dir)
 
-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
+configure_logging()
+LOG = logging.getLogger("axolotl.scripts")
 
-logging.basicConfig(level=os.getenv("LOG_LEVEL", "INFO"))
-DEFAULT_DATASET_PREPARED_PATH = "last_run_prepared"
+os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
 
 
-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 print_axolotl_text_art():
+    ascii_art = """
+                           dP            dP   dP
+                           88            88   88
+.d8888b. dP.  .dP .d8888b. 88 .d8888b. d8888P 88
+88'  `88  `8bd8'  88'  `88 88 88'  `88   88   88
+88.  .88  .d88b.  88.  .88 88 88.  .88   88   88
+`88888P8 dP'  `dP `88888P' dP `88888P'   dP   dP
+"""
+
+    if is_main_process():
+        print(ascii_art)
 
 
-def 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()
+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
 
-    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()
+def do_inference(cfg, model, tokenizer, prompter: Optional[str]):
+    default_tokens = {"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>"}
 
-    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,
-            )
+    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})
 
-            replace_peft_model_with_int4_lora_model()
-
-        from peft import (
-            LoraConfig,
-            get_peft_model,
-            prepare_model_for_int8_training,
-            PeftModel,
+    prompter_module = None
+    if prompter:
+        prompter_module = getattr(
+            importlib.import_module("axolotl.prompters"), prompter
         )
-    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
+    if cfg.landmark_attention:
+        from axolotl.monkeypatch.llama_landmark_attn import set_model_mem_id
 
-            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,
+        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:
-            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,
+            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,
             )
-    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"
-
-    cfg.device = get_device()
-    if cfg.device == "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")
+            streamer = TextStreamer(tokenizer)
+            generated = model.generate(
+                inputs=batch["input_ids"].to(cfg.device),
+                generation_config=generation_config,
+                streamer=streamer,
             )
-            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 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>"})
-
-    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)
-
-    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,
-        )
-    print(tokenizer.decode(generated["sequences"].cpu().tolist()[0]))
+        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 +151,8 @@ 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(
@@ -421,198 +160,155 @@ def train(
     prepare_ds_only: bool = False,
     **kwargs,
 ):
+    print_axolotl_text_art()
     if Path(config).is_dir():
         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]
 
-    # setup some derived config / hyperparams
-    cfg.gradient_accumulation_steps = cfg.batch_size // cfg.micro_batch_size
-    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
-    if cfg.ddp:
-        cfg.device_map = {"": int(os.environ.get("LOCAL_RANK", 0))}
-        cfg.gradient_accumulation_steps = (
-            cfg.gradient_accumulation_steps // cfg.world_size
-        )
+    validate_config(cfg)
+
+    normalize_config(cfg)
+
     setup_wandb_env_vars(cfg)
-    if cfg.device == "mps":
-        cfg.load_in_8bit = False
-        cfg.tf32 = False
-        if cfg.bf16:
-            cfg.fp16 = True
-        cfg.bf16 = False
 
-    # Load the model and tokenizer
-    logging.info("loading model, tokenizer, and lora_config...")
-    model, tokenizer, lora_config = load_model(
-        cfg.base_model,
-        cfg.base_model_config,
-        cfg.model_type,
-        cfg.tokenizer_type,
-        cfg,
-        adapter=cfg.adapter,
-        inference=("inference" in kwargs),
-    )
+    # load the tokenizer first
+    LOG.info(f"loading tokenizer... {cfg.tokenizer_config or cfg.base_model_config}")
+    tokenizer = load_tokenizer(cfg)
 
-    if "inference" in kwargs:
-        logging.info("calling do_inference function")
-        do_inference(cfg, model, tokenizer)
+    if (
+        check_not_in(["shard", "merge_lora"], kwargs) and not cfg.inference
+    ):  # don't need to load dataset for these
+        train_dataset, eval_dataset, total_num_steps = prepare_dataset(cfg, tokenizer)
+
+    if cfg.debug or "debug" in kwargs:
+        LOG.info("check_dataset_labels...")
+        check_dataset_labels(
+            train_dataset.select(
+                [random.randrange(0, len(train_dataset) - 1) for _ in range(5)]  # nosec
+            ),
+            tokenizer,
+        )
+
+    if prepare_ds_only:
+        LOG.info("Finished preparing dataset. Exiting...")
         return
 
-    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
-    )
+    # Load the model and tokenizer
+    LOG.info("loading model and (optionally) peft_config...")
+    model, peft_config = load_model(cfg, tokenizer)
 
-    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
+    safe_serialization = cfg.save_safetensors is True
 
-            # 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:
+        LOG.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)
+            LOG.info("saving merged model")
+            model.save_pretrained(
+                str(Path(cfg.output_dir) / "merged"),
+                safe_serialization=safe_serialization,
+            )
+            tokenizer.save_pretrained(str(Path(cfg.output_dir) / "merged"))
+        return
 
-        if prepare_ds_only:
-            logging.info("Finished preparing dataset. Exiting...")
-            return
+    if cfg.inference:
+        LOG.info("calling do_inference function")
+        prompter: Optional[str] = "AlpacaPrompter"
+        if "prompter" in kwargs:
+            if kwargs["prompter"] == "None":
+                prompter = None
+            else:
+                prompter = kwargs["prompter"]
+        do_inference(cfg, model, tokenizer, prompter=prompter)
+        return
 
-    train_dataset = dataset["train"]
-    eval_dataset = dataset["test"]
+    if "shard" in kwargs:
+        model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)
+        return
 
-    if cfg.debug:
-        check_dataset_labels(
-            train_dataset.select([random.randrange(0, len(train_dataset) - 1)]),
-            tokenizer,
-        )
-
-    trainer = setup_trainer(cfg, train_dataset, eval_dataset, model, tokenizer)
+    trainer = setup_trainer(
+        cfg, train_dataset, eval_dataset, model, tokenizer, total_num_steps
+    )
 
     model.config.use_cache = False
 
     if torch.__version__ >= "2" and sys.platform != "win32":
-        logging.info("Compiling torch model")
+        LOG.info("Compiling torch model")
         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:
+        LOG.info(f"Pre-saving adapter config to {cfg.output_dir}")
+        peft_config.save_pretrained(cfg.output_dir)
 
     # In case we want to stop early with ctrl+c, this is a nice to have to save the pretrained model
     if cfg.local_rank == 0:
+
+        def terminate_handler(_, __, model):
+            if cfg.flash_optimum:
+                model = BetterTransformer.reverse(model)
+            model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)
+            sys.exit(0)
+
         signal.signal(
-            signal.SIGINT,
-            lambda 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)
+    LOG.info("Starting trainer...")
+    if cfg.group_by_length:
+        LOG.info("hang tight... sorting dataset for group_by_length")
+    resume_from_checkpoint = cfg.resume_from_checkpoint
+    if cfg.resume_from_checkpoint is None and cfg.auto_resume_from_checkpoints:
+        possible_checkpoints = [
+            str(cp) for cp in Path(cfg.output_dir).glob("checkpoint-*")
+        ]
+        if len(possible_checkpoints) > 0:
+            sorted_paths = sorted(
+                possible_checkpoints,
+                key=lambda path: int(path.split("-")[-1]),
+            )
+            resume_from_checkpoint = sorted_paths[-1]
+            LOG.info(
+                f"Using Auto-resume functionality to start with checkpoint at {resume_from_checkpoint}"
+            )
 
-    if 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}"
-        )
-        model.save_pretrained(cfg.output_dir)
+    if not Path(cfg.output_dir).is_dir():
+        os.makedirs(cfg.output_dir, exist_ok=True)
+    tokenizer.save_pretrained(cfg.output_dir)
+    if cfg.flash_optimum:
+        with torch.backends.cuda.sdp_kernel(
+            enable_flash=True, enable_math=True, enable_mem_efficient=True
+        ):
+            trainer.train(resume_from_checkpoint=resume_from_checkpoint)
+    else:
+        trainer.train(resume_from_checkpoint=resume_from_checkpoint)
+
+    LOG.info(f"Training Completed!!! Saving pre-trained model to {cfg.output_dir}")
+
+    # TODO do we need this fix? https://huggingface.co/docs/accelerate/usage_guides/fsdp#saving-and-loading
+    # only save on rank 0, otherwise it corrupts output on multi-GPU when multiple processes attempt to write the same file
+    if cfg.fsdp:
+        trainer.save_model(cfg.output_dir)
+    elif cfg.local_rank == 0:
+        if cfg.flash_optimum:
+            model = BetterTransformer.reverse(model)
+        model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)
 
 
 if __name__ == "__main__":

scripts/runpod-entrypoint.sh

@@ -0,0 +1,21 @@
+#!/bin/bash
+
+# Export specific ENV variables to /etc/rp_environment
+echo "Exporting environment variables..."
+printenv | grep -E '^RUNPOD_|^PATH=|^_=' | sed 's/^\(.*\)=\(.*\)$/export \1="\2"/' >> /etc/rp_environment
+echo 'source /etc/rp_environment' >> ~/.bashrc
+
+if [[ $PUBLIC_KEY ]]
+then
+    mkdir -p ~/.ssh
+    chmod 700 ~/.ssh
+    echo $PUBLIC_KEY >> ~/.ssh/authorized_keys
+    chmod 700 -R ~/.ssh
+    # Start the SSH service in the background
+    service ssh start
+else
+    echo "No PUBLIC_KEY ENV variable provided, not starting openSSH daemon"
+fi
+
+# Execute the passed arguments (CMD)
+exec "$@"

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 +1,13 @@
+"""Module containing Dataset functionality"""
+
+import logging
+import os
 from typing import List
 
 import torch
-from datasets import IterableDataset
-from .prompt_tokenizers import PromptTokenizingStrategy, InvalidDataException
+from datasets import Dataset, IterableDataset
 
+from .prompt_tokenizers import 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
@@ -11,24 +15,35 @@ from .prompt_tokenizers import PromptTokenizingStrategy, InvalidDataException
 # let's check to ensure we don't truncate an item in the middle, we'll use
 # the collators later on to pad the datasets
 
+LOG = logging.getLogger("axolotl")
 
-class TokenizedPromptDataset(IterableDataset):
-    def __init__(
+
+class TokenizedPromptDataset(Dataset):
+    """
+    Dataset that returns tokenized prompts from a stream of text files.
+        Args:
+            prompt_tokenizer (PromptTokenizingStrategy): The prompt tokenizing method for proccessing the data.
+            dataset (dataset.Dataset): Dataset with text files.
+    """
+
+    def __init__(  # pylint: disable=super-init-not-called
         self,
         prompt_tokenizer: PromptTokenizingStrategy,
         dataset: IterableDataset,
+        **kwargs,
     ):
         self.prompt_tokenizer = prompt_tokenizer
-        self.dataset = dataset
+        super().__init__(self.process(dataset).data, **kwargs)
+
+    def process(self, dataset):
+        features = dataset.features.keys()
+        num_proc = min(64, os.cpu_count())
+        return dataset.map(
+            self.prompt_tokenizer.tokenize_prompt,
+            num_proc=num_proc,
+            remove_columns=features,
+        )
 
-    def __iter__(self):
-        iterator = iter(self.dataset)
-        # Loop through the entire dataset
-        for example in iterator:
-            try:
-                yield self.prompt_tokenizer.tokenize_prompt(example)
-            except InvalidDataException:
-                pass
 
 # 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,
@@ -60,14 +76,21 @@ class ConstantLengthDataset(IterableDataset):
             self.tokens_dtype = torch.int64
 
     def __iter__(self):
-        buffer = {"input_ids": [], "attention_mask": [], "labels": []}
+        buffer = {
+            "input_ids": [],
+            "attention_mask": [],
+            "labels": [],
+            "position_ids": [],
+        }
         buffer_len = 0
         for dataset in self.datasets:
+            idx = 0
             iterator = iter(dataset)
             more_examples = True
             while more_examples:
                 try:
                     example = next(iterator)
+                    idx += 1
                 except StopIteration:
                     more_examples = False
                     example = None
@@ -79,10 +102,8 @@ 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)[
@@ -91,16 +112,34 @@ class ConstantLengthDataset(IterableDataset):
                         attention_mask = torch.cat(buffer["attention_mask"], dim=-1)[
                             : self.seq_length
                         ]
+                        position_ids = torch.cat(buffer["position_ids"], dim=-1)[
+                            : 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,
+                                "position_ids": position_ids,
+                            }
+                        else:
+                            LOG.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": [],
+                        "position_ids": [],
+                    }
                     buffer_len = 0
+                    idx = 1
 
                 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"]
@@ -112,13 +151,21 @@ class ConstantLengthDataset(IterableDataset):
                             attention_mask.append(1)
                             labels.append(self.concat_token_id)
 
-                        input_ids_with_concat = torch.tensor(input_ids, dtype=self.tokens_dtype)
-                        attention_mask_with_concat = torch.tensor(
-                            attention_mask, dtype=self.tokens_dtype
+                        input_ids_with_concat = torch.tensor(
+                            input_ids, dtype=self.tokens_dtype
+                        )
+                        attention_mask_with_concat = torch.tensor(
+                            [idx * m for m in attention_mask], dtype=torch.int16
+                        )
+                        labels_with_concat = torch.tensor(
+                            labels, dtype=self.tokens_dtype
+                        )
+                        position_ids = torch.arange(
+                            len(input_ids), 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)
                         buffer["labels"].append(labels_with_concat)
+                        buffer["position_ids"].append(position_ids)
                         buffer_len += len(input_ids)

src/axolotl/logging_config.py

@@ -0,0 +1,33 @@
+"""Logging configuration settings"""
+
+import os
+import sys
+from logging.config import dictConfig
+from typing import Any, Dict
+
+DEFAULT_LOGGING_CONFIG: Dict[str, Any] = {
+    "version": 1,
+    "formatters": {
+        "simple": {
+            "format": "[%(asctime)s] [%(levelname)s] [%(name)s.%(funcName)s:%(lineno)d] [PID:%(process)d] %(message)s",
+        },
+    },
+    "filters": {},
+    "handlers": {
+        "console": {
+            "class": "logging.StreamHandler",
+            "formatter": "simple",
+            "filters": [],
+            "stream": sys.stdout,
+        },
+    },
+    "root": {"handlers": ["console"], "level": os.getenv("LOG_LEVEL", "INFO")},
+    "loggers": {
+        "axolotl": {"handlers": ["console"], "level": "DEBUG", "propagate": False},
+    },
+}
+
+
+def configure_logging():
+    """Configure with default logging"""
+    dictConfig(DEFAULT_LOGGING_CONFIG)

src/axolotl/monkeypatch/llama_attn_hijack_flash.py

@@ -1,17 +1,24 @@
+"""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 einops import rearrange
+from flash_attn.bert_padding import pad_input, unpad_input
+
+try:
+    from flash_attn.flash_attn_interface import flash_attn_varlen_qkvpacked_func
+except ImportError:
+    from flash_attn.flash_attn_interface import (
+        flash_attn_unpadded_qkvpacked_func as flash_attn_varlen_qkvpacked_func,
+    )
+
 from transformers.models.llama.modeling_llama import apply_rotary_pos_emb
 
-from einops import rearrange
-
-from flash_attn.flash_attn_interface import flash_attn_unpadded_qkvpacked_func
-from flash_attn.bert_padding import unpad_input, pad_input
+from axolotl.monkeypatch.utils import get_cu_seqlens_from_pos_ids
 
 
 def forward(
@@ -27,6 +34,7 @@ def forward(
 
     attention_mask: [bsz, q_len]
     """
+    # pylint: disable=duplicate-code
     bsz, q_len, _ = hidden_states.size()
 
     query_states = (
@@ -74,43 +82,79 @@ 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(
+        output = flash_attn_varlen_qkvpacked_func(
+            qkv, cu_q_lens, max_s, 0.0, softmax_scale=None, causal=True
+        )
+        output = rearrange(output, "(b s) ... -> b s ...", b=bsz)
+    elif attention_mask.shape[0] == 1:
+        # special handling using sample packing
+        qkv = rearrange(qkv, "b s ... -> (b s) ...")
+        cu_q_lens, max_s = get_cu_seqlens_from_pos_ids(position_ids)
+        cu_q_lens = cu_q_lens.squeeze()
+
+        output = flash_attn_varlen_qkvpacked_func(
             qkv, cu_q_lens, max_s, 0.0, softmax_scale=None, causal=True
         )
         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
+        output_unpad = flash_attn_varlen_qkvpacked_func(
+            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 +1,284 @@
+"""
+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 torch.nn.functional as F
+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()
+
+    if not hasattr(self, "pretraining_tp"):
+        self.pretraining_tp = 1
+
+    if self.pretraining_tp > 1:
+        key_value_slicing = (
+            self.num_key_value_heads * self.head_dim
+        ) // self.pretraining_tp
+        query_slices = self.q_proj.weight.split(
+            (self.num_heads * self.head_dim) // self.pretraining_tp, dim=0
+        )
+        key_slices = self.k_proj.weight.split(key_value_slicing, dim=0)
+        value_slices = self.v_proj.weight.split(key_value_slicing, dim=0)
+
+        query_states = [
+            F.linear(hidden_states, query_slices[i]) for i in range(self.pretraining_tp)
+        ]
+        query_states = torch.cat(query_states, dim=-1)
+
+        key_states = [
+            F.linear(hidden_states, key_slices[i]) for i in range(self.pretraining_tp)
+        ]
+        key_states = torch.cat(key_states, dim=-1)
+
+        value_states = [
+            F.linear(hidden_states, value_slices[i]) for i in range(self.pretraining_tp)
+        ]
+        value_states = torch.cat(value_states, dim=-1)
+
+    else:
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+    query_states = query_states.view(
+        bsz, q_len, self.num_heads, self.head_dim
+    ).transpose(1, 2)
+    key_states = key_states.view(
+        bsz, q_len, self.num_key_value_heads, self.head_dim
+    ).transpose(1, 2)
+    value_states = value_states.view(
+        bsz, q_len, self.num_key_value_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
+
+    # repeat k/v heads if n_kv_heads < n_heads
+    key_states = transformers.models.llama.modeling_llama.repeat_kv(
+        key_states, self.num_key_value_groups
+    )
+    value_states = transformers.models.llama.modeling_llama.repeat_kv(
+        value_states, self.num_key_value_groups
+    )
+
+    # 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=attention_mask,
+                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).contiguous()
+        # end x-formers vs. not x-formers if-else block
+
+    attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+
+    if self.pretraining_tp > 1:
+        attn_output = attn_output.split(self.hidden_size // self.pretraining_tp, dim=2)
+        o_proj_slices = self.o_proj.weight.split(
+            self.hidden_size // self.pretraining_tp, dim=1
+        )
+        attn_output = sum(
+            F.linear(attn_output[i], o_proj_slices[i])
+            for i in range(self.pretraining_tp)
+        )
+    else:
+        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:
+        with torch.backends.cuda.sdp_kernel():
+            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/llama_expand_mask.py

@@ -0,0 +1,52 @@
+"""
+expands the binary attention mask per 3.2.2 of https://arxiv.org/pdf/2107.02027.pdf
+"""
+from typing import Optional
+
+import torch
+
+
+def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
+    """
+    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
+    This expansion handles packed sequences so that sequences share the same attention mask integer value
+    when they attend to each other within that sequence.
+    This expansion transforms the mask to lower triangular form to prevent future peeking.
+    """
+    bsz, src_len = mask.size()
+    tgt_len = tgt_len if tgt_len is not None else src_len
+
+    mask = mask.unsqueeze(1).unsqueeze(2)
+    mask = mask.expand(bsz, 1, tgt_len, src_len)
+
+    # Create a binary mask from the original mask where zeros remain zeros and all other values are set to one
+    binary_mask = torch.where(
+        mask != 0,
+        torch.tensor(1).to(dtype),
+        torch.tensor(0).to(dtype),
+    )
+
+    # Create a block-diagonal mask.
+    # we multiply by the binary mask so that 0's in the original mask are correctly excluded
+    zero_one_mask = torch.eq(mask, mask.transpose(-1, -2)).int() * binary_mask
+
+    # Now let's create a lower triangular mask of ones that will zero out the upper triangular part
+    lower_triangular_ones = torch.tril(torch.ones((tgt_len, src_len), dtype=dtype)).to(
+        mask.device
+    )
+
+    # Use the lower triangular mask to zero out the upper triangular part of the zero_one_mask
+    masked_zero_one_mask = zero_one_mask * lower_triangular_ones
+    inverted_mask = 1.0 - masked_zero_one_mask
+
+    return inverted_mask.masked_fill(
+        inverted_mask.to(torch.bool), torch.finfo(dtype).min
+    )
+
+
+def hijack_expand_mask():
+    import transformers
+
+    transformers.models.llama.modeling_llama._expand_mask = (  # pylint: disable=protected-access
+        _expand_mask
+    )

src/axolotl/monkeypatch/utils.py

@@ -0,0 +1,103 @@
+"""
+Shared utils for the monkeypatches
+"""
+import torch
+
+
+def get_cu_seqlens(attn_mask):
+    """generate a cumulative sequence length mask for flash attention using attn mask"""
+    if len(attn_mask.shape) == 1:
+        attn_mask = attn_mask.unsqueeze(0)
+
+    device = attn_mask.device
+    results = []
+    max_seq_lens = []
+
+    for row in attn_mask:
+        # Exclude zeros to avoid adding their positions to the mask
+        t_non_zeros = row[row != 0]
+        # Find where the sequence number changes (including the first position)
+        seq_change = torch.cat(
+            [
+                torch.tensor([1], dtype=torch.int32, device=device),
+                t_non_zeros[1:] != t_non_zeros[:-1],
+            ]
+        )
+        # Get the indices where the sequence changes
+        change_indices = torch.cat(
+            [
+                (seq_change == 1).nonzero(as_tuple=True)[0],
+                torch.tensor([len(t_non_zeros)], dtype=torch.int32, device=device),
+            ]
+        )
+        # Calculate the sequence lengths
+        seq_lengths = change_indices[1:] - change_indices[:-1]
+        # Calculate the length of the final sequence or padding
+        final_seq_length = len(row) - change_indices[-1]
+        # Append the length of the final sequence or padding to seq_lengths
+        if final_seq_length.item():
+            seq_lengths = torch.cat(
+                [
+                    seq_lengths,
+                    torch.tensor(
+                        [final_seq_length.item()], dtype=torch.int32, device=device
+                    ),
+                ]
+            )
+        # Calculate the cumulative sequence lengths
+        cu_seqlens = torch.cat(
+            [torch.tensor([0], dtype=torch.int32, device=device), seq_lengths.cumsum(0)]
+        )
+        max_seq_len = (cu_seqlens[1:] - cu_seqlens[:-1]).max()
+        results.append(cu_seqlens)
+        max_seq_lens.append(max_seq_len)
+
+    return torch.stack(results).to(dtype=torch.int32), torch.stack(max_seq_lens)
+
+
+def get_cu_seqlens_from_pos_ids(position_ids):
+    """generate a cumulative sequence length mask for flash attention using pos ids"""
+    if len(position_ids.shape) == 1:
+        position_ids = position_ids.unsqueeze(0)
+
+    device = position_ids.device
+    results = []
+    max_seq_lens = []
+
+    for row in position_ids:
+        # Count the number of consecutive zeros from the right side
+        padding_length = (row == 0).int().flip(dims=[0]).cumprod(dim=0).sum().item()
+
+        # Adjust the row to exclude padding
+        adjusted_row = row[:-padding_length] if padding_length else row.clone()
+
+        # Find where the position resets to 0 (indicating a new sequence)
+        seq_starts = torch.cat(
+            [
+                torch.tensor([True], dtype=torch.bool, device=device),
+                adjusted_row[1:] == 0,
+            ]
+        )
+        # Get the indices where the sequence starts
+        start_indices = torch.cat(
+            [
+                (seq_starts).nonzero(as_tuple=True)[0],
+                torch.tensor([len(adjusted_row)], dtype=torch.int32, device=device),
+            ]
+        )
+        # Calculate the sequence lengths
+        seq_lengths = start_indices[1:] - start_indices[:-1]
+        # Calculate the cumulative sequence lengths
+        cu_seqlens = torch.cat(
+            [torch.tensor([0], dtype=torch.int32, device=device), seq_lengths.cumsum(0)]
+        )
+        # Append the padding length to the cumulative sequence lengths
+        if padding_length:
+            cu_seqlens = torch.cat(
+                [cu_seqlens, torch.tensor([len(row)], dtype=torch.int32, device=device)]
+            )
+        max_seq_len = (cu_seqlens[1:] - cu_seqlens[:-1]).max()
+        results.append(cu_seqlens)
+        max_seq_lens.append(max_seq_len)
+
+    return torch.stack(results).to(dtype=torch.int32), torch.stack(max_seq_lens)

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 +1,72 @@
+"""Module loading the AlpacaInstructPromptTokenizingStrategy class"""
+import logging
+
+from axolotl.prompt_tokenizers import AlpacaPromptTokenizingStrategy
+from axolotl.prompters import AlpacaPrompter, PromptStyle, UnpromptedPrompter
+
+LOG = logging.getLogger("axolotl.prompt_strategies.alpaca_instruct")
+
+
+class LatentSpaceAlpacaPromptTokenizingStrategy(AlpacaPromptTokenizingStrategy):
+    """
+    Overrides the tokenization to include additional padding tokens as
+    latent space on the inputs
+    """
+
+    def _tokenize(self, prompt: str, add_eos_token=True, strip_bos_token=False):
+        # pylint: disable=duplicate-code
+        result = self.tokenizer(
+            prompt,
+            truncation=True,
+            max_length=self.sequence_len,
+            padding=False,
+            return_tensors=None,
+        )
+        if len(result["input_ids"]) == 0:
+            LOG.warning("Tokenizer result is empty. You may want to audit your dataset")
+        if (
+            len(result["input_ids"]) > 0
+            and 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:]
+
+        # latent space
+        if add_eos_token and not strip_bos_token:
+            result["input_ids"].extend([self.tokenizer.pad_token_id] * 100)
+
+        result["labels"] = result["input_ids"].copy()
+        return result
+
+
+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,
+    )
+
+
+def load_latent_space(tokenizer, cfg):
+    return LatentSpaceAlpacaPromptTokenizingStrategy(
+        AlpacaPrompter(PromptStyle.INSTRUCT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )

src/axolotl/prompt_strategies/alpaca_w_system.py

@@ -0,0 +1,161 @@
+"""
+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.
+        formatted_sys_prompt = (
+            self.system_format.format(system=system)
+            if system and self.system_format
+            else ""
+        )
+        if input:
+            res = formatted_sys_prompt + self.turn_format.format(
+                instruction=instruction, input=input
+            )
+        else:
+            res = formatted_sys_prompt + self.turn_no_input_format.format(
+                instruction=instruction
+            )
+        if output:
+            res = f"{res}{output}"
+        yield res
+
+
+class OpenOrcaSystemDataPrompter(SystemDataPrompter):
+    """
+    Alpaca Style Prompter that uses system prompts from the dataset, with OpenOrca prompts
+    """
+
+    def match_prompt_style(self):
+        # pylint: disable=duplicate-code
+        if self.prompt_style == PromptStyle.INSTRUCT.value:
+            self.turn_format = "### Human:\n{instruction}\n### Additional Context:\n{input}\n### Assistant:\n"
+            self.turn_no_input_format = "### Human:\n{instruction}\n### Assistant:\n"
+            self.system_format = "### System:\n{system}\n"
+        if self.prompt_style == PromptStyle.CHAT.value:
+            self.turn_format = "USER: {instruction}\n{input}\nASSISTANT:"
+            self.turn_no_input_format = "USER: {instruction}\nASSISTANT:"
+            self.system_format = "SYSTEM: {system}\n"
+        if self.prompt_style == PromptStyle.CHATML.value:
+            self.turn_format = "<|im_start|>user\n{instruction}\n{input}<|im_end|>\n<|im_start|>assistant\n"
+            self.turn_no_input_format = (
+                "<|im_start|>user\n{instruction}<|im_end|>\n<|im_start|>assistant\n"
+            )
+            self.system_format = "<|im_start|>system\n{system}<|im_end|>\n"
+
+
+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(
+        OpenOrcaSystemDataPrompter(PromptStyle.INSTRUCT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+def load_open_orca_chatml(tokenizer, cfg):
+    return OpenOrcaPromptTokenizingStrategy(
+        OpenOrcaSystemDataPrompter(PromptStyle.CHATML.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/llama2_chat.py

@@ -0,0 +1,205 @@
+"""
+Prompt Strategy for finetuning Llama2 chat models
+see also https://github.com/facebookresearch/llama/blob/6c7fe276574e78057f917549435a2554000a876d/llama/generation.py#L213 for ma reference implementation.
+
+This implementation is based on the Vicuna PR and the fastchat repo, see also:
+https://github.com/lm-sys/FastChat/blob/cdd7730686cb1bf9ae2b768ee171bdf7d1ff04f3/fastchat/conversation.py#L847
+
+Use dataset type: "llama2_chat" in conig.yml to use this prompt style.
+
+E.g. in the config.yml:
+```
+datasets:
+  - path: llama_finetune_train.jsonl
+    type: llama2_chat
+```
+
+The dataset itself should look like this:
+```
+{'conversations':[{"from": "human", "value": "Who are you?"}, {"from": "gpt", "value": "I am Vicuna"},...]}
+```
+in a jsonl file. The first message should be from the human, the second from gpt.
+For a custom system message, the first "from" can be "system" (followed by alternating "human" and "gpt" turns).
+
+Important: Don't use "special_tokens:" in your config.yml if you are not sure what you are doing!
+"""
+
+import logging
+from dataclasses import dataclass, field
+from typing import Generator, List, Sequence
+
+from axolotl.prompt_tokenizers import PromptTokenizingStrategy
+from axolotl.prompters import IGNORE_TOKEN_ID, SHAREGPT_ASSERTION_FAILED_ROLE
+
+
+@dataclass
+class Llama2ChatConversation:
+    """A class that manages prompt templates and keeps all conversation history.
+    copied from https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py"""
+
+    name: str = "llama2"
+    # The system prompt
+    system: str = (
+        "[INST] <<SYS>>\nYou are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. "
+        "Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. "
+        "Please ensure that your responses are socially unbiased and positive in nature.\n\n"
+        "If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. "
+        "If you don't know the answer to a question, please don't share false information.\n<</SYS>>\n\n"
+    )
+    roles: Sequence[str] = ("[INST]", "[/INST]")
+    messages: List[List[str]] = field(default_factory=list)
+    offset: int = 0
+    sep = " "
+    sep2 = " </s><s>"
+    stop_token_ids = [2]
+
+    def get_prompt(self) -> str:
+        """Get the prompt for generation."""
+        seps = [self.sep, self.sep2]
+        ret = ""
+        for i, (role, message) in enumerate(self.messages):
+            if (i == len(self.messages) - 1) and (role == self.roles[0]):
+                # last message is from user (due to length),
+                #  return prompt without it for training
+                return ret
+            if i == 0:
+                ret += self.system + message.strip()
+            else:
+                ret += role + " " + message.strip() + seps[i % 2]
+        return ret
+
+    def append_message(self, role: str, message: str):
+        """Append a new message."""
+        self.messages.append([role, message])
+
+
+class LLama2ChatTokenizingStrategy(PromptTokenizingStrategy):
+    """
+    Tokenizing strategy for ShareGPT prompts.
+    adapted from https://github.com/lm-sys/FastChat/blob/main/fastchat/train/train.py
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.sequence_len = 4096
+        self.tokenizer.add_special_tokens({"pad_token": "<pad>"})
+        # https://huggingface.co/meta-llama/Llama-2-7b-chat-hf/blob/main/added_tokens.json
+
+    def tokenize_prompt(self, prompt):
+        conv = next(self.prompter.build_prompt(prompt))
+        conversation_str = conv.get_prompt()
+
+        # Tokenize conversations
+        input_ids = self.tokenizer(
+            conversation_str,
+            return_tensors="pt",
+            padding="max_length",
+            max_length=self.sequence_len,
+            truncation=True,
+        ).input_ids[0]
+        target = input_ids.clone()
+
+        # Mask targets. Only compute loss on the assistant outputs.
+        sep = conv.roles[1]
+
+        total_len = int(target.ne(self.tokenizer.pad_token_id).sum())
+
+        turns = conversation_str.split(conv.sep2)
+        cur_len = 1
+        target[:cur_len] = IGNORE_TOKEN_ID
+        for turn in turns:
+            if turn == "":
+                break
+            turn_len = len(self.tokenizer(turn).input_ids)
+
+            parts = turn.split(sep)
+            if len(parts) != 2:
+                break
+            parts[0] += sep
+            # "-1" is hardcoded for the LLaMA tokenizer to make the offset correct.
+            instruction_len = len(self.tokenizer(parts[0]).input_ids) - 1
+
+            # Ignore the user instructions
+            target[cur_len - 1 : cur_len + instruction_len] = IGNORE_TOKEN_ID
+            cur_len += turn_len + 2  # due to length of role token
+
+        target[cur_len:] = IGNORE_TOKEN_ID
+
+        if cur_len < self.sequence_len:
+            if cur_len != total_len:
+                target[:] = IGNORE_TOKEN_ID
+                logging.warning(
+                    f"WARNING: tokenization mismatch: {cur_len} vs. {total_len}."
+                    f" (ignored)"
+                )
+
+        attention_mask = input_ids.ne(self.tokenizer.pad_token_id).tolist()
+        input_ids = input_ids.tolist()
+        target = target.tolist()
+        # this is a fix for the tokenizer which tokenizes [ differently with eos tokens and
+        # follows the original llama implementation
+        for i in range(2, total_len - 2):
+            if input_ids[i] == 29961:
+                input_ids[i] = 518
+            if target[i] == 29961:
+                target[i] = 518
+        return {
+            "input_ids": input_ids,
+            "labels": target,
+            "attention_mask": attention_mask,
+        }
+
+
+class Llama2ChatPrompter:  # pylint: disable=too-few-public-methods
+    """
+    A prompter that generates prompts for Llama2 models.
+    """
+
+    system_prompt = (
+        "[INST] <<SYS>>\nYou are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. "
+        "Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. "
+        "Please ensure that your responses are socially unbiased and positive in nature.\n\n"
+        "If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. "
+        "If you don't know the answer to a question, please don't share false information.\n<</SYS>>\n\n"
+    )
+
+    def build_prompt(self, source) -> Generator[Llama2ChatConversation, None, None]:
+        # see https://github.com/lm-sys/FastChat/blob/da0641e567cf93756b0978ab5a6b092e96f06240/fastchat/train/train.py#L78
+        source = source["conversations"]  # fix data structure for datasets
+
+        # if system prompt provided, use it
+        if source[0]["from"] == "system":
+            system = f"[INST] <<SYS>>\n{source[0]['value']}\n<</SYS>>\n\n"
+            source = source[1:]
+        else:
+            system = self.system_prompt
+
+        conv = Llama2ChatConversation(system=system)
+
+        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
+
+        roles = {"human": conv.roles[0], "gpt": conv.roles[1]}
+
+        if roles[source[0]["from"]] != conv.roles[0]:
+            # Skip the first one if it is not from human
+            source = source[1:]
+
+        conv.messages = []  # pylint: disable=R0801
+        for j, sentence in enumerate(source):
+            role = roles[sentence["from"]]
+            assert role == conv.roles[j % 2], SHAREGPT_ASSERTION_FAILED_ROLE
+            if sentence["value"]:
+                conv.append_message(role, sentence["value"])
+        yield conv
+
+
+def load(tokenizer, cfg) -> LLama2ChatTokenizingStrategy:
+    return LLama2ChatTokenizingStrategy(
+        Llama2ChatPrompter(),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )

src/axolotl/prompt_strategies/orcamini.py

@@ -0,0 +1,46 @@
+"""
+Prompt Strategy for finetuning Orca Mini (v2) models
+see also https://huggingface.co/psmathur/orca_mini_v2_7b for more information
+
+Use dataset type: orcamini in conig.yml to use this prompt style.
+
+Compared to the alpaca_w_system.open_orca dataset type,
+this one specifies the system prompt with "### System:".
+
+Not suited/tested for multiple-turn conversations without further adjustments.
+"""
+from typing import Generator, Union
+
+from axolotl.prompt_strategies.alpaca_w_system import OpenOrcaPromptTokenizingStrategy
+from axolotl.prompters import AlpacaPrompter
+
+
+class OrcaMiniPrompter(AlpacaPrompter):
+    """Adjusted Prompter for Orca Mini (v2) datasets"""
+
+    def match_prompt_style(self):
+        self.turn_no_input_format = (
+            "### System:\n{system}\n\n### User:\n{instruction}\n\n### Response:\n"
+        )
+
+    def build_prompt_w_system(
+        self,
+        system: str,
+        instruction: str,
+        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.
+        res = self.turn_no_input_format.format(system=system, instruction=instruction)
+        if output:
+            res = f"{res}{output}"
+        yield res
+
+
+def load(tokenizer, cfg):
+    return OpenOrcaPromptTokenizingStrategy(
+        OrcaMiniPrompter(),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )

src/axolotl/prompt_strategies/pygmalion.py

@@ -0,0 +1,101 @@
+"""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,
+)
+
+LOG = logging.getLogger("axolotl")
+
+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:
+                LOG.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 +1,113 @@
+"""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,
+    )
+
+
+def load_latent_space(tokenizer, cfg):
+    return LatentSpaceShareGPTPromptTokenizingStrategy(
+        ShareGPTPrompter(PromptStyle.CHAT.value),
+        tokenizer,
+        cfg.train_on_inputs,
+        cfg.sequence_len,
+    )
+
+
+class LatentSpaceShareGPTPromptTokenizingStrategy(ShareGPTPromptTokenizingStrategy):
+    """
+    latent space padded sharegpt strategy to grab conversations from the sample row
+    """
+
+    def get_conversation_thread(self, prompt):
+        return prompt["conversations"]
+
+    def _tokenize(self, prompt, add_eos_token=True, strip_bos_token=False):
+        # pylint: disable=duplicate-code
+        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:]
+
+        # latent space
+        if add_eos_token and not strip_bos_token:
+            result["input_ids"].extend([self.tokenizer.pad_token_id] * 100)
+
+        result["labels"] = result["input_ids"].copy()
+        return result
+
+
+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 +1,35 @@
+"""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
+
+LOG = logging.getLogger("axolotl")
+
 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 +46,27 @@ class PromptTokenizingStrategy(abc.ABC):
     def tokenize_prompt(self, prompt):
         pass
 
+    @functools.lru_cache(maxsize=128)
+    def _get_user_token(self):
+        try:
+            id_or_ids = self.tokenizer.convert_tokens_to_ids("<|USER|>")
+            if isinstance(id_or_ids, (int,)):
+                return id_or_ids
+        except KeyError:
+            pass
+        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):
+        try:
+            id_or_ids = self.tokenizer.convert_tokens_to_ids("<|ASSISTANT|>")
+            if isinstance(id_or_ids, (int,)):
+                return id_or_ids
+        except KeyError:
+            pass
+        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,
@@ -68,20 +74,83 @@ class InstructionPromptTokenizingStrategy(PromptTokenizingStrategy):
             padding=False,
             return_tensors=None,
         )
+        if len(result["input_ids"]) == 0:
+            LOG.warning("Tokenizer result is empty. You may want to audit your dataset")
         if (
-            result["input_ids"][-1] != self.tokenizer.eos_token_id
+            len(result["input_ids"]) > 0
+            and 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
 
 
+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 +158,38 @@ 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 +197,25 @@ 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 +223,64 @@ 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 +291,22 @@ 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 +327,11 @@ 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 +340,131 @@ 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:
+                        LOG.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 +1,213 @@
-import copy
-import dataclasses
-from enum import auto, Enum
-from typing import List, Tuple, Any, Union
+"""Module containing prompters"""
 
+import dataclasses
+import logging
+from enum import Enum, auto
+from typing import Generator, List, Optional, Tuple, Union
+
+LOG = logging.getLogger("axolotl")
 IGNORE_TOKEN_ID = -100
 
 
+class PromptStyle(Enum):
+    """
+    Enum for prompt styles
+    """
+
+    INSTRUCT = "instruct"
+    CHAT = "chat"
+    CHATML = "chatml"
+
+
 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"
+    system_format: str
+    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):
+        # pylint: disable=duplicate-code
+        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"
+            )
+            self.system_format = "### System:\n{system}\n\n"
+        if self.prompt_style == PromptStyle.CHAT.value:
+            self.turn_format = "USER: {instruction}\n{input}\nASSISTANT:"
+            self.turn_no_input_format = "USER: {instruction}\nASSISTANT:"
+            self.system_format = "SYSTEM: {system}\n"
+        if self.prompt_style == PromptStyle.CHATML.value:
+            self.turn_format = "<|im_start|>user\n{instruction}\n{input}<|im_end|>\n<|im_start|>assistant\n"
+            self.turn_no_input_format = (
+                "<|im_start|>user\n{instruction}<|im_end|>\n<|im_start|>assistant\n"
+            )
+            self.system_format = "<|im_start|>system\n{system}<|im_end|>\n"
 
     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 +215,13 @@ 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 +243,18 @@ 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
+                LOG.warning(f"role with empty message: {role}")
+                yield (role + ":", "")
 
     def copy(self):
         return Conversation(
@@ -110,26 +271,52 @@ 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>",
+SHAREGPT_ASSERTION_FAILED_ROLE = (
+    "Role did not alternate between turns (gpt and human). Please check your data."
 )
 
 
-class ShareGPTPrompter:
-    def build_prompt(self, source, tokenizer):
+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)
+
         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
+            raise IndexError(
+                f"A conversation entry has less than 2 messages :\n{source}"
+            )
 
-        conv = conv_vicuna_v1_1.copy()
+        conv = self._conversation.copy()
         roles = {"human": conv.roles[0], "gpt": conv.roles[1]}
 
         try:
@@ -140,55 +327,15 @@ 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]
+            assert role == conv.roles[j % 2], SHAREGPT_ASSERTION_FAILED_ROLE
             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/bench.py

@@ -0,0 +1,43 @@
+"""Benchmarking and measurement utilities"""
+
+import pynvml
+import torch
+
+
+def gpu_memory_usage(device=0):
+    return torch.cuda.memory_allocated(device) / 1024.0**3
+
+
+def gpu_memory_usage_all(device=0):
+    usage = torch.cuda.memory_allocated(device) / 1024.0**3
+    reserved = torch.cuda.memory_reserved(device) / 1024.0**3
+    smi = gpu_memory_usage_smi(device)
+    return usage, reserved - usage, max(0, smi - reserved)
+
+
+def gpu_memory_usage_smi(device=0):
+    if isinstance(device, torch.device):
+        device = device.index
+    if isinstance(device, str) and device.startswith("cuda:"):
+        device = int(device[5:])
+
+    pynvml.nvmlInit()
+    handle = pynvml.nvmlDeviceGetHandleByIndex(device)
+    info = pynvml.nvmlDeviceGetMemoryInfo(handle)
+    return info.used / 1024.0**3
+
+
+def log_gpu_memory_usage(log, msg, device):
+    if not torch.cuda.is_available():
+        return (0, 0, 0)
+
+    usage, cache, misc = gpu_memory_usage_all(device)
+    extras = []
+    if cache > 0:
+        extras.append(f"+{cache:.03f}GB cache")
+    if misc > 0:
+        extras.append(f"+{misc:.03f}GB misc")
+    log.info(
+        f"GPU memory usage {msg}: {usage:.03f}GB ({', '.join(extras)})", stacklevel=2
+    )
+    return usage, cache, misc

src/axolotl/utils/callbacks.py

@@ -0,0 +1,96 @@
+"""Callbacks for Trainer class"""
+
+import logging
+import os
+
+from optimum.bettertransformer import BetterTransformer
+from transformers import (
+    TrainerCallback,
+    TrainerControl,
+    TrainerState,
+    TrainingArguments,
+)
+from transformers.trainer_utils import PREFIX_CHECKPOINT_DIR, IntervalStrategy
+
+from axolotl.utils.bench import log_gpu_memory_usage
+
+LOG = logging.getLogger("axolotl.callbacks")
+
+
+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
+
+
+class GPUStatsCallback(
+    TrainerCallback
+):  # pylint: disable=too-few-public-methods disable=unused-argument
+    """Callback to track GPU utilization"""
+
+    def __init__(self, cfg):
+        self.cfg = cfg
+        self.logged = False
+
+    def on_step_end(
+        self,
+        args: TrainingArguments,
+        state: TrainerState,
+        control: TrainerControl,
+        **kwargs,
+    ):
+        if not self.logged and state.global_step > 1:
+            log_gpu_memory_usage(LOG, "while training", self.cfg.device)
+            self.logged = True
+        return control

src/axolotl/utils/collators.py

@@ -0,0 +1,121 @@
+"""
+DataCollator for axolotl to pad labels and position_ids for packed sequences
+"""
+from dataclasses import dataclass
+from typing import Any, Optional, Union
+
+import numpy as np
+from transformers import PreTrainedTokenizerBase
+from transformers.utils import PaddingStrategy
+
+
+@dataclass
+class DataCollatorForSeq2Seq:
+    """
+    Data collator that will dynamically pad the inputs received, as well as the labels and position_ids
+
+    Args:
+        tokenizer ([`PreTrainedTokenizer`] or [`PreTrainedTokenizerFast`]):
+            The tokenizer used for encoding the data.
+        model ([`PreTrainedModel`]):
+            The model that is being trained. If set and has the *prepare_decoder_input_ids_from_labels*, use it to
+            prepare the *decoder_input_ids*
+
+            This is useful when using *label_smoothing* to avoid calculating loss twice.
+        padding (`bool`, `str` or [`~utils.PaddingStrategy`], *optional*, defaults to `True`):
+            Select a strategy to pad the returned sequences (according to the model's padding side and padding index)
+            among:
+
+            - `True` or `'longest'` (default): Pad to the longest sequence in the batch (or no padding if only a single
+              sequence is provided).
+            - `'max_length'`: Pad to a maximum length specified with the argument `max_length` or to the maximum
+              acceptable input length for the model if that argument is not provided.
+            - `False` or `'do_not_pad'`: No padding (i.e., can output a batch with sequences of different lengths).
+        max_length (`int`, *optional*):
+            Maximum length of the returned list and optionally padding length (see above).
+        pad_to_multiple_of (`int`, *optional*):
+            If set will pad the sequence to a multiple of the provided value.
+
+            This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability >=
+            7.5 (Volta).
+        label_pad_token_id (`int`, *optional*, defaults to -100):
+            The id to use when padding the labels (-100 will be automatically ignored by PyTorch loss functions).
+        return_tensors (`str`):
+            The type of Tensor to return. Allowable values are "np", "pt" and "tf".
+    """
+
+    tokenizer: PreTrainedTokenizerBase
+    model: Optional[Any] = None
+    padding: Union[bool, str, PaddingStrategy] = True
+    max_length: Optional[int] = None
+    pad_to_multiple_of: Optional[int] = None
+    label_pad_token_id: int = -100
+    position_pad_token_id: int = 0
+    return_tensors: str = "pt"
+
+    def __call__(self, features, return_tensors=None):
+        labels = None
+        if return_tensors is None:
+            return_tensors = self.return_tensors
+
+        for feature_name, pad_token_id in [
+            ("labels", self.label_pad_token_id),
+            ("position_ids", self.position_pad_token_id),
+        ]:
+            feat = (
+                [feature[feature_name] for feature in features]
+                if feature_name in features[0].keys()
+                else None
+            )
+            labels = feat if feat and feature_name == "labels" else labels
+            # We have to pad the labels before calling `tokenizer.pad` as this method won't pad them and needs them of the
+            # same length to return tensors.
+            if feat is not None:
+                max_feature_length = max(len(l) for l in feat)  # noqa: E741
+                if self.pad_to_multiple_of is not None:
+                    max_feature_length = (
+                        (max_feature_length + self.pad_to_multiple_of - 1)
+                        // self.pad_to_multiple_of
+                        * self.pad_to_multiple_of
+                    )
+
+                padding_side = self.tokenizer.padding_side
+                for feature in features:
+                    remainder = [pad_token_id] * (
+                        max_feature_length - len(feature[feature_name])
+                    )
+                    if isinstance(feature[feature_name], list):
+                        feature[feature_name] = (
+                            feature[feature_name] + remainder
+                            if padding_side == "right"
+                            else remainder + feature[feature_name]
+                        )
+                    elif padding_side == "right":
+                        feature[feature_name] = np.concatenate(
+                            [feature[feature_name], remainder]
+                        ).astype(np.int64)
+                    else:
+                        feature[feature_name] = np.concatenate(
+                            [remainder, feature[feature_name]]
+                        ).astype(np.int64)
+
+        features = self.tokenizer.pad(
+            features,
+            padding=self.padding,
+            max_length=self.max_length,
+            pad_to_multiple_of=self.pad_to_multiple_of,
+            return_tensors=return_tensors,
+        )
+
+        # prepare decoder_input_ids
+        if (
+            labels is not None
+            and self.model is not None
+            and hasattr(self.model, "prepare_decoder_input_ids_from_labels")
+        ):
+            decoder_input_ids = self.model.prepare_decoder_input_ids_from_labels(
+                labels=features["labels"]
+            )
+            features["decoder_input_ids"] = decoder_input_ids
+
+        return features

src/axolotl/utils/config.py

@@ -0,0 +1,199 @@
+"""Module for working with config dicts"""
+
+import logging
+import os
+
+import torch
+
+from axolotl.utils.bench import log_gpu_memory_usage
+
+LOG = logging.getLogger("axolotl")
+
+
+def choose_device(cfg):
+    def get_device():
+        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_map != "auto":
+        if cfg.device.startswith("cuda"):
+            cfg.device_map = {"": cfg.local_rank}
+        else:
+            cfg.device_map = {"": cfg.device}
+
+    # in `accelerate launch`, we need to not pass through any device map and let
+    # accelerate figure out which parts of the model to put on which gpu
+    accelerate_vars = [var for var in os.environ if var.startswith("ACCELERATE_USE_")]
+    if accelerate_vars:
+        cfg.device_map = None
+
+
+def normalize_config(cfg):
+    # setup some derived config / hyperparams
+    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.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.batch_size = cfg.batch_size * cfg.world_size
+
+    if cfg.device == "mps":
+        cfg.load_in_8bit = False
+        cfg.tf32 = False
+        if cfg.bf16:
+            cfg.fp16 = True
+        cfg.bf16 = False
+    else:
+        torch.backends.cuda.matmul.allow_tf32 = cfg.tf32 or False
+
+    log_gpu_memory_usage(LOG, "baseline", cfg.device)
+
+
+def validate_config(cfg):
+    if cfg.max_packed_sequence_len and cfg.sample_packing:
+        raise ValueError(
+            "please set only one of max_packed_sequence_len (deprecated soon) or sample_packing"
+        )
+    if cfg.max_packed_sequence_len:
+        LOG.warning(
+            str(
+                PendingDeprecationWarning(
+                    "max_packed_sequence_len will be deprecated in favor of sample_packing"
+                )
+            )
+        )
+
+    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:
+        LOG.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":
+        LOG.warning("We recommend setting `load_in_8bit: true` for LORA finetuning")
+
+    if cfg.trust_remote_code:
+        LOG.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:
+            LOG.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:
+            LOG.warning(
+                "You should probably set bfloat16 or float16 to true to "
+                "load the model in float16 for BetterTransformers"
+            )
+        if int(torch.__version__.split(".", maxsplit=1)[0]) < 2:
+            LOG.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:
+        LOG.warning(
+            "You probably want to disable group_by_length as it will force a streamed dataset to download completely."
+        )
+
+    if any([cfg.adam_beta1, cfg.adam_beta2, cfg.adam_epsilon]) and (
+        not cfg.optimizer or "adamw" not in cfg.optimizer
+    ):
+        LOG.warning("adamw hyperparameters found, but no adamw optimizer set")
+
+    if cfg.push_to_hub_model_id:
+        raise ValueError(
+            "push_to_hub_model_id is deprecated. Please use hub_model_id instead."
+        )
+
+    if cfg.gptq and cfg.model_revision:
+        raise ValueError(
+            "model_revision is not supported for GPTQ models. "
+            + "Please download the model from HuggingFace Hub manually for correct branch, "
+            + "point to its path, and remove model_revision from the config."
+        )
+
+    if cfg.sample_packing and cfg.sdp_attention:
+        # incompatible due to bug w/ accelerate causing 0.0 loss when using llama2
+        raise ValueError(
+            "sample_packing not compatible with sdp_attention. Use flash_attention"
+        )
+
+    if cfg.sample_packing and cfg.xformers_attention:
+        raise ValueError(
+            "sample_packing not compatible with xformers_attention. Use flash_attention"
+        )
+
+    # 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/data.py

@@ -0,0 +1,608 @@
+"""Module containing data utilities"""
+import functools
+import hashlib
+import logging
+from hashlib import md5
+from pathlib import Path
+from typing import Tuple, Union
+
+import torch
+from datasets import (
+    Dataset,
+    DatasetDict,
+    concatenate_datasets,
+    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,
+)
+from axolotl.utils.distributed import is_main_process, zero_first
+from axolotl.utils.trainer import (
+    calculate_total_num_steps,
+    process_datasets_for_packing,
+)
+
+LOG = logging.getLogger("axolotl")
+DEFAULT_DATASET_PREPARED_PATH = "last_run_prepared"
+
+
+def prepare_dataset(cfg, tokenizer):
+    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 or 42,
+        )
+        # 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
+
+    with zero_first(is_main_process()):
+        train_dataset, eval_dataset = process_datasets_for_packing(
+            cfg, train_dataset, eval_dataset
+        )
+    if cfg.max_steps:
+        total_num_steps = min(
+            calculate_total_num_steps(cfg, train_dataset, tokenizer), cfg.max_steps
+        )
+        LOG.info(f"Maximum number of steps set at {total_num_steps}")
+    else:
+        total_num_steps = calculate_total_num_steps(cfg, train_dataset, tokenizer)
+    return train_dataset, eval_dataset, total_num_steps
+
+
+def load_tokenized_prepared_datasets(
+    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])
+                )
+                + "|"
+                + 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["train"]
+    except Exception:  # pylint: disable=broad-except # nosec
+        pass
+
+    if dataset:
+        ...
+    elif any(prepared_ds_path.glob("*")):
+        LOG.info(f"Loading prepared dataset from disk at {prepared_ds_path}...")
+        dataset = load_from_disk(str(prepared_ds_path))
+        LOG.info("Prepared dataset loaded from disk...")
+    else:
+        LOG.info(f"Unable to find prepared dataset in {prepared_ds_path}")
+        LOG.info("Loading raw datasets...")
+
+        if cfg.seed:
+            seed = cfg.seed
+        else:
+            LOG.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,
+                    name=d.name,
+                    streaming=True,
+                    use_auth_token=use_auth_token,
+                )
+                ds_from_hub = True
+            except FileNotFoundError:
+                pass
+
+            # prefer local dataset, even if hub exists
+            local_path = Path(d.path)
+            if local_path.exists():
+                if local_path.is_dir():
+                    # TODO dirs with arrow or parquet files could be loaded with `load_from_disk`
+                    ds = load_dataset(
+                        d.path,
+                        name=d.name,
+                        data_files=d.data_files,
+                        streaming=False,
+                        split=None,
+                    )
+                elif local_path.is_file():
+                    ds = load_dataset(
+                        "json",
+                        name=d.name,
+                        data_files=d.path,
+                        streaming=False,
+                        split=None,
+                    )
+                else:
+                    raise ValueError(
+                        "unhandled dataset load: local path exists, but is neither a directory or a file"
+                    )
+            elif ds_from_hub:
+                ds = load_dataset(
+                    d.path,
+                    name=d.name,
+                    streaming=False,
+                    data_files=d.data_files,
+                    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", name=d.name, 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 "train" in ds:
+                    ds = ds.shuffle(seed=seed)["train"].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 "train" in ds:
+                ds = ds["train"]
+            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_')}?"
+                LOG.error(f"unhandled prompt tokenization strategy: {d.type}. {suffix}")
+                raise ValueError(
+                    f"unhandled prompt tokenization strategy: {d.type} {suffix}"
+                )
+        LOG.info("merging datasets")
+        dataset = concatenate_datasets(datasets)
+
+        if len(datasets) > 1:
+            LOG.info("shuffle merged datasets")
+            dataset = dataset.shuffle(seed=seed)
+        if cfg.local_rank == 0:
+            LOG.info(f"Saving merged prepared dataset to disk... {prepared_ds_path}")
+            dataset.save_to_disk(prepared_ds_path)
+            if cfg.push_dataset_to_hub:
+                LOG.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:
+                LOG.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,
+                )
+                dataset = dataset["train"]
+        except Exception:  # pylint: disable=broad-except # nosec
+            pass
+
+        if dataset:
+            ...
+        elif any(prepared_ds_path.glob("*")):
+            LOG.info(
+                f"Loading prepared packed dataset from disk at {prepared_ds_path}..."
+            )
+            dataset = load_from_disk(str(prepared_ds_path))
+            LOG.info("Prepared packed dataset loaded from disk...")
+            if cfg.push_dataset_to_hub:
+                LOG.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 = load_tokenized_prepared_datasets(
+                tokenizer, cfg, default_dataset_prepared_path
+            )
+
+            if cfg.seed:
+                dataset = dataset.shuffle(seed=cfg.seed)
+
+            constant_len_dataset = ConstantLengthDataset(
+                tokenizer,
+                [dataset],
+                seq_length=max_packed_sequence_len,
+            )
+            LOG.info(f"packing master dataset to len: {cfg.max_packed_sequence_len}")
+            dataset = Dataset.from_list(list(constant_len_dataset))
+
+            # filter out bad data
+            # TODO convert to dataset.filter(...)
+            dataset = Dataset.from_list(
+                [
+                    d
+                    for d in dataset
+                    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"])
+                ]
+            )
+
+            if cfg.local_rank == 0:
+                LOG.info(
+                    f"Saving packed prepared dataset to disk... {prepared_ds_path}"
+                )
+                dataset.save_to_disk(prepared_ds_path)
+                if cfg.push_dataset_to_hub:
+                    LOG.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 = load_tokenized_prepared_datasets(
+            tokenizer, cfg, default_dataset_prepared_path
+        )
+
+    if cfg.dataset_shard_num and cfg.dataset_shard_idx is not None:
+        LOG.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:
+        # ensure we end up with the same fingerprint by doing rank0 first and being able to cache
+        to_hash_train = (
+            dataset._fingerprint  # pylint: disable=protected-access
+            + "|"
+            + str(cfg.val_set_size)
+            + "|"
+            + "train"
+            + "|"
+            + str(cfg.seed or 42)
+        )
+        to_hash_test = (
+            dataset._fingerprint  # pylint: disable=protected-access
+            + "|"
+            + str(cfg.val_set_size)
+            + "|"
+            + "test"
+            + "|"
+            + str(cfg.seed or 42)
+        )
+        train_fingerprint = hashlib.md5(
+            to_hash_train.encode(), usedforsecurity=False
+        ).hexdigest()
+        test_fingerprint = hashlib.md5(
+            to_hash_test.encode(), usedforsecurity=False
+        ).hexdigest()
+
+        with zero_first(is_main_process()):
+            dataset = dataset.train_test_split(
+                test_size=cfg.val_set_size,
+                shuffle=False,
+                seed=cfg.seed or 42,
+                train_new_fingerprint=train_fingerprint,
+                test_new_fingerprint=test_fingerprint,
+            )
+
+        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],
+    }
+
+    LOG.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/dataloader.py

@@ -0,0 +1,288 @@
+# pylint: skip-file
+import hashlib
+import itertools
+import logging
+import math
+from typing import Any, Callable, List, Union
+
+import numba
+import numpy as np
+from torch.utils.data import DistributedSampler, Sampler
+
+LOG = logging.getLogger("axolotl.utils.dataloader")
+
+
+@numba.njit
+def ffd_check(a: np.ndarray, c: int, n: int):
+    # First-fit-decreasing bin packing
+    # Check if a[] could fit in n bins with capacity c
+    # https://en.wikipedia.org/wiki/First-fit-decreasing_bin_packing
+
+    a = np.sort(a)[::-1]
+    bins = np.full((n,), c, dtype=a.dtype)
+    for size in a:
+        not_found = True
+        for idx in range(n):
+            if bins[idx] >= size:
+                bins[idx] -= size
+                not_found = False
+                break
+
+        if not_found:
+            return False
+
+    return True
+
+
+@numba.njit
+def ffd_with_result(a: np.ndarray, c: int, start_index: int):
+    # First-fit-decreasing bin packing (with result return)
+
+    indices = np.argsort(a)[::-1]
+    a = a[indices]
+
+    bins: List[Any] = []
+    bins_result: List[Any] = []
+    for a_id, size in enumerate(a):
+        add_new = True
+        for idx in range(len(bins)):
+            if bins[idx] >= size:
+                bins[idx] -= size
+                bins_result[idx].append(indices[a_id] + start_index)
+                add_new = False
+                break
+
+        if add_new:
+            bins.append(c - size)
+            bins_result.append([indices[a_id] + start_index])
+
+    return bins_result, len(a)
+
+
+@numba.njit
+def allocate(
+    lengths: np.ndarray, lengths_cumsum: np.ndarray, rank: int, c: int, n: int
+):
+    """
+    :param lengths: array of lengths of each sample
+    :param lengths_cumsum: cumulative sum of consecutive lengths
+    :param rank: rank for this process
+    :param c: length of tokens per batch
+    :param n: number of ranks
+    :return:
+    """
+    # Dynamic batch allocator, similar to Multifit
+    # https://en.wikipedia.org/wiki/Multifit_algorithm
+    # ~99.5% efficiency on OpenChat training set (12 * 2048 ctx len)
+
+    s = 0
+    start_index = 0
+    result = []
+    result_totseqs = []
+
+    while True:
+        # binary search [left, right)
+        left = 1
+        right = 1 + np.searchsorted(lengths_cumsum[start_index:], s + c * n, "right")
+
+        while right - left > 1:
+            mid = (left + right) // 2
+            if ffd_check(lengths[start_index : start_index + mid], c, n):
+                left = mid
+            else:
+                right = mid
+
+        # use length left
+        batch, tot_seqs = ffd_with_result(
+            lengths[start_index : start_index + left], c, start_index
+        )
+        if len(batch) < n:
+            break
+
+        start_index += left
+        s = lengths_cumsum[start_index - 1]
+
+        # add local rank
+        result.append(batch[rank])
+        # add total seqs for all ranks
+        result_totseqs.append(tot_seqs)
+        # yield batch[rank], tot_seqs, s, len(result) * c * n
+    return result, result_totseqs, s, len(result) * c * n
+
+
+def chunk(iterable, n):
+    """
+    Chunk data into tuples of length n
+    """
+    # batched('ABCDEFG', 3) --> ABC DEF G
+    if n < 1:
+        raise ValueError("n must be at least one")
+    it = iter(iterable)
+    while batch := tuple(itertools.islice(it, n)):
+        yield batch
+
+
+def hash_indices(lst: List[int]) -> str:
+    # Convert the list of integers to a string representation
+    concatenated = ",".join(map(str, lst))
+
+    # Generate the hash
+    sha256 = hashlib.sha256()
+    sha256.update(concatenated.encode())
+
+    return sha256.hexdigest()
+
+
+class MultipackDistributedDataloader:
+    """Unpadded data loading using Multipack.
+    Adapted from https://github.com/imoneoi/openchat/blob/v3_fix_mle_loss/ochat/training_deepspeed/multipack_dataloader.py
+    Approximate (at most ~1.22x) the optimal solution of the identical-machines scheduling problem, which is NP-hard.
+    """
+
+    def __init__(
+        self,
+        dataset: Any,
+        collate_fn: Callable,
+        seq_max_length: int = 2048,
+        batch_size: int = 1,
+        sampler: Union[Sampler, DistributedSampler] = None,
+        packing_efficiency_estimate: float = 1.0,
+        sample_packing_seq_len_multiplier: int = 1,
+        device_count: int = 1,
+    ):
+        # Dataset
+        self.dataset = dataset
+        self.lengths = (
+            dataset.data.column("position_ids")
+            .to_pandas()
+            .apply(lambda x: x[-1] + 1)
+            .values
+        )
+        assert isinstance(self.lengths, np.ndarray)
+        assert batch_size % sample_packing_seq_len_multiplier == 0
+        assert batch_size >= sample_packing_seq_len_multiplier
+        self.sampler = sampler
+        self.batch_size = batch_size
+        self.sample_packing_seq_len_multiplier = sample_packing_seq_len_multiplier
+        self.seq_max_length = seq_max_length
+        self.batch_max_length = batch_size * seq_max_length
+        self.collate_fn = collate_fn
+
+        self.num_replicas = 1
+        self.rank = 0
+
+        # statistics
+        self.eff_total_used = 0
+        self.eff_total_slots = 0
+        self.packing_efficiency_estimate = packing_efficiency_estimate or 1.0
+        self.device_count = device_count
+
+    def generate_batches(self, set_stats=False):
+        LOG.info("generating packed batches")
+        if self.sampler:
+            indices = [idx for idx in self.sampler]
+        else:
+            indices = range(0, len(self.dataset))
+
+        LOG.info(hash_indices(indices))
+        lengths = self.lengths[indices]
+        lengths_cumsum = np.cumsum(lengths)
+
+        batches, totseqs, total_used, total_slots = allocate(
+            lengths=lengths,
+            lengths_cumsum=lengths_cumsum,
+            rank=self.rank,
+            # c=self.batch_max_length,
+            c=self.seq_max_length * self.sample_packing_seq_len_multiplier,
+            n=self.num_replicas,
+        )
+
+        batches = [[indices[b_idx] for b_idx in batch] for batch in batches]
+
+        # statistics
+        if set_stats:
+            self.eff_total_used += total_used
+            self.eff_total_slots += total_slots
+
+        return batches, totseqs
+
+    def __iter__(self):
+        if hasattr(self.sampler, "set_epoch"):
+            new_epoch = self.sampler.epoch + 1
+            self.sampler.set_epoch(new_epoch)
+            LOG.info(f"calling sampler.set_epoch({new_epoch})")
+        all_batches, _ = self.generate_batches(set_stats=True)
+        features = self.dataset.features.keys()
+        len_remaining = self._len_est()
+        for batches in chunk(
+            all_batches, self.batch_size // self.sample_packing_seq_len_multiplier
+        ):
+            chunked_data = []
+            attn_mask_cum_idx = 0
+            for batch in batches:
+                concatenated = {}
+                batched_data = [self.dataset[batch_idx] for batch_idx in batch]
+                for feature in features:
+                    if feature == "attention_mask":
+                        arrays = [
+                            (attn_mask_cum_idx + idx + 1) * np.array(item[feature])
+                            for idx, item in enumerate(batched_data)
+                            if feature in item
+                        ]
+                        attn_mask_cum_idx += len(batched_data)
+                        concatenated[feature] = np.concatenate(arrays)
+                    else:
+                        arrays = [
+                            np.array(item[feature])
+                            for item in batched_data
+                            if feature in item
+                        ]
+                        concatenated[feature] = np.concatenate(arrays)
+                chunked_data.append(concatenated)
+            yield self.collate_fn(chunked_data)
+            len_remaining -= 1
+            if not len_remaining:
+                return
+
+    def _len_est(self):
+        lengths_sum = np.sum(self.lengths)
+        lengths_sum_per_device = lengths_sum // self.device_count
+        LOG.info(
+            f"packing_efficiency_estimate: {self.packing_efficiency_estimate} "
+            f"total_num_tokens per device: {lengths_sum_per_device}"
+        )
+
+        # shave off 1% + 1 for dealing with variance in packing from random sampler to sampler
+        return (
+            math.floor(
+                0.99
+                * lengths_sum_per_device
+                / self.packing_efficiency_estimate
+                // self.seq_max_length
+                // self.batch_size
+            )
+            - 1
+        )
+
+    def __len__(self):
+        # this doesn't return the actual length b/c with distributed samplers, not all dataloaders get
+        # the same share of total tokens
+        # if not self.eff_total_used:
+        #     batches, _ = self.generate_batches(set_stats=True)
+        # LOG.info(
+        #     f"packing_efficiency_estimate: {self.packing_efficiency_estimate} "
+        #     f"actual packing efficiency: {self.efficiency()}"
+        # )
+        return max(1, self._len_est())
+
+    def len_w_stats(self):
+        if not self.eff_total_used:
+            batches, _ = self.generate_batches(set_stats=True)
+        LOG.info(
+            f"packing_efficiency_estimate: {self.packing_efficiency_estimate} "
+            f"actual packing efficiency: {self.efficiency()}"
+        )
+        return max(1, self._len_est())
+
+    def efficiency(self):
+        return self.eff_total_used / self.eff_total_slots