Fix logic errors

.github/CODE_OF_CONDUCT.md

@@ -1,129 +0,0 @@
-# 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

@@ -1,76 +0,0 @@
-# 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

@@ -1,13 +0,0 @@
-# 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

@@ -1,112 +0,0 @@
-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: config
-    attributes:
-      label: Config yaml
-      description: |
-        Please attach the config yaml!
-
-  - 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

@@ -1,7 +0,0 @@
-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

@@ -1,46 +0,0 @@
-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

@@ -1,63 +0,0 @@
-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

@@ -1,22 +0,0 @@
-<!--- 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

@@ -1,9 +0,0 @@
-# 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

@@ -1,10 +0,0 @@
-# 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/workflows/base.yml

@@ -25,11 +25,6 @@ jobs:
             python_version: "3.10"
             pytorch: 2.0.1
             torch_cuda_arch_list: "7.0 7.5 8.0 8.6 9.0+PTX"
-          - cuda: "118"
-            cuda_version: 11.8.0
-            python_version: "3.10"
-            pytorch: 2.1.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 9.0+PTX"
     steps:
       - name: Checkout
         uses: actions/checkout@v3

.github/workflows/main.yml

@@ -4,6 +4,7 @@ on:
   push:
     branches:
       - "main"
+      - "dev"
 
 jobs:
   build-axolotl:
@@ -13,23 +14,22 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - cuda: 118
+          - cuda: cu118
             cuda_version: 11.8.0
             python_version: "3.9"
             pytorch: 2.0.1
             axolotl_extras:
-          - cuda: 118
+          - cuda: cu118
             cuda_version: 11.8.0
             python_version: "3.10"
             pytorch: 2.0.1
             axolotl_extras:
-            is_latest: true
-          - cuda: 118
+          - cuda: cu118
             cuda_version: 11.8.0
-            python_version: "3.10"
-            pytorch: 2.1.0
-            axolotl_extras:
-    runs-on: [self-hosted, gpu, docker]
+            python_version: "3.9"
+            pytorch: 2.0.1
+            axolotl_extras: gptq
+    runs-on: self-hosted
     steps:
       - name: Checkout
         uses: actions/checkout@v3
@@ -50,14 +50,10 @@ jobs:
         with:
           context: .
           build-args: |
-            BASE_TAG=${{ github.ref_name }}-base-py${{ matrix.python_version }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}
-            CUDA=${{ matrix.cuda }}
-            PYTORCH_VERSION=${{ matrix.pytorch }}
+            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 }}-cu${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
-            ${{ (matrix.is_latest) && format('{0}-latest', steps.metadata.outputs.tags) || '' }}
+          tags: ${{ steps.metadata.outputs.tags }}-py${{ matrix.python_version }}-${{ matrix.cuda }}-${{ matrix.pytorch }}${{ matrix.axolotl_extras != '' && '-' || '' }}${{ matrix.axolotl_extras }}
           labels: ${{ steps.metadata.outputs.labels }}
   build-axolotl-runpod:
     needs: build-axolotl
@@ -76,13 +72,12 @@ jobs:
             python_version: "3.10"
             pytorch: 2.0.1
             axolotl_extras:
-            is_latest: true
           - cuda: 118
             cuda_version: 11.8.0
-            python_version: "3.10"
-            pytorch: 2.1.0
-            axolotl_extras:
-    runs-on: [self-hosted, gpu, docker]
+            python_version: "3.9"
+            pytorch: 2.0.1
+            axolotl_extras: gptq
+    runs-on: self-hosted
     steps:
       - name: Checkout
         uses: actions/checkout@v3
@@ -107,7 +102,5 @@ jobs:
             CUDA=${{ 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) || '' }}
+          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 }}

.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/pypi.yml

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

.github/workflows/tests.yml

@@ -1,32 +1,10 @@
-name: Tests
+name: PyTest
 on:
-  # check on push/merge to main, PRs, and manual triggers
   push:
-    branches:
-      - "main"
-    paths:
-      - '**.py'
-      - 'requirements.txt'
   pull_request:
-      paths:
-       - '**.py'
-       - 'requirements.txt'
-  workflow_dispatch:
 
 jobs:
-  pre-commit:
-    name: pre-commit
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v3
-      - uses: actions/setup-python@v4
-        with:
-          python-version: "3.9"
-          cache: 'pip' # caching pip dependencies
-      - uses: pre-commit/action@v3.0.0
-
-  pytest:
-    name: PyTest
+  test:
     runs-on: ubuntu-latest
     strategy:
       fail-fast: false
@@ -46,35 +24,9 @@ jobs:
 
       - name: Install dependencies
         run: |
-          pip3 install -U -e .
-          pip3 install -r requirements-tests.txt
+          pip install -e .
+          pip install -r requirements-tests.txt
 
       - name: Run tests
         run: |
-          pytest --ignore=tests/e2e/ tests/
-
-  e2e-test:
-    name: E2E Tests
-    runs-on: [self-hosted, gpu]
-    timeout-minutes: 20
-    needs: [pre-commit, pytest]
-
-    steps:
-      - name: Check out repository code
-        uses: actions/checkout@v3
-
-      - name: Setup Python
-        uses: actions/setup-python@v4
-        with:
-          python-version: "3.10"
-#          cache: 'pip' # caching pip dependencies
-
-      - name: Install dependencies
-        run: |
-          pip3 uninstall -y transformers accelerate
-          pip3 install -U -e .[flash-attn]
-          pip3 install -r requirements-tests.txt
-
-      - name: Run e2e tests
-        run: |
-          pytest tests/e2e/
+          pytest tests/

.gitignore

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

.isort.cfg

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

.mypy.ini

@@ -8,9 +8,6 @@ ignore_missing_imports = True
 [mypy-axolotl.monkeypatch.*]
 ignore_errors = True
 
-[mypy-axolotl.models.phi.*]
-ignore_errors = True
-
 [mypy-flash_attn.*]
 ignore_missing_imports = True
 
@@ -23,9 +20,6 @@ ignore_missing_imports = True
 [mypy-peft]
 ignore_missing_imports = True
 
-[mypy-wandb]
-ignore_missing_imports = True
-
 [mypy-bitsandbytes]
 ignore_missing_imports = True
 

README.md

@@ -1,112 +1,60 @@
 # Axolotl
 
-Axolotl is a tool designed to streamline the fine-tuning of various AI models, offering support for multiple configurations and architectures.
-
-Features:
-- Train various Huggingface models such as llama, pythia, falcon, mpt
-- Supports fullfinetune, lora, qlora, relora, and gptq
-- Customize configurations using a simple yaml file or CLI overwrite
-- Load different dataset formats, use custom formats, or bring your own tokenized datasets
-- Integrated with xformer, flash attention, rope scaling, and multipacking
-- Works with single GPU or multiple GPUs via FSDP or Deepspeed
-- Easily run with Docker locally or on the cloud
-- Log results and optionally checkpoints to wandb
-- And more!
-
-
-<table>
-<tr>
-<td>
-
-## Table of Contents
-- [Introduction](#axolotl)
-- [Supported Features](#axolotl-supports)
-- [Quickstart](#quickstart-)
-- [Installation](#installation)
-  - [Docker](#docker)
-  - [Conda/Pip venv](#condapip-venv)
-  - [LambdaLabs](#lambdalabs)
-  - [Windows](#windows)
-- [Dataset](#dataset)
-  - [How to Add Custom Prompts](#how-to-add-custom-prompts)
-  - [How to Use Custom Pretokenized Dataset](#how-to-use-your-custom-pretokenized-dataset)
-- [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>
+      <b>One repo to finetune them all! </b>
     </p>
     <p>
-      Go ahead and Axolotl questions!!
+      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 | ✅         | ✅    | ✅     | ❌             | ❌                 | ❌          | ❓            |
-| btlm     | ✅         | ✅    | ✅     | ❌             | ❌                 | ❌          | ❓            |
-| mpt      | ✅         | ❌    | ❓     | ❌             | ❌                 | ❌          | ❓            |
-| falcon   | ✅         | ✅    | ✅     | ❌             | ❌                 | ❌          | ❓            |
-| gpt-j    | ✅         | ✅    | ✅     | ❌             | ❌                 | ❓          | ❓            |
-| XGen     | ✅         | ❓    | ✅     | ❓             | ❓                 | ❓          | ✅            |
-| phi      | ✅         | ✅    | ✅     | ❓             | ❓                 | ❓          | ❓            |
+|          | fp16/fp32 | lora | qlora | gptq | gptq w/ lora | 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
-cd axolotl
 
-pip3 install packaging
-pip3 install -e '.[flash-attn,deepspeed]'
+pip3 install -e .
 pip3 install -U git+https://github.com/huggingface/peft.git
 
 # finetune lora
-accelerate launch -m axolotl.cli.train examples/openllama-3b/lora.yml
+accelerate launch scripts/finetune.py examples/openllama-3b/lora.yml
 
 # inference
-accelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \
-    --lora_model_dir="./lora-out"
+accelerate launch scripts/finetune.py examples/openllama-3b/lora.yml \
+    --inference --lora_model_dir="./lora-out"
 ```
 
 ## Installation
 
 ### Environment
 
-#### Docker
+- Docker
   ```bash
   docker run --gpus '"all"' --rm -it winglian/axolotl:main-py3.10-cu118-2.0.1
   ```
-  - `winglian/axolotl-runpod:main-latest`: for runpod or use this [direct link](https://runpod.io/gsc?template=v2ickqhz9s&ref=6i7fkpdz)
+  - `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:
 
@@ -114,42 +62,27 @@ accelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \
   docker compose up -d
   ```
 
-  <details>
-
-  <summary>Docker advanced</summary>
-
-  A more powerful Docker command to run would be this:
-
-  ```bash
-  docker run --gpus '"all"' --rm -it --name axolotl --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 --mount type=volume,src=axolotl,target=/workspace/axolotl -v ${HOME}/.cache/huggingface:/root/.cache/huggingface winglian/axolotl:main-py3.10-cu118-2.0.1
-  ```
-
-  It additionally:
-  * Prevents memory issues when running e.g. deepspeed (e.g. you could hit SIGBUS/signal 7 error) through `--ipc` and `--ulimit` args.
-  * Persists the downloaded HF data (models etc.) and your modifications to axolotl code through `--mount`/`-v` args.
-  * The `--name` argument simply makes it easier to refer to the container in vscode (`Dev Containers: Attach to Running Container...`) or in your terminal.
-
-  [More information on nvidia website](https://docs.nvidia.com/deeplearning/frameworks/user-guide/index.html#setincshmem)
-
-  </details>
-
-#### Conda/Pip venv
-  1. Install python >=**3.9**
+- Conda/Pip venv
+  1. Install python **3.9**
 
   2. Install pytorch stable https://pytorch.org/get-started/locally/
 
-  3. Install Axolotl along with python dependencies
+  3. Install python dependencies with ONE of the following:
+      - Recommended, supports QLoRA, NO gptq/int4 support
         ```bash
-        pip3 install packaging
-        pip3 install -e '.[flash-attn,deepspeed]'
+        pip3 install -e .
+        pip3 install -U git+https://github.com/huggingface/peft.git
         ```
-  4. (Optional) Login to Huggingface to use gated models/datasets.
+      - gptq/int4 support, NO QLoRA
         ```bash
-        huggingface-cli login
+        pip3 install -e .[gptq]
+        ```
+      - same as above but not recommended
+        ```bash
+        pip3 install -e .[gptq_triton]
         ```
-        Get the token at huggingface.co/settings/tokens
 
-#### LambdaLabs
+- LambdaLabs
   <details>
 
   <summary>Click to Expand</summary>
@@ -181,10 +114,12 @@ accelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \
   git clone https://github.com/OpenAccess-AI-Collective/axolotl
   cd axolotl
 
-  pip3 install packaging
-  pip3 install -e '.[flash-attn,deepspeed]'
+  pip3 install -e . # change depend on needs
   pip3 install protobuf==3.20.3
-  pip3 install -U --ignore-installed requests Pillow psutil scipy
+  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
@@ -193,19 +128,15 @@ accelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \
   ```
   </details>
 
-#### Windows
-Please use WSL or Docker!
-
 ### 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`: conversations where `from` is `human`/`gpt`
+- `sharegpt:chat`: conversations
   ```json
   {"conversations": [{"from": "...", "value": "..."}]}
   ```
@@ -270,10 +201,6 @@ Have dataset(s) in one of the following format (JSONL recommended):
   ```json
   {"article": "...", "question": "...", "answer": "..."}
   ```
-- `context_qa.load_v2`: in context question answering (alternate)
-  ```json
-  {"context": "...", "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": "..."}
@@ -294,18 +221,10 @@ Have dataset(s) in one of the following format (JSONL recommended):
   ```json
   {"conversations": [{"role": "...", "value": "..."}]}
   ```
-- `metharme`: instruction, adds additional eos tokens
-  ```json
-  {"prompt": "...", "generation": "..."}
-  ```
-- `sharegpt.load_role`: conversations where `role` is used instead of `from`
+- `sharegpt_simple.load_role`: conversations where `role` is used instead of `from`
   ```json
   {"conversations": [{"role": "...", "value": "..."}]}
   ```
-- `sharegpt.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": "..."}]}
@@ -315,28 +234,11 @@ Have dataset(s) in one of the following format (JSONL recommended):
 
 #### How to add custom prompts
 
-For a dataset that is preprocessed for instruction purposes:
+  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>`.
 
-```json
-{"instruction": "...", "output": "..."}
-```
+Optionally, download some datasets, see [data/README.md](data/README.md)
 
-You can use this example in your YAML config:
-
-```yaml
-datasets:
-  - path: repo
-    type:
-      system_prompt: ""
-      field_system: system
-      format: "[INST] {instruction} [/INST]"
-      no_input_format: "[INST] {instruction} [/INST]"
-```
-
-#### How to use your custom pretokenized dataset
-
-- Do not pass a `type:`
-- Columns in Dataset must be exactly `input_ids`, `attention_mask`, `labels`
 
 
 ### Config
@@ -363,35 +265,12 @@ See [examples](examples) for quick start. It is recommended to duplicate and mod
     - path: EleutherAI/pile
       name: enron_emails
       type: completion # format from earlier
-      field: text # Optional[str] default: text, field to use for completion data
-
-  # huggingface repo with multiple named configurations/subsets
-  datasets:
-    - path: bigcode/commitpackft
-      name:
-        - ruby
-        - python
-        - typescript
-      type: ... # unimplemented custom format
-
-  # fastchat conversation
-  # See 'conversation' options: https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py
-  datasets:
-    - path: ...
-      type: sharegpt
-      conversation: chatml
 
   # local
   datasets:
-    - path: data.jsonl # or json
-      ds_type: json # see other options below
-      type: alpaca
-
-  # dataset with splits, but no train split
-  dataset:
-    - path: knowrohit07/know_sql
-      type: context_qa.load_v2
-      train_on_split: validation
+    - path: json
+      data_files: data.jsonl # or json
+      type: alpaca # format from earlier
   ```
 
 - loading
@@ -419,18 +298,18 @@ See [examples](examples) for quick start. It is recommended to duplicate and mod
 
 <details>
 
-<summary>All yaml options (click me)</summary>
+<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
+# 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)
+# 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
+# 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
+# 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
@@ -443,26 +322,18 @@ tokenizer_type: AutoTokenizer
 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 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:
 
-# Used to identify which the model is based on
-is_falcon_derived_model:
-is_llama_derived_model:
-# Please note that if you set this to true, `padding_side` will be set to "left" by default
-is_mistral_derived_model:
-
-# Whether you are training a 4-bit GPTQ quantized model
+# 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
+# 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
+# use bitsandbytes 4 bit
 load_in_4bit:
 
 # Use CUDA bf16
@@ -472,97 +343,52 @@ fp16: true
 # Use CUDA tf32
 tf32: true # require >=ampere
 
-# No AMP (automatic mixed precision)
-bfloat16: true # require >=ampere
-float16: true
-
-# A list of one or more datasets to finetune the model with
+# a list of one or more datasets to finetune the model with
 datasets:
-  # HuggingFace dataset repo | "json" for local dataset, make sure to fill data_files
+  # hf dataset repo | "json" for local dataset, make sure to fill data_files
   - path: vicgalle/alpaca-gpt4
   # 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>
-    ds_type: # Optional[str] (json|arrow|parquet|text|csv) defines the datatype when path is a file
-    data_files: # Optional[str] path to source data files
-    shards: # Optional[int] number of shards to split data into
-    name: # Optional[str] name of dataset configuration to load
+    data_files: # path to source data files
+    shards: # number of shards to split data into
+    name: # name of dataset configuration to load
 
-    # Optional[str] fastchat conversation type, only used with type: sharegpt
-    conversation:  # Options (see Conversation 'name'): https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py
-
-  # Custom user prompt
-  - path: repo
-    type:
-      # The below are defaults. only set what's needed.
-      system_prompt: ""
-      system_format: "{system}"
-      field_system: system
-      field_instruction: instruction
-      field_input: input
-      field_output: output
-
-      # Customizable to be single line or multi-line
-      # 'format' can include {input}
-      format: |-
-        User: {instruction} {input}
-        Assistant:
-      # 'no_input_format' cannot include {input}
-      no_input_format: "{instruction} "
-
-      # For `completion` datsets only, uses the provided field instead of `text` column
-      field:
-
-# Axolotl attempts to save the dataset as an arrow after packing the data together so
+# 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 prepared dataset to hub
 push_dataset_to_hub: # repo path
-# The maximum number of processes to use while preprocessing your input dataset. This defaults to `os.cpu_count()`
-# if not set.
-dataset_processes: # defaults to os.cpu_count() if not set
 # 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`
+# 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.
+# How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc
 val_set_size: 0.04
 # Num shards for whole dataset
 dataset_shard_num:
 # Index of shard to use for whole dataset
 dataset_shard_idx:
 
-# The maximum length of an input to train with, this should typically be less than 2048
+# 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
-# Pad inputs so each step uses constant sized buffers
-# This will reduce memory fragmentation and may prevent OOMs, by re-using memory more efficiently
-pad_to_sequence_len:
-# Max sequence length to concatenate training samples together up to
-# Inspired by StackLLaMA. see https://huggingface.co/blog/stackllama#supervised-fine-tuning
+# 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'
+# use efficient multi-packing with block diagonal attention and per sequence position_ids. Recommend set to 'true'
 sample_packing:
-# Set to 'false' if getting errors during eval with sample_packing on.
-eval_sample_packing:
-# You can set these packing optimizations AFTER starting a training at least once.
+# 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
+# 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.
-# This means after training, if you want to test the model, you should set this to the value of `lora_out_dir`.
+# if you already have a lora model trained that you want to load, put that here
+# lora hyperparameters
 lora_model_dir:
-
-# LoRA hyperparameters
-# For more details about the following options, see:
-# https://www.anyscale.com/blog/fine-tuning-llms-lora-or-full-parameter-an-in-depth-analysis-with-llama-2
 lora_r: 8
 lora_alpha: 16
 lora_dropout: 0.05
@@ -574,122 +400,67 @@ lora_target_modules:
 #  - gate_proj
 #  - down_proj
 #  - up_proj
-lora_target_linear: # If true, will target all linear layers
-
-# If you added new tokens to the tokenizer, you may need to save some LoRA modules because they need to know the new tokens.
-# For LLaMA and Mistral, you need to save `embed_tokens` and `lm_head`. It may vary for other models.
-# `embed_tokens` converts tokens to embeddings, and `lm_head` converts embeddings to token probabilities.
-# https://github.com/huggingface/peft/issues/334#issuecomment-1561727994
+lora_target_linear: # if true, will target all linear layers
 lora_modules_to_save:
 #  - embed_tokens
 #  - lm_head
-
-# Once you complete training, the model will be saved to the following directory.
-# If you merge the adapter to the base model, a subdirectory `merged` will be created under this directory.
-# Make sure `lora_model_dir` points to this directory if you want to use the trained model.
 lora_out_dir:
 lora_fan_in_fan_out: false
 
-# ReLoRA configuration
-# Must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed
-relora_steps: # Number of steps per ReLoRA restart
-relora_warmup_steps: # Number of per-restart warmup steps
-relora_cpu_offload: # True to perform lora weight merges on cpu during restarts, for modest gpu memory savings
-
 # 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_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_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 full-finetuned model to
+# where to save the finished model to
 output_dir: ./completed-model
 
-# Whether to use torch.compile and which backend to use
-torch_compile:  # bool
-torch_compile_backend:  # Optional[str]
-
-# Training hyperparameters
-
-# If greater than 1, backpropagation will be skipped and the gradients will be accumulated for the given number of steps.
+# training hyperparameters
 gradient_accumulation_steps: 1
-# The number of samples to include in each batch. This is the number of samples sent to each GPU.
 micro_batch_size: 2
-eval_batch_size:
-num_epochs: 4
+eval_batch_size: 2
+num_epochs: 3
 warmup_steps: 100
 learning_rate: 0.00003
-lr_quadratic_warmup:
 logging_steps:
-save_strategy: # Set to `no` to skip checkpoint saves
-save_steps: # Leave empty to save at each epoch
-eval_steps: # Leave empty to eval at each epoch, integers for every N steps. decimal for fraction of total steps
-save_total_limit: # Checkpoints saved at a time
-# Maximum number of iterations to train for. It precedes num_epochs which means that
-# if both are set, num_epochs will not be guaranteed.
-# e.g., when 1 epoch is 1000 steps => `num_epochs: 2` and `max_steps: 100` will train for 100 steps
-max_steps:
+save_steps:
+eval_steps:
+save_total_limit:
 
-eval_table_size: # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0
-eval_table_max_new_tokens: # Total number of tokens generated for predictions sent to wandb. Default is 128
-
-# Save model as safetensors (require safetensors package)
+# save model as safetensors (require safetensors package)
 save_safetensors:
 
-# Whether to mask out or include the human's prompt from the training labels
+# 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 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
+# 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
+# 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 one_cycle optim
+lr_div_factor: # learning rate div factor
 
-# For log_sweep optim
+# for log_sweep optim
 log_sweep_min_lr:
 log_sweep_max_lr:
 
-# Specify optimizer
-# Valid values are driven by the Transformers OptimizerNames class, see:
-# https://github.com/huggingface/transformers/blob/95b374952dc27d8511541d6f5a4e22c9ec11fb24/src/transformers/training_args.py#L134
-#
-# Note that not all optimizers may be available in your environment, ex: 'adamw_anyprecision' is part of
-# torchdistx, 'adamw_bnb_8bit' is part of bnb.optim.Adam8bit, etc. When in doubt, it is recommended to start with the optimizer used
-# in the examples/ for your model and fine-tuning use case.
-#
-# Valid values for 'optimizer' include:
-# - adamw_hf
-# - adamw_torch
-# - adamw_torch_fused
-# - adamw_torch_xla
-# - adamw_apex_fused
-# - adafactor
-# - adamw_anyprecision
-# - sgd
-# - adagrad
-# - adamw_bnb_8bit
-# - lion_8bit
-# - lion_32bit
-# - paged_adamw_32bit
-# - paged_adamw_8bit
-# - paged_lion_32bit
-# - paged_lion_8bit
+# specify optimizer
 optimizer:
-# Specify weight decay
+# specify weight decay
 weight_decay:
 # adamw hyperparams
 adam_beta1:
@@ -698,68 +469,55 @@ adam_epsilon:
 # Gradient clipping max norm
 max_grad_norm:
 
-# Augmentation techniques
-# NEFT https://arxiv.org/abs/2310.05914, set this to a number (paper default is 5) to add noise to embeddings
-# currently only supported on Llama and Mistral
-noisy_embedding_alpha:
-
-# Whether to bettertransformers
+# whether to bettertransformers
 flash_optimum:
-# Whether to use xformers attention patch https://github.com/facebookresearch/xformers:
+# 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:
-flash_attn_cross_entropy:  # Whether to use flash-attention cross entropy implementation - advanced use only
-flash_attn_rms_norm:  # Whether to use flash-attention rms norm implementation - advanced use only
-flash_attn_fuse_qkv: # Whether to fuse QKV into a single operation
-flash_attn_fuse_mlp: # Whether to fuse part of the MLP into a single operation
-# Whether to use scaled-dot-product attention
+# whether to use flash attention patch https://github.com/HazyResearch/flash-attention:
+flash_attention:  # require a100 for llama
+# whether to use scaled-dot-product attention
 # https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html
 sdp_attention:
 # Landmark attention (only llama)
 landmark_attention:
 # xpos RoPE see https://github.com/kaiokendev/cutoff-len-is-context-len/blob/main/util/xpos_rope_llama_monkey_patch.py
-# LLaMA only
+# 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 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.
+# 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
+# don't mess with this, it's here for accelerate and torchrun
 local_rank:
 
-# Add or change special tokens.
-# If you add tokens here, you don't need to add them to the `tokens` list.
+# add or change special tokens
 special_tokens:
   # bos_token: "<s>"
   # eos_token: "</s>"
   # unk_token: "<unk>"
-
-# Add extra tokens.
+# add extra tokens
 tokens:
 
 # FSDP
 fsdp:
 fsdp_config:
 
-# Deepspeed config path. e.g., deepspeed/zero3.json
+# Deepspeed
 deepspeed:
 
-# Advanced DDP Arguments
-ddp_timeout:
-ddp_bucket_cap_mb:
-ddp_broadcast_buffers:
-
 # 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:
 
@@ -775,108 +533,21 @@ strict:
 
 </details>
 
-<details>
-<summary> Understanding of batch size and gradient accumulation steps </summary>
-<br/>
-Gradient accumulation means accumulating gradients over several mini-batches and updating the model weights afterward. When the samples in each batch are diverse, this technique doesn't significantly impact learning.
-
-This method allows for effective training with larger effective batch sizes without needing proportionally larger memory. Here's why:
-
-1. **Memory Consumption with Batch Size**: The primary reason increasing the batch size impacts memory is due to the storage requirements for intermediate activations. When you forward propagate a batch through a network, you have to store the activations at each layer for each sample in the batch, because these activations are used during backpropagation to compute gradients. Therefore, larger batches mean more activations, leading to greater GPU memory consumption.
-
-2. **Gradient Accumulation**: With gradient accumulation, you're effectively simulating a larger batch size by accumulating gradients over several smaller batches (or micro-batches). However, at any given time, you're only forward and backward propagating a micro-batch. This means you only store activations for the micro-batch, not the full accumulated batch. As a result, you can simulate the effect of a larger batch size without the memory cost of storing activations for a large batch.
-
-**Example 1:**
-Micro batch size: 3
-Gradient accumulation steps: 2
-Number of GPUs: 3
-Total batch size = 3 * 2 * 3 = 18
-
-```
-| GPU 1          | GPU 2          | GPU 3          |
-|----------------|----------------|----------------|
-| S1, S2, S3     | S4, S5, S6     | S7, S8, S9     |
-| e1, e2, e3     | e4, e5, e6     | e7, e8, e9     |
-|----------------|----------------|----------------|
-| → (accumulate) | → (accumulate) | → (accumulate) |
-|----------------|----------------|----------------|
-| S10, S11, S12  | S13, S14, S15  | S16, S17, S18  |
-| e10, e11, e12  | e13, e14, e15  | e16, e17, e18  |
-|----------------|----------------|----------------|
-| → (apply)      | → (apply)      | → (apply)      |
-
-Accumulated gradient for the weight w1 after the second iteration (considering all GPUs):
-Total gradient for w1 = e1 + e2 + e3 + e4 + e5 + e6 + e7 + e8 + e9 + e10 + e11 + e12 + e13 + e14 + e15 + e16 + e17 + e18
-
-Weight update for w1:
-w1_new = w1_old - learning rate x (Total gradient for w1 / 18)
-```
-
-**Example 2:**
-Micro batch size: 2
-Gradient accumulation steps: 1
-Number of GPUs: 3
-Total batch size = 2 * 1 * 3 = 6
-
-```
-| GPU 1     | GPU 2     | GPU 3     |
-|-----------|-----------|-----------|
-| S1, S2    | S3, S4    | S5, S6    |
-| e1, e2    | e3, e4    | e5, e6    |
-|-----------|-----------|-----------|
-| → (apply) | → (apply) | → (apply) |
-
-Accumulated gradient for the weight w1 (considering all GPUs):
-Total gradient for w1 = e1 + e2 + e3 + e4 + e5 + e6
-
-Weight update for w1:
-w1_new = w1_old - learning rate × (Total gradient for w1 / 6)
-```
-
-</details>
-
 ### Train
 
 Run
 ```bash
-accelerate launch -m axolotl.cli.train your_config.yml
-```
-
-#### Preprocess dataset
-
-You can optionally pre-tokenize dataset with the following before finetuning.
-This is recommended for large datasets.
-
-- Set `push_dataset_to_hub: hf_user/repo` to push it to Huggingface.
-- Use `--debug` to see preprocessed examples.
-
-```bash
-python -m axolotl.cli.preprocess your_config.yml
+accelerate launch scripts/finetune.py configs/your_config.yml
 ```
 
 #### Multi-GPU
 
-Below are the options available in axolotl for training with multiple GPUs. Note that DeepSpeed
-is the recommended multi-GPU option currently because FSDP may experience
-[loss instability](https://github.com/huggingface/transformers/issues/26498).
-
-##### DeepSpeed
-
-Deepspeed is an optimization suite for multi-gpu systems allowing you to train much larger models than you
-might typically be able to fit into your GPU's VRAM. More information about the various optimization types
-for deepspeed is available at https://huggingface.co/docs/accelerate/main/en/usage_guides/deepspeed#what-is-integrated
-
-We provide several default deepspeed JSON configurations for ZeRO stage 1, 2, and 3.
-
-```yaml
-deepspeed: deepspeed/zero1.json
+You can optionally pre-tokenize dataset with the following before finetuning:
+```bash
+CUDA_VISIBLE_DEVICES="" accelerate ... --prepare_ds_only
 ```
 
-```shell
-accelerate launch -m axolotl.cli.train examples/llama-2/config.py --deepspeed deepspeed/zero1.json
-```
-
-##### FSDP
+##### Config
 
 - llama FSDP
 ```yaml
@@ -889,6 +560,8 @@ fsdp_config:
   fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
 ```
 
+- llama Deepspeed: append `ACCELERATE_USE_DEEPSPEED=true` in front of finetune command
+
 ##### Weights & Biases Logging
 
 - wandb options
@@ -907,41 +580,35 @@ Pass the appropriate flag to the train command:
 
 - Pretrained LORA:
   ```bash
-  python -m axolotl.cli.inference examples/your_config.yml --lora_model_dir="./lora-output-dir"
+  --inference --lora_model_dir="./lora-output-dir"
   ```
 - Full weights finetune:
   ```bash
-  python -m axolotl.cli.inference examples/your_config.yml --base_model="./completed-model"
+  --inference --base_model="./completed-model"
   ```
 - Full weights finetune w/ a prompt from a text file:
   ```bash
-  cat /tmp/prompt.txt | python -m axolotl.cli.inference examples/your_config.yml \
-    --base_model="./completed-model" --prompter=None --load_in_8bit=True
+  cat /tmp/prompt.txt | python scripts/finetune.py configs/your_config.yml \
+    --base_model="./completed-model" --inference --prompter=None --load_in_8bit=True
   ```
 
-Please use `--sample_packing False` if you have it on and receive the error similar to below:
-
-> RuntimeError: stack expects each tensor to be equal size, but got [1, 32, 1, 128] at entry 0 and [1, 32, 8, 128] at entry 1
-
 ### Merge LORA to base
 
 Add below flag to train command above
 
 ```bash
-python3 -m axolotl.cli.merge_lora examples/your_config.yml --lora_model_dir="./completed-model" --load_in_8bit=False --load_in_4bit=False
+--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 -m axolotl.cli.merge_lora ...
+CUDA_VISIBLE_DEVICES="" python3 scripts/finetune.py ...
 ```
 
 ## Common Errors 🧰
 
-See also the [FAQ's](./docs/faq.md).
-
-> 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:
+> Cuda out of memory
 
 Please reduce any below
   - `micro_batch_size`
@@ -949,12 +616,6 @@ Please reduce any below
   - `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`
@@ -967,10 +628,6 @@ Try to turn off xformers.
 
 It's safe to ignore it.
 
-> NCCL Timeouts during training
-
-See the [NCCL](docs/nccl.md) guide.
-
 ## Need help? 🙋♂️
 
 Join our [Discord server](https://discord.gg/HhrNrHJPRb) where we can help you
@@ -987,8 +644,6 @@ Building something cool with Axolotl? Consider adding a badge to your model card
 
 ## 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)
@@ -999,9 +654,7 @@ PocketDoc Labs
 
 ## 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.
+Bugs? Please check for open issue else create a new [Issue](https://github.com/OpenAccess-AI-Collective/axolotl/issues/new).
 
 PRs are **greatly welcome**!
 

data/README.md

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

data/raw/.gitignore

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

deepspeed/zero1.json

@@ -1,41 +0,0 @@
-{
-  "zero_optimization": {
-    "stage": 1,
-    "overlap_comm": true
-  },
-  "bf16": {
-    "enabled": "auto"
-  },
-  "fp16": {
-    "enabled": "auto",
-    "auto_cast": false,
-    "loss_scale": 0,
-    "initial_scale_power": 32,
-    "loss_scale_window": 1000,
-    "hysteresis": 2,
-    "min_loss_scale": 1
-  },
-  "optimizer": {
-    "type": "AdamW",
-    "params": {
-      "lr": "auto",
-      "betas": "auto",
-      "eps": "auto",
-      "weight_decay": "auto"
-    }
-  },
-  "scheduler": {
-    "type": "WarmupDecayLR",
-    "params": {
-      "warmup_min_lr": "auto",
-      "warmup_max_lr": "auto",
-      "warmup_num_steps": "auto",
-      "warmup_type": "linear",
-      "total_num_steps": "auto"
-    }
-  },
-  "gradient_accumulation_steps": "auto",
-  "train_batch_size": "auto",
-  "train_micro_batch_size_per_gpu": "auto",
-  "wall_clock_breakdown": false
-}

deepspeed/zero2.json

@@ -1,45 +0,0 @@
-{
-  "zero_optimization": {
-    "stage": 2,
-    "offload_optimizer": {
-      "device": "cpu"
-    },
-    "contiguous_gradients": true,
-    "overlap_comm": true
-  },
-  "bf16": {
-    "enabled": "auto"
-  },
-  "fp16": {
-    "enabled": "auto",
-    "auto_cast": false,
-    "loss_scale": 0,
-    "initial_scale_power": 32,
-    "loss_scale_window": 1000,
-    "hysteresis": 2,
-    "min_loss_scale": 1
-  },
-  "optimizer": {
-    "type": "AdamW",
-    "params": {
-      "lr": "auto",
-      "betas": "auto",
-      "eps": "auto",
-      "weight_decay": "auto"
-    }
-  },
-  "scheduler": {
-    "type": "WarmupDecayLR",
-    "params": {
-      "warmup_min_lr": "auto",
-      "warmup_max_lr": "auto",
-      "warmup_num_steps": "auto",
-      "warmup_type": "linear",
-      "total_num_steps": "auto"
-    }
-  },
-  "gradient_accumulation_steps": "auto",
-  "train_batch_size": "auto",
-  "train_micro_batch_size_per_gpu": "auto",
-  "wall_clock_breakdown": false
-}

deepspeed/zero3.json

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

docker-compose.yaml

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

docker/Dockerfile

@@ -5,29 +5,25 @@ ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
 ARG AXOLOTL_EXTRAS=""
 ARG CUDA="118"
 ENV BNB_CUDA_VERSION=$CUDA
-ARG PYTORCH_VERSION="2.0.1"
-
-ENV PYTORCH_VERSION=$PYTORCH_VERSION
 
 RUN apt-get update && \
     apt-get install -y vim curl
 
 WORKDIR /workspace
 
+RUN pip3 install --force-reinstall "peft @ git+https://github.com/huggingface/peft.git@main"
 RUN git clone --depth=1 https://github.com/OpenAccess-AI-Collective/axolotl.git
-
-WORKDIR /workspace/axolotl
-
 # If AXOLOTL_EXTRAS is set, append it in brackets
-RUN sed -i "s/torch==.*/torch==$PYTORCH_VERSION/" requirements.txt
-RUN if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
-        pip install -e .[flash-attn,$AXOLOTL_EXTRAS]; \
+RUN cd axolotl && \
+    if [ "$AXOLOTL_EXTRAS" != "" ] ; then \
+        pip install -e .[$AXOLOTL_EXTRAS]; \
     else \
-        pip install -e .[flash-attn]; \
+        pip install -e .; \
     fi
 
 # fix so that git fetch/pull from remote works
-RUN git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
+RUN cd axolotl && \
+    git config remote.origin.fetch "+refs/heads/*:refs/remotes/origin/*" && \
     git config --get remote.origin.fetch
 
 # helper for huggingface-login cli

docker/Dockerfile-base

@@ -13,14 +13,16 @@ ARG CUDA="118"
 
 ENV PYTHON_VERSION=$PYTHON_VERSION
 
-RUN apt-get update \
-    && apt-get install -y wget git build-essential ninja-build git-lfs libaio-dev && rm -rf /var/lib/apt/lists/* \
-    && wget \
+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 \
-    && conda create -n "py${PYTHON_VERSION}" python="${PYTHON_VERSION}"
+    && 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}"
 
@@ -29,6 +31,26 @@ 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 v2.0.1  && \
+    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"
@@ -37,15 +59,13 @@ WORKDIR /workspace
 
 RUN git clone https://github.com/microsoft/DeepSpeed.git && \
     cd DeepSpeed && \
-    MAX_CONCURRENCY=8 DS_BUILD_SPARSE_ATTN=0 DS_BUILD_OPS=1 DS_BUILD_EVOFORMER_ATTN=0 python3 setup.py bdist_wheel
+    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=$CUDA
-ARG MAX_JOBS="-1"
-ENV MAX_JOBS=$MAX_JOBS
 
 RUN git clone https://github.com/TimDettmers/bitsandbytes.git && \
     cd bitsandbytes && \
@@ -57,6 +77,12 @@ 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
 
@@ -64,8 +90,13 @@ 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
+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 && \

docs/faq.md

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

docs/multi-node.md

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

docs/multipack.md

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

docs/nccl.md

@@ -1,46 +0,0 @@
-# NCCL
-
-NVIDIA NCCL is a library to facilitate and optimize multi-GPU communication operations, such as broadcast, all-gather, reduce, all-reduce, etc. Broadly, NCCL configuration is highly environment-specific and is configured via several [environment variables](https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/env.html). A common NCCL-related problem occurs when a long-running operation times out causing the training process to abort:
-
-```text
-Watchdog caught collective operation timeout: WorkNCCL(SeqNum=42, OpType=ALLGATHER, Timeout(ms)=1800000) ran for 1806948 milliseconds before timing out.
-```
-
-Often, this timeout will happen after 30 minutes (the default setting) and is accompanied by below-average power consumption with near 100% GPU utilization before the error is raised. Nvidia recommends [disabling PCI access control services (ACS)](https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/troubleshooting.html#pci-access-control-services-acs) as a possible solution if this is available to you.
-
-Forcing cross-GPU communication via [NVLink](https://en.wikipedia.org/wiki/NVLink) may help without increasing timeouts. To verify that your configuration is leveraging NVLink run the following command:
-
-```shell
-nvidia-smi nvlink --status
-```
-
-To force NCCL to use NVLink, simply set this in the environment:
-
-```shell
-export NCCL_P2P_LEVEL=NVL
-```
-
-If NVLink is not available in your environment there are other options for ``NCCL_P2P_LEVEL`` in the table below:
-
-| NCCL_P2P_LEVEL | Description |
-| -------------- | ----------- |
-| PIX | P2P data transfers through no more than a single PCIe bridge. Faster data transfer rates vs to paths involving multiple bridges, but slower compared to direct GPU-to-GPU communication. |
-| PXB | P2P data transfers through multiple PCIe bridges but not going through the PCIe Host Bridge; this path involves a complex routing process, potentially incurring a moderate level of latency. |
-| PHB | P2P data transfers occur over the PCIe and through a PCIe Host Bridge, typically involving the CPU, which can facilitate direct memory access but might introduce additional latency compared to more direct paths (ex PIX, NVL) |
-
-To validate that acceptable data transfer speeds exist for your training job, running [NCCL Tests](https://github.com/NVIDIA/nccl-tests/blob/master/README.md) can help pinpoint bottlenecks, for example:
-
-```shell
-./build/all_reduce_perf -b 8 -e 128M -f 2 -g 3
-```
-
-It can be useful when debugging NCCL communication timeouts to activate additional logging in both PyTorch and NCCL:
-
-```shell
-export NCCL_DEBUG=INFO
-export NCCL_DEBUG_SUBSYS=ALL
-export TORCH_DISTRIBUTED_DEBUG=INFO
-export TORCHELASTIC_ERROR_FILE=/PATH/TO/torcherror.log
-```
-
-Finally, if you believe your training job needs more time you can increase the timeout past 30 minutes by setting the ``ddp_timeout`` value in the Axolotl configuration. See [PyTorch init_process_group](https://pytorch.org/docs/stable/distributed.html#torch.distributed.init_process_group) for documentation on this value.

examples/cerebras/btlm-ft.yml

@@ -1,89 +0,0 @@
-base_model: cerebras/btlm-3b-8k-base
-model_type: AutoModelForCausalLM
-tokenizer_type: GPT2Tokenizer
-trust_remote_code: true
-tokenizer_use_fast: true
-tokenizer_legacy: true
-
-load_in_8bit: false
-load_in_4bit: false
-strict: false
-push_dataset_to_hub:
-hf_use_auth_token: true
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-dataset_prepared_path: last_prepared_run
-val_set_size: 0.01
-
-adapter:
-lora_model_dir:
-sequence_len: 2048
-max_packed_sequence_len:
-sample_packing: false
-sample_packing_eff_est:
-sample_packing_seq_len_multiplier:
-total_num_tokens:
-
-lora_r:
-lora_alpha:
-lora_dropout:
-lora_target_modules:
-lora_target_linear:
-lora_fan_in_fan_out:
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-
-output_dir: btlm-out
-gradient_accumulation_steps: 1
-micro_batch_size: 1
-num_epochs: 1
-optimizer: adamw_torch
-adam_beta2: 0.95
-adam_eps: 0.000000001
-max_grad_norm: 1.0
-
-torchdistx_path:
-lr_scheduler: cosine
-lr_quadratic_warmup: true
-learning_rate: 0.000085
-train_on_inputs: true
-group_by_length: false
-bf16: true
-fp16: false
-tf32: true
-
-gradient_checkpointing: false
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-
-xformers_attention:
-flash_attention: true
-sdp_attention:
-flash_optimum:
-
-gptq_groupsize:
-gptq_model_v1:
-
-warmup_steps: 32
-eval_steps:
-save_steps:
-save_total_limit:
-
-debug:
-deepspeed:
-weight_decay: 0.1
-special_tokens:
-  pad_token: "<|endoftext|>"
-fsdp:
-#  - full_shard
-#  - auto_wrap
-fsdp_config:
-#  fsdp_state_dict_type: FULL_STATE_DICT
-#  fsdp_transformer_layer_cls_to_wrap: BTLMBlock

examples/cerebras/qlora.yml

@@ -1,4 +1,5 @@
 base_model: cerebras/Cerebras-GPT-1.3B
+base_model_config: cerebras/Cerebras-GPT-1.3B
 load_in_8bit: false
 load_in_4bit: true
 strict: false
@@ -6,7 +7,7 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.01
 adapter: qlora
 lora_model_dir:
@@ -49,7 +50,7 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-eval_steps: 0.05
+eval_steps: 20
 save_steps:
 debug:
 deepspeed:

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

@@ -1,67 +0,0 @@
-base_model: codellama/CodeLlama-13b-hf
-model_type: LlamaForCausalLM
-tokenizer_type: CodeLlamaTokenizer
-is_llama_derived_model: true
-
-load_in_8bit: true
-load_in_4bit: false
-strict: false
-
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.01
-output_dir: ./lora-out
-
-sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
-
-adapter: lora
-lora_model_dir:
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-lora_fan_in_fan_out:
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 4
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: true
-fp16: false
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_steps: 10
-eval_steps: 0.05
-save_steps:
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"

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

@@ -1,69 +0,0 @@
-base_model: codellama/CodeLlama-13b-hf
-model_type: LlamaForCausalLM
-tokenizer_type: CodeLlamaTokenizer
-is_llama_derived_model: true
-
-load_in_8bit: false
-load_in_4bit: true
-strict: false
-
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.01
-output_dir: ./qlora-out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules:
-lora_target_linear: true
-lora_fan_in_fan_out:
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 4
-optimizer: paged_adamw_32bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: true
-fp16: false
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_steps: 10
-eval_steps: 0.05
-save_steps:
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"

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

@@ -1,67 +0,0 @@
-base_model: codellama/CodeLlama-34b-hf
-model_type: LlamaForCausalLM
-tokenizer_type: CodeLlamaTokenizer
-is_llama_derived_model: true
-
-load_in_8bit: true
-load_in_4bit: false
-strict: false
-
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.01
-output_dir: ./lora-out
-
-sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
-
-adapter: lora
-lora_model_dir:
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-lora_fan_in_fan_out:
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 4
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: true
-fp16: false
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_steps: 10
-eval_steps: 0.05
-save_steps:
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"

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

@@ -1,69 +0,0 @@
-base_model: codellama/CodeLlama-34b-hf
-model_type: LlamaForCausalLM
-tokenizer_type: CodeLlamaTokenizer
-is_llama_derived_model: true
-
-load_in_8bit: false
-load_in_4bit: true
-strict: false
-
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.01
-output_dir: ./qlora-out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules:
-lora_target_linear: true
-lora_fan_in_fan_out:
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 4
-optimizer: paged_adamw_32bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: true
-fp16: false
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_steps: 10
-eval_steps: 0.05
-save_steps:
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"

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

@@ -1,67 +0,0 @@
-base_model: codellama/CodeLlama-7b-hf
-model_type: LlamaForCausalLM
-tokenizer_type: CodeLlamaTokenizer
-is_llama_derived_model: true
-
-load_in_8bit: true
-load_in_4bit: false
-strict: false
-
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.01
-output_dir: ./lora-out
-
-sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
-
-adapter: lora
-lora_model_dir:
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-lora_fan_in_fan_out:
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 4
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: true
-fp16: false
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_steps: 10
-eval_steps: 0.05
-save_steps:
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"

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

@@ -1,69 +0,0 @@
-base_model: codellama/CodeLlama-7b-hf
-model_type: LlamaForCausalLM
-tokenizer_type: CodeLlamaTokenizer
-is_llama_derived_model: true
-
-load_in_8bit: false
-load_in_4bit: true
-strict: false
-
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.01
-output_dir: ./qlora-out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules:
-lora_target_linear: true
-lora_fan_in_fan_out:
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 4
-optimizer: paged_adamw_32bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: true
-fp16: false
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_steps: 10
-eval_steps: 0.05
-save_steps:
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"

examples/code-llama/README.md

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

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

@@ -1,8 +1,8 @@
 base_model: tiiuae/falcon-7b
+base_model_config: tiiuae/falcon-7b
 trust_remote_code: true
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
-is_falcon_derived_model: true
 load_in_8bit: true
 load_in_4bit: false
 gptq: false
@@ -11,7 +11,7 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca:chat
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.01
 adapter: lora
 lora_model_dir:

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

@@ -1,11 +1,11 @@
 # 1b: tiiuae/falcon-rw-1b
 # 40b: tiiuae/falcon-40b
 base_model: tiiuae/falcon-7b
+base_model_config: tiiuae/falcon-7b
 # required by falcon custom model code: https://huggingface.co/tiiuae/falcon-7b/tree/main
 trust_remote_code: true
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
-is_falcon_derived_model: true
 load_in_8bit: false
 # enable 4bit for QLoRA
 load_in_4bit: true
@@ -17,7 +17,7 @@ datasets:
     data_files:
       - Chain-of-Thought/formatted_cot_data/gsm8k_train.json
     type: "alpaca:chat"
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.01
 # enable QLoRA
 adapter: qlora
@@ -53,7 +53,7 @@ output_dir: ./qlora-out
 # decrease if OOM, increase for max VRAM utilization
 micro_batch_size: 1
 gradient_accumulation_steps: 2
-num_epochs: 4
+num_epochs: 3
 # Optimizer for QLoRA
 optimizer: paged_adamw_32bit
 torchdistx_path:

examples/falcon/config-7b.yml

@@ -1,8 +1,8 @@
 base_model: tiiuae/falcon-7b
+base_model_config: tiiuae/falcon-7b
 trust_remote_code: true
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
-is_falcon_derived_model: true
 load_in_8bit: false
 load_in_4bit: false
 gptq: false
@@ -11,7 +11,7 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca:chat
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.01
 adapter:
 lora_model_dir:

examples/gptj/qlora.yml

@@ -1,4 +1,5 @@
 base_model: EleutherAI/gpt-j-6b
+base_model_config: EleutherAI/gpt-j-6b
 load_in_8bit: false
 load_in_4bit: true
 strict: false
@@ -6,7 +7,7 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.01
 adapter: qlora
 lora_model_dir:
@@ -46,7 +47,7 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-eval_steps: 0.05
+eval_steps: 20
 save_steps:
 debug:
 deepspeed:

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

@@ -1,11 +1,12 @@
 base_model: huggyllama/llama-7b
+base_model_config: huggyllama/llama-7b
 model_type: LlamaForCausalLM
 tokenizer_type: LlamaTokenizer
 load_in_8bit: false
 datasets:
   - path: openaccess-ai-collective/jeopardy
     type: jeopardy
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.02
 adapter:
 lora_model_dir:
@@ -24,7 +25,7 @@ wandb_log_model:
 output_dir: ./jeopardy-bot-7b
 gradient_accumulation_steps: 1
 micro_batch_size: 1
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine

examples/llama-2/README.md

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

examples/llama-2/fft_optimized.yml

@@ -1,72 +0,0 @@
-base_model: NousResearch/Llama-2-7b-hf
-model_type: LlamaForCausalLM
-tokenizer_type: LlamaTokenizer
-is_llama_derived_model: true
-
-load_in_8bit: false
-load_in_4bit: false
-strict: false
-
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
-output_dir: ./out
-
-sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
-
-adapter:
-lora_model_dir:
-lora_r:
-lora_alpha:
-lora_dropout:
-lora_target_linear:
-lora_fan_in_fan_out:
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-
-gradient_accumulation_steps: 1
-micro_batch_size: 1
-num_epochs: 1
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: true
-fp16: false
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-flash_attn_cross_entropy: false
-flash_attn_rms_norm: true
-flash_attn_fuse_qkv: false
-flash_attn_fuse_mlp: true
-
-warmup_steps: 100
-eval_steps: 0.05
-eval_table_size:
-save_steps:
-debug:
-deepspeed: #deepspeed/zero2.json # multi-gpu only
-weight_decay: 0.1
-fsdp:
-fsdp_config:
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"

examples/llama-2/gptq-lora.yml

@@ -1,73 +0,0 @@
-base_model: TheBloke/Llama-2-7B-GPTQ
-is_llama_derived_model: false
-gptq: true
-gptq_disable_exllama: true
-model_type: AutoModelForCausalLM
-tokenizer_type: LlamaTokenizer
-tokenizer_use_fast: true
-tokenizer_legacy: true
-load_in_8bit: false
-load_in_4bit: false
-strict: false
-push_dataset_to_hub:
-hf_use_auth_token: true
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.01
-adapter: lora
-lora_model_dir:
-sequence_len: 4096
-sample_packing:
-lora_r: 8
-lora_alpha: 32
-lora_dropout: 0.05
-lora_target_modules:
-  - k_proj
-  - o_proj
-  - q_proj
-  - v_proj
-lora_target_linear:
-lora_fan_in_fan_out:
-wandb_project:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-output_dir: ./model-out
-gradient_accumulation_steps: 1
-micro_batch_size: 1
-num_epochs: 4
-optimizer: adamw_torch
-adam_beta2: 0.95
-adam_eps: 0.00001
-max_grad_norm: 1.0
-torchdistx_path:
-lr_scheduler: cosine
-lr_quadratic_warmup: true
-learning_rate: 0.000017
-train_on_inputs: false
-group_by_length: false
-bf16: false
-fp16: false
-float16: true
-tf32: true
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention:
-sdp_attention:
-flash_optimum:
-warmup_steps: 100
-eval_steps:
-save_steps:
-debug:
-deepspeed:
-weight_decay: 0.1
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"

examples/llama-2/lora.yml

@@ -1,7 +1,7 @@
-base_model: NousResearch/Llama-2-7b-hf
+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
@@ -10,13 +10,12 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.01
 output_dir: ./lora-out
 
 sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
+max_packed_sequence_len: 4096
 
 adapter: lora
 lora_model_dir:
@@ -34,7 +33,7 @@ wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -50,13 +49,11 @@ early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
 logging_steps: 1
-xformers_attention:
-flash_attention: true
+xformers_attention: true
+flash_attention:
 
 warmup_steps: 10
-eval_steps: 0.05
-eval_table_size:
-eval_table_max_new_tokens: 128
+eval_steps: 20
 save_steps:
 debug:
 deepspeed:
@@ -67,3 +64,4 @@ special_tokens:
   bos_token: "<s>"
   eos_token: "</s>"
   unk_token: "<unk>"
+  pad_token: "<pad>"

examples/llama-2/qlora.yml

@@ -1,7 +1,7 @@
-base_model: NousResearch/Llama-2-7b-hf
+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
@@ -10,7 +10,7 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.01
 output_dir: ./qlora-out
 
@@ -18,9 +18,7 @@ adapter: qlora
 lora_model_dir:
 
 sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
-
+max_packed_sequence_len: 4096
 lora_r: 32
 lora_alpha: 16
 lora_dropout: 0.05
@@ -36,7 +34,7 @@ wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 4
+num_epochs: 3
 optimizer: paged_adamw_32bit
 lr_scheduler: cosine
 learning_rate: 0.0002
@@ -52,12 +50,11 @@ early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
 logging_steps: 1
-xformers_attention:
-flash_attention: true
+xformers_attention: true
+flash_attention:
 
 warmup_steps: 10
-eval_steps: 0.05
-eval_table_size:
+eval_steps: 20
 save_steps:
 debug:
 deepspeed:
@@ -68,3 +65,4 @@ special_tokens:
   bos_token: "<s>"
   eos_token: "</s>"
   unk_token: "<unk>"
+  pad_token: "<pad>"

examples/llama-2/relora.yml

@@ -1,73 +0,0 @@
-base_model: NousResearch/Llama-2-7b-hf
-model_type: LlamaForCausalLM
-tokenizer_type: LlamaTokenizer
-is_llama_derived_model: true
-
-load_in_8bit: false
-load_in_4bit: true
-strict: false
-
-datasets:
-  - path: teknium/GPT4-LLM-Cleaned
-    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.01
-output_dir: ./relora-out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
-
-lora_r: 8
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_modules:
-lora_target_linear: true
-lora_fan_in_fan_out:
-
-relora_steps: 150
-relora_warmup_steps: 10
-relora_cpu_offload: false
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 4
-num_epochs: 4
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: true
-fp16: false
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_steps: 10
-eval_steps: 0.05
-save_steps: 50
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"

examples/llama-2/tiny-llama.yml

@@ -1,68 +0,0 @@
-base_model: PY007/TinyLlama-1.1B-step-50K-105b
-
-model_type: LlamaForCausalLM
-tokenizer_type: LlamaTokenizer
-is_llama_derived_model: true
-
-load_in_8bit: true
-load_in_4bit: false
-strict: false
-
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.01
-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: 4
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: true
-fp16: false
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_steps: 10
-eval_steps: 0.05
-eval_table_size:
-save_steps:
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"

examples/mistral/README.md

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

examples/mistral/config.yml

@@ -1,61 +0,0 @@
-base_model: mistralai/Mistral-7B-v0.1
-model_type: MistralForCausalLM
-tokenizer_type: LlamaTokenizer
-is_mistral_derived_model: true
-
-load_in_8bit: false
-load_in_4bit: false
-strict: false
-
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-dataset_prepared_path:
-val_set_size: 0.01
-output_dir: ./out
-
-sequence_len: 8192
-sample_packing: true
-pad_to_sequence_len: true
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 4
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.000005
-
-train_on_inputs: false
-group_by_length: false
-bf16: true
-fp16: false
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_steps: 10
-eval_steps: 0.05
-eval_table_size:
-eval_table_max_new_tokens: 128
-save_steps:
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
-  bos_token: "<s>"
-  eos_token: "</s>"
-  unk_token: "<unk>"

examples/mistral/qlora.yml

@@ -1,78 +0,0 @@
-base_model: mistralai/Mistral-7B-v0.1
-model_type: MistralForCausalLM
-tokenizer_type: LlamaTokenizer
-is_mistral_derived_model: true
-
-load_in_8bit: false
-load_in_4bit: true
-strict: false
-
-datasets:
-  - path: mhenrichsen/alpaca_2k_test
-    type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.01
-output_dir: ./qlora-out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 8192
-sample_packing: true
-pad_to_sequence_len: true
-
-lora_r: 32
-lora_alpha: 16
-lora_dropout: 0.05
-lora_target_linear: true
-lora_fan_in_fan_out:
-lora_target_modules:
-  - gate_proj
-  - down_proj
-  - up_proj
-  - q_proj
-  - v_proj
-  - k_proj
-  - o_proj
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-
-gradient_accumulation_steps: 4
-micro_batch_size: 2
-num_epochs: 1
-optimizer: adamw_bnb_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-train_on_inputs: false
-group_by_length: false
-bf16: true
-fp16: false
-tf32: false
-
-gradient_checkpointing: true
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_steps: 10
-eval_steps: 0.05
-eval_table_size:
-eval_table_max_new_tokens: 128
-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/config.yml

@@ -1,11 +1,12 @@
 base_model: mosaicml/mpt-7b
+base_model_config: mosaicml/mpt-7b
 tokenizer_type: AutoTokenizer
 trust_remote_code: true  # required for mpt as their model class is not merged into transformers yet
 load_in_8bit: false
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.02
 adapter:
 lora_model_dir:
@@ -26,7 +27,7 @@ wandb_log_model:
 output_dir: ./mpt-alpaca-7b
 gradient_accumulation_steps: 1
 micro_batch_size: 1
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine

examples/openllama-3b/config.yml

@@ -1,4 +1,5 @@
-base_model: openlm-research/open_llama_3b_v2
+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
@@ -8,12 +9,12 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.02
 adapter:
 lora_model_dir:
-sequence_len: 1024
-sample_packing: true
+sequence_len: 256
+max_packed_sequence_len:
 lora_r:
 lora_alpha:
 lora_dropout:
@@ -28,11 +29,11 @@ wandb_log_model:
 output_dir: ./openllama-out
 gradient_accumulation_steps: 1
 micro_batch_size: 1
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
-learning_rate: 0.000003
+learning_rate: 0.00001
 train_on_inputs: false
 group_by_length: false
 float16: true
@@ -44,12 +45,12 @@ early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
 logging_steps: 1
-xformers_attention:
-flash_attention: true
+xformers_attention: true
+flash_attention:
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 20
-eval_steps: 0.05
+warmup_steps: 10
+eval_steps: 50
 save_steps:
 debug:
 deepspeed:

examples/openllama-3b/lora.yml

@@ -1,4 +1,5 @@
-base_model: openlm-research/open_llama_3b_v2
+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
@@ -8,12 +9,12 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.02
 adapter: lora
 lora_model_dir:
-sequence_len: 1024
-sample_packing: true
+sequence_len: 256
+max_packed_sequence_len:
 lora_r: 8
 lora_alpha: 16
 lora_dropout: 0.0
@@ -32,9 +33,9 @@ wandb_watch:
 wandb_run_id:
 wandb_log_model:
 output_dir: ./lora-out
-gradient_accumulation_steps: 1
-micro_batch_size: 2
-num_epochs: 4
+batch_size: 16
+micro_batch_size: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine
@@ -49,16 +50,16 @@ early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
 logging_steps: 1
-xformers_attention:
-flash_attention: true
+xformers_attention: true
+flash_attention:
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 20
-eval_steps: 0.05
+warmup_steps: 10
+eval_steps: 50
 save_steps:
 debug:
 deepspeed:
-weight_decay: 0.1
+weight_decay: 0.0
 fsdp:
 fsdp_config:
 special_tokens:

examples/openllama-3b/qlora.yml

@@ -1,4 +1,5 @@
-base_model: openlm-research/open_llama_3b_v2
+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
@@ -8,12 +9,12 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.01
 adapter: qlora
 lora_model_dir:
-sequence_len: 1024
-sample_packing: true
+sequence_len: 2048
+max_packed_sequence_len: 2048
 lora_r: 8
 lora_alpha: 32
 lora_dropout: 0.05
@@ -26,33 +27,33 @@ wandb_watch:
 wandb_run_id:
 wandb_log_model:
 output_dir: ./qlora-out
-gradient_accumulation_steps: 1
-micro_batch_size: 2
-num_epochs: 4
+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: false
-fp16: true
-tf32: false
+bf16: true
+fp16: false
+tf32: true
 gradient_checkpointing: true
 early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
 logging_steps: 1
-xformers_attention:
-flash_attention: true
+xformers_attention: true
+flash_attention:
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 20
-eval_steps: 0.05
+warmup_steps: 10
+eval_steps: 20
 save_steps:
 debug:
 deepspeed:
-weight_decay: 0.1
+weight_decay: 0.0
 fsdp:
 fsdp_config:
 special_tokens:

examples/phi/README.md

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

examples/phi/phi-ft.yml

@@ -1,74 +0,0 @@
-base_model: microsoft/phi-1_5
-model_type: MixFormerSequentialForCausalLM
-tokenizer_type: AutoTokenizer
-is_llama_derived_model: false
-trust_remote_code: true
-
-load_in_8bit: false
-load_in_4bit: false
-strict: false
-
-datasets:
-  - path: garage-bAInd/Open-Platypus
-    type: alpaca
-
-dataset_prepared_path:
-val_set_size: 0.05
-output_dir: ./phi-sft-out
-
-sequence_len: 2048
-sample_packing: true
-pad_to_sequence_len:
-
-adapter:
-lora_model_dir:
-lora_r:
-lora_alpha:
-lora_dropout:
-lora_target_linear:
-lora_fan_in_fan_out:
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-
-gradient_accumulation_steps: 1
-micro_batch_size: 1
-num_epochs: 4
-optimizer: adamw_torch
-adam_beta2: 0.95
-adam_epsilon: 0.00001
-max_grad_norm: 1.0
-lr_scheduler: cosine
-learning_rate: 0.000003
-
-train_on_inputs: false
-group_by_length: true
-bf16: true
-fp16: false
-tf32: true
-
-gradient_checkpointing:
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention:
-
-warmup_steps: 100
-eval_steps: 0.05
-save_steps:
-debug:
-deepspeed:
-weight_decay: 0.1
-fsdp:
-fsdp_config:
-resize_token_embeddings_to_32x: true
-special_tokens:
-  bos_token: "<|endoftext|>"
-  eos_token: "<|endoftext|>"
-  unk_token: "<|endoftext|>"
-  pad_token: "<|endoftext|>"

examples/phi/phi-qlora.yml

@@ -1,74 +0,0 @@
-base_model: microsoft/phi-1_5
-model_type: AutoModelForCausalLM
-tokenizer_type: AutoTokenizer
-is_llama_derived_model: false
-trust_remote_code: true
-
-load_in_8bit: false
-load_in_4bit: true
-strict: false
-
-datasets:
-  - path: garage-bAInd/Open-Platypus
-    type: alpaca
-
-dataset_prepared_path:
-val_set_size: 0.05
-output_dir: ./phi-sft-out
-
-sequence_len: 1024
-sample_packing: false  # not CURRENTLY compatible with LoRAs
-pad_to_sequence_len:
-
-adapter: qlora
-lora_model_dir:
-lora_r: 64
-lora_alpha: 32
-lora_dropout: 0.05
-lora_target_linear: true
-lora_fan_in_fan_out:
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_run_id:
-wandb_log_model:
-
-gradient_accumulation_steps: 1
-micro_batch_size: 1
-num_epochs: 4
-optimizer: adamw_torch
-adam_beta2: 0.95
-adam_epsilon: 0.00001
-max_grad_norm: 1.0
-lr_scheduler: cosine
-learning_rate: 0.000003
-
-train_on_inputs: false
-group_by_length: true
-bf16: true
-fp16: false
-tf32: true
-
-gradient_checkpointing:
-early_stopping_patience:
-resume_from_checkpoint:
-local_rank:
-logging_steps: 1
-xformers_attention:
-flash_attention:
-
-warmup_steps: 100
-eval_steps: 0.05
-save_steps:
-debug:
-deepspeed:
-weight_decay: 0.1
-fsdp:
-fsdp_config:
-resize_token_embeddings_to_32x: true
-special_tokens:
-  bos_token: "<|endoftext|>"
-  eos_token: "<|endoftext|>"
-  unk_token: "<|endoftext|>"
-  pad_token: "<|endoftext|>"

examples/pythia-12b/config.yml

@@ -1,4 +1,5 @@
 base_model: EleutherAI/pythia-12b-deduped
+base_model_config: EleutherAI/pythia-12b-deduped
 base_model_ignore_patterns: pytorch*  # prefer safetensors
 model_type: GPTNeoXForCausalLM
 tokenizer_type: AutoTokenizer
@@ -9,7 +10,7 @@ device_map: auto
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.05
 adapter:
 lora_model_dir:
@@ -46,3 +47,4 @@ local_rank:
 gradient_checkpointing: true
 fsdp:
 fsdp_config:
+collator_pad_to_longest: true

examples/pythia/lora.yml

@@ -1,9 +1,10 @@
 base_model: EleutherAI/pythia-1.4b-deduped
+base_model_config: EleutherAI/pythia-1.4b-deduped
 load_in_8bit: true
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.05
 adapter: lora
 lora_model_dir:
@@ -23,15 +24,15 @@ wandb_log_model:
 output_dir: ./lora-alpaca-pythia
 gradient_accumulation_steps: 1
 micro_batch_size: 4
-num_epochs: 4
+num_epochs: 3
 learning_rate: 0.00001
 train_on_inputs: false
 group_by_length: false
-bf16: true
-tf32: true
+bf16: True
+tf32: True
 early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
 weight_decay: 0.1
-eval_steps: 0.05
+eval_steps: 20
 logging_steps: 1

examples/redpajama/config-3b.yml

@@ -1,4 +1,5 @@
 base_model: togethercomputer/RedPajama-INCITE-Chat-3B-v1
+base_model_config: togethercomputer/RedPajama-INCITE-Chat-3B-v1
 model_type: GPTNeoXForCausalLM
 tokenizer_type: AutoTokenizer
 trust_remote_code:
@@ -6,7 +7,7 @@ load_in_8bit: false
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.02
 adapter:
 lora_model_dir:
@@ -27,7 +28,7 @@ wandb_log_model:
 output_dir: ./redpajama-alpaca-3b
 batch_size: 4
 micro_batch_size: 1
-num_epochs: 4
+num_epochs: 3
 optimizer: adamw_bnb_8bit
 torchdistx_path:
 lr_scheduler: cosine

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

@@ -1,10 +1,11 @@
 base_model: replit/replit-code-v1-3b
+base_model_config: replit/replit-code-v1-3b
 trust_remote_code: true
 load_in_8bit: false
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.05
 adapter: lora
 lora_model_dir:
@@ -26,7 +27,7 @@ wandb_log_model:
 output_dir: ./lora-replit
 batch_size: 8
 micro_batch_size: 1
-num_epochs: 4
+num_epochs: 3
 optimizer:
 torchdistx_path:
 lr_scheduler:

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

@@ -1,6 +1,7 @@
 # An example finetuning Saleforce's XGen-7b model with 8k context using qlora
 # on Tim Dettmer's Guanaco dataset.
 base_model: Salesforce/xgen-7b-8k-base
+base_model_config: Salesforce/xgen-7b-8k-base
 trust_remote_code: true
 model_type: AutoModelForCausalLM
 tokenizer_type: AutoTokenizer
@@ -15,7 +16,7 @@ datasets:
     data_files:
       - openassistant_best_replies_train.jsonl
     type: "completion"
-dataset_prepared_path:
+dataset_prepared_path: last_run_prepared
 val_set_size: 0.01
 # enable QLoRA
 adapter: qlora
@@ -51,7 +52,7 @@ output_dir: ./qlora-out
 # decrease if OOM, increase for max VRAM utilization
 micro_batch_size: 1
 gradient_accumulation_steps: 1
-num_epochs: 4
+num_epochs: 3
 # Optimizer for QLoRA
 optimizer: paged_adamw_32bit
 torchdistx_path:

gitbook/README.md

@@ -1 +0,0 @@
-# Page

gitbook/SUMMARY.md

@@ -1,4 +0,0 @@
-# Table of contents
-
-* [Page](README.md)
-* [Small dev details](small-dev-details.md)

gitbook/small-dev-details.md

@@ -1,3 +0,0 @@
-# Small dev details
-
-/

requirements.txt

@@ -1,27 +1,20 @@
---extra-index-url https://download.pytorch.org/whl/cu118
---extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/
-torch==2.0.1
-auto-gptq
-packaging
 peft @ git+https://github.com/huggingface/peft.git
-transformers @ git+https://github.com/huggingface/transformers.git@acc394c4f5e1283c19783581790b3dc3105a3697
+transformers @ git+https://github.com/huggingface/transformers.git
 bitsandbytes>=0.41.1
-accelerate @ git+https://github.com/huggingface/accelerate@80da9cfb09bb3cc9f1b385cb55d6b90d025a5fd9
-deepspeed
+accelerate @ git+https://github.com/huggingface/accelerate@2a289f6108e77a77a4efffb3f6316bc98538413b
 addict
 fire
-PyYAML>=6.0
+PyYAML==6.0
 datasets
-flash-attn>=2.3.0
+accelerate>=0.19.0
 sentencepiece
 wandb
 einops
-xformers>=0.0.22
+xformers
 optimum
 hf_transfer
-colorama
 numba
-numpy>=1.24.4
+numpy==1.24.4
 # qlora things
 bert-score==0.3.13
 evaluate==0.4.0
@@ -29,6 +22,3 @@ rouge-score==0.1.2
 scipy
 scikit-learn==1.2.2
 pynvml
-art
-fschat==0.2.29
-tensor_parallel

scripts/alpaca_json_to_jsonl.py

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

scripts/finetune.py

@@ -1,52 +1,384 @@
 """Prepare and train a model on a dataset. Can also infer from a model or merge lora"""
+
+import importlib
 import logging
+import os
+import random
+import signal
+import sys
 from pathlib import Path
+from typing import Any, Dict, List, Optional, Union
 
 import fire
-import transformers
+import torch
+import yaml
 
-from axolotl.cli import (
-    check_accelerate_default_config,
-    check_user_token,
-    do_inference,
-    do_merge_lora,
-    load_cfg,
-    load_datasets,
-    print_axolotl_text_art,
+# add src to the pythonpath so we don't need to pip install this
+from optimum.bettertransformer import BetterTransformer
+from transformers import GenerationConfig, TextStreamer
+
+from axolotl.logging_config import configure_logging
+from axolotl.utils.bench import log_gpu_memory_usage
+from axolotl.utils.data import load_prepare_datasets, load_pretraining_dataset
+from axolotl.utils.dict import DictDefault
+from axolotl.utils.distributed import barrier, 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 (
+    calculate_total_num_steps,
+    process_datasets_for_packing,
+    setup_trainer,
 )
-from axolotl.cli.shard import shard
-from axolotl.common.cli import TrainerCliArgs
-from axolotl.train import train
+from axolotl.utils.validation import validate_config
+from axolotl.utils.wandb import setup_wandb_env_vars
 
-LOG = logging.getLogger("axolotl.scripts.finetune")
+project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
+src_dir = os.path.join(project_root, "src")
+sys.path.insert(0, src_dir)
+
+configure_logging()
+LOG = logging.getLogger("axolotl.scripts")
 
 
-def do_cli(config: Path = Path("examples/"), **kwargs):
-    print_axolotl_text_art()
-    LOG.warning(
-        str(
-            PendingDeprecationWarning(
-                "scripts/finetune.py will be replaced with calling axolotl.cli.train"
-            )
+DEFAULT_DATASET_PREPARED_PATH = "last_run_prepared"
+os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
+
+
+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}
+
+
+def get_multi_line_input() -> Optional[str]:
+    print("Give me an instruction (Ctrl + D to finish): ")
+    instruction = ""
+    for line in sys.stdin:
+        instruction += line  # pylint: disable=consider-using-join
+    # instruction = pathlib.Path("/proc/self/fd/0").read_text()
+    return instruction
+
+
+def do_inference(cfg, model, tokenizer, prompter: Optional[str]):
+    default_tokens = {"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>"}
+
+    for token, symbol in default_tokens.items():
+        # If the token isn't already specified in the config, add it
+        if not (cfg.special_tokens and token in cfg.special_tokens):
+            tokenizer.add_special_tokens({token: symbol})
+
+    prompter_module = None
+    if prompter:
+        prompter_module = getattr(
+            importlib.import_module("axolotl.prompters"), prompter
         )
+
+    if cfg.landmark_attention:
+        from axolotl.monkeypatch.llama_landmark_attn import set_model_mem_id
+
+        set_model_mem_id(model, tokenizer)
+        model.set_mem_cache_args(
+            max_seq_len=255, mem_freq=50, top_k=5, max_cache_size=None
+        )
+
+    while True:
+        print("=" * 80)
+        # support for multiline inputs
+        instruction = get_multi_line_input()
+        if not instruction:
+            return
+        if prompter_module:
+            prompt: str = next(
+                prompter_module().build_prompt(instruction=instruction.strip("\n"))
+            )
+        else:
+            prompt = instruction.strip()
+        batch = tokenizer(prompt, return_tensors="pt", add_special_tokens=True)
+
+        print("=" * 40)
+        model.eval()
+        with torch.no_grad():
+            generation_config = GenerationConfig(
+                repetition_penalty=1.1,
+                max_new_tokens=1024,
+                temperature=0.9,
+                top_p=0.95,
+                top_k=40,
+                bos_token_id=tokenizer.bos_token_id,
+                eos_token_id=tokenizer.eos_token_id,
+                pad_token_id=tokenizer.pad_token_id,
+                do_sample=True,
+                use_cache=True,
+                return_dict_in_generate=True,
+                output_attentions=False,
+                output_hidden_states=False,
+                output_scores=False,
+            )
+            streamer = TextStreamer(tokenizer)
+            generated = model.generate(
+                inputs=batch["input_ids"].to(cfg.device),
+                generation_config=generation_config,
+                streamer=streamer,
+            )
+        print("=" * 40)
+        print(tokenizer.decode(generated["sequences"].cpu().tolist()[0]))
+
+
+def choose_config(path: Path):
+    yaml_files = list(path.glob("*.yml"))
+
+    if not yaml_files:
+        raise ValueError(
+            "No YAML config files found in the specified directory. Are you using a .yml extension?"
+        )
+
+    print("Choose a YAML file:")
+    for idx, file in enumerate(yaml_files):
+        print(f"{idx + 1}. {file}")
+
+    chosen_file = None
+    while chosen_file is None:
+        try:
+            choice = int(input("Enter the number of your choice: "))
+            if 1 <= choice <= len(yaml_files):
+                chosen_file = yaml_files[choice - 1]
+            else:
+                print("Invalid choice. Please choose a number from the list.")
+        except ValueError:
+            print("Invalid input. Please enter a number.")
+
+    return chosen_file
+
+
+def check_not_in(list1: List[str], list2: Union[Dict[str, Any], List[str]]) -> bool:
+    return not any(el in list2 for el in list1)
+
+
+def train(
+    config: Path = Path("configs/"),
+    prepare_ds_only: bool = False,
+    **kwargs,
+):
+    if Path(config).is_dir():
+        config = choose_config(config)
+
+    # load the config from the yaml file
+    with open(config, encoding="utf-8") as file:
+        cfg: DictDefault = DictDefault(yaml.safe_load(file))
+    # if there are any options passed in the cli, if it is something that seems valid from the yaml,
+    # then overwrite the value
+    cfg_keys = cfg.keys()
+    for k, _ in kwargs.items():
+        # if not strict, allow writing to cfg even if it's not in the yml already
+        if k in cfg_keys or not cfg.strict:
+            # handle booleans
+            if isinstance(cfg[k], bool):
+                cfg[k] = bool(kwargs[k])
+            else:
+                cfg[k] = kwargs[k]
+
+    validate_config(cfg)
+
+    # setup some derived config / hyperparams
+    cfg.gradient_accumulation_steps = cfg.gradient_accumulation_steps or (
+        cfg.batch_size // cfg.micro_batch_size
     )
-    parsed_cfg = load_cfg(config, **kwargs)
-    check_accelerate_default_config()
-    check_user_token()
-    parser = transformers.HfArgumentParser((TrainerCliArgs))
-    parsed_cli_args, _ = parser.parse_args_into_dataclasses(
-        return_remaining_strings=True
+    cfg.batch_size = (
+        cfg.batch_size or cfg.micro_batch_size * cfg.gradient_accumulation_steps
     )
-    if parsed_cli_args.inference:
-        do_inference(cfg=parsed_cfg, cli_args=parsed_cli_args)
-    elif parsed_cli_args.merge_lora:
-        do_merge_lora(cfg=parsed_cfg, cli_args=parsed_cli_args)
-    elif parsed_cli_args.shard:
-        shard(cfg=parsed_cfg, cli_args=parsed_cli_args)
+    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
+
+    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
+
+    if cfg.tf32:
+        torch.backends.cuda.matmul.allow_tf32 = True
+
+    # load the tokenizer first
+    tokenizer_config = cfg.tokenizer_config or cfg.base_model_config
+    LOG.info(f"loading tokenizer... {tokenizer_config}")
+    tokenizer = load_tokenizer(tokenizer_config, cfg.tokenizer_type, cfg)
+
+    if (
+        check_not_in(["shard", "merge_lora"], kwargs) and not cfg.inference
+    ):  # don't need to load dataset for these
+        if not cfg.pretraining_dataset:
+            train_dataset, eval_dataset = load_prepare_datasets(
+                tokenizer, cfg, DEFAULT_DATASET_PREPARED_PATH
+            )
+        else:
+            train_dataset = load_pretraining_dataset(
+                cfg.pretraining_dataset,
+                tokenizer,
+                max_tokens=cfg.sequence_len,
+                seed=cfg.seed 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
+
+        if is_main_process():
+            # process on rank 0 first so it gets cached so other ranks load from cache
+            train_dataset, eval_dataset = process_datasets_for_packing(
+                cfg, train_dataset, eval_dataset
+            )
+        barrier()
+        if not is_main_process():
+            train_dataset, eval_dataset = process_datasets_for_packing(
+                cfg, train_dataset, eval_dataset
+            )
+        barrier()
+        total_num_steps = calculate_total_num_steps(cfg, train_dataset, 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
+
+    log_gpu_memory_usage(LOG, "baseline", cfg.device)
+
+    # Load the model and tokenizer
+    LOG.info("loading model and (optionally) peft_config...")
+    model, peft_config = load_model(cfg, tokenizer)
+
+    safe_serialization = cfg.save_safetensors is True
+
+    if "merge_lora" in kwargs and cfg.adapter is not None:
+        LOG.info("running merge of LoRA with base model")
+        model = model.merge_and_unload()
+        model.to(dtype=torch.float16)
+
+        if cfg.local_rank == 0:
+            LOG.info("saving merged model")
+            model.save_pretrained(
+                str(Path(cfg.output_dir) / "merged"),
+                safe_serialization=safe_serialization,
+            )
+            tokenizer.save_pretrained(str(Path(cfg.output_dir) / "merged"))
+        return
+
+    if cfg.inference:
+        LOG.info("calling do_inference function")
+        prompter: Optional[str] = "AlpacaPrompter"
+        if "prompter" in kwargs:
+            if kwargs["prompter"] == "None":
+                prompter = None
+            else:
+                prompter = kwargs["prompter"]
+        do_inference(cfg, model, tokenizer, prompter=prompter)
+        return
+
+    if "shard" in kwargs:
+        model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)
+        return
+
+    trainer = setup_trainer(
+        cfg, train_dataset, eval_dataset, model, tokenizer, total_num_steps
+    )
+
+    model.config.use_cache = False
+
+    if torch.__version__ >= "2" and sys.platform != "win32":
+        LOG.info("Compiling torch model")
+        model = torch.compile(model)
+
+    # go ahead and presave, so we have the adapter config available to inspect
+    if peft_config:
+        LOG.info(f"Pre-saving adapter config to {cfg.output_dir}")
+        peft_config.save_pretrained(cfg.output_dir)
+
+    # In case we want to stop early with ctrl+c, this is a nice to have to save the pretrained model
+    if cfg.local_rank == 0:
+
+        def terminate_handler(_, __, model):
+            if cfg.flash_optimum:
+                model = BetterTransformer.reverse(model)
+            model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)
+            sys.exit(0)
+
+        signal.signal(
+            signal.SIGINT, lambda signum, frame: terminate_handler(signum, frame, model)
+        )
+
+    LOG.info("Starting trainer...")
+    if cfg.group_by_length:
+        LOG.info("hang tight... sorting dataset for group_by_length")
+    resume_from_checkpoint = cfg.resume_from_checkpoint
+    if cfg.resume_from_checkpoint is None and cfg.auto_resume_from_checkpoints:
+        possible_checkpoints = [
+            str(cp) for cp in Path(cfg.output_dir).glob("checkpoint-*")
+        ]
+        if len(possible_checkpoints) > 0:
+            sorted_paths = sorted(
+                possible_checkpoints,
+                key=lambda path: int(path.split("-")[-1]),
+            )
+            resume_from_checkpoint = sorted_paths[-1]
+            LOG.info(
+                f"Using Auto-resume functionality to start with checkpoint at {resume_from_checkpoint}"
+            )
+
+    if not Path(cfg.output_dir).is_dir():
+        os.makedirs(cfg.output_dir, exist_ok=True)
+    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:
-        dataset_meta = load_datasets(cfg=parsed_cfg, cli_args=parsed_cli_args)
-        train(cfg=parsed_cfg, cli_args=parsed_cli_args, dataset_meta=dataset_meta)
+        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)
+
+        if cfg.adapter == "lora" and cfg.relora_steps:
+            model = model.merge_and_unload()
+
+        model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)
+
+    # trainer.save_model(cfg.output_dir)  # TODO this may be needed for deepspeed to work? need to review another time
 
 
 if __name__ == "__main__":
-    fire.Fire(do_cli)
+    fire.Fire(train)

setup.py

@@ -2,53 +2,31 @@
 
 from setuptools import find_packages, setup
 
-
-def parse_requirements():
-    _install_requires = []
-    _dependency_links = []
-    with open("./requirements.txt", encoding="utf-8") as requirements_file:
-        lines = [r.strip() for r in requirements_file.readlines()]
-        for line in lines:
-            if line.startswith("--extra-index-url"):
-                # Handle custom index URLs
-                _, url = line.split()
-                _dependency_links.append(url)
-            elif (
-                "flash-attn" not in line
-                and "deepspeed" not in line
-                and line
-                and line[0] != "#"
-            ):
-                # Handle standard packages
-                _install_requires.append(line)
-
-    # TODO(wing) remove once xformers release supports torch 2.1.0
-    if "torch==2.1.0" in _install_requires:
-        _install_requires.pop(_install_requires.index("xformers>=0.0.22"))
-        _install_requires.append(
-            "xformers @ git+https://github.com/facebookresearch/xformers.git@main"
-        )
-
-    return _install_requires, _dependency_links
-
-
-install_requires, dependency_links = parse_requirements()
-
+install_requires = []
+with open("./requirements.txt", encoding="utf-8") as requirements_file:
+    # don't include peft yet until we check the int4
+    # need to manually install peft for now...
+    reqs = [r.strip() for r in requirements_file.readlines() if "peft" not in r]
+    reqs = [r for r in reqs if r and r[0] != "#"]
+    for r in reqs:
+        install_requires.append(r)
 
 setup(
     name="axolotl",
-    version="0.3.0",
-    description="LLM Trainer",
-    long_description="Axolotl is a tool designed to streamline the fine-tuning of various AI models, offering support for multiple configurations and architectures.",
+    version="0.1",
+    description="You know you're going to axolotl questions",
     package_dir={"": "src"},
     packages=find_packages(),
     install_requires=install_requires,
-    dependency_links=dependency_links,
     extras_require={
-        "flash-attn": [
-            "flash-attn>=2.3.0",
+        "gptq": [
+            "alpaca_lora_4bit @ git+https://github.com/winglian/alpaca_lora_4bit.git@setup_pip",
         ],
-        "deepspeed": [
+        "gptq_triton": [
+            "alpaca_lora_4bit[triton] @ git+https://github.com/winglian/alpaca_lora_4bit.git@setup_pip",
+        ],
+        "extras": [
+            "flash-attn",
             "deepspeed",
         ],
     },

src/axolotl/cli/__init__.py

@@ -1,271 +0,0 @@
-"""Prepare and train a model on a dataset. Can also infer from a model or merge lora"""
-
-import importlib
-import logging
-import os
-import random
-import sys
-from pathlib import Path
-from typing import Any, Dict, List, Optional, Union
-
-import torch
-import yaml
-
-# add src to the pythonpath so we don't need to pip install this
-from accelerate.commands.config import config_args
-from art import text2art
-from huggingface_hub import HfApi
-from huggingface_hub.utils import LocalTokenNotFoundError
-from transformers import GenerationConfig, TextStreamer
-
-from axolotl.common.cli import TrainerCliArgs, load_model_and_tokenizer
-from axolotl.logging_config import configure_logging
-from axolotl.train import TrainDatasetMeta
-from axolotl.utils.config import normalize_config, validate_config
-from axolotl.utils.data import prepare_dataset
-from axolotl.utils.dict import DictDefault
-from axolotl.utils.distributed import is_main_process
-from axolotl.utils.models import load_tokenizer
-from axolotl.utils.tokenization import check_dataset_labels
-from axolotl.utils.wandb_ import setup_wandb_env_vars
-
-project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
-src_dir = os.path.join(project_root, "src")
-sys.path.insert(0, src_dir)
-
-configure_logging()
-LOG = logging.getLogger("axolotl.scripts")
-
-os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
-
-
-def print_axolotl_text_art(suffix=None):
-    font = "nancyj"
-    ascii_text = "  axolotl"
-    if suffix:
-        ascii_text += f"  x  {suffix}"
-    ascii_art = text2art(" axolotl", font=font)
-
-    if is_main_process():
-        print(ascii_art)
-
-
-def get_multi_line_input() -> Optional[str]:
-    print("Give me an instruction (Ctrl + D to submit): ")
-    instruction = ""
-    for line in sys.stdin:
-        instruction += line  # pylint: disable=consider-using-join
-    # instruction = pathlib.Path("/proc/self/fd/0").read_text()
-    return instruction
-
-
-def do_merge_lora(
-    *,
-    cfg: DictDefault,
-    cli_args: TrainerCliArgs,
-):
-    model, tokenizer = load_model_and_tokenizer(cfg=cfg, cli_args=cli_args)
-    safe_serialization = cfg.save_safetensors is True
-
-    LOG.info("running merge of LoRA with base model")
-    model = model.merge_and_unload()
-    model.to(dtype=torch.float16)
-
-    if cfg.local_rank == 0:
-        LOG.info(f"saving merged model to: {str(Path(cfg.output_dir) / 'merged')}")
-        model.save_pretrained(
-            str(Path(cfg.output_dir) / "merged"),
-            safe_serialization=safe_serialization,
-        )
-        tokenizer.save_pretrained(str(Path(cfg.output_dir) / "merged"))
-
-
-def do_inference(
-    *,
-    cfg: DictDefault,
-    cli_args: TrainerCliArgs,
-):
-    model, tokenizer = load_model_and_tokenizer(cfg=cfg, cli_args=cli_args)
-    prompter = cli_args.prompter
-    default_tokens = {"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>"}
-
-    for token, symbol in default_tokens.items():
-        # If the token isn't already specified in the config, add it
-        if not (cfg.special_tokens and token in cfg.special_tokens):
-            tokenizer.add_special_tokens({token: symbol})
-
-    prompter_module = None
-    if prompter:
-        prompter_module = getattr(
-            importlib.import_module("axolotl.prompters"), prompter
-        )
-
-    if cfg.landmark_attention:
-        from axolotl.monkeypatch.llama_landmark_attn import set_model_mem_id
-
-        set_model_mem_id(model, tokenizer)
-        model.set_mem_cache_args(
-            max_seq_len=255, mem_freq=50, top_k=5, max_cache_size=None
-        )
-
-    model = model.to(cfg.device)
-
-    while True:
-        print("=" * 80)
-        # support for multiline inputs
-        instruction = get_multi_line_input()
-        if not instruction:
-            return
-        if prompter_module:
-            prompt: str = next(
-                prompter_module().build_prompt(instruction=instruction.strip("\n"))
-            )
-        else:
-            prompt = instruction.strip()
-        batch = tokenizer(prompt, return_tensors="pt", add_special_tokens=True)
-
-        print("=" * 40)
-        model.eval()
-        with torch.no_grad():
-            generation_config = GenerationConfig(
-                repetition_penalty=1.1,
-                max_new_tokens=1024,
-                temperature=0.9,
-                top_p=0.95,
-                top_k=40,
-                bos_token_id=tokenizer.bos_token_id,
-                eos_token_id=tokenizer.eos_token_id,
-                pad_token_id=tokenizer.pad_token_id,
-                do_sample=True,
-                use_cache=True,
-                return_dict_in_generate=True,
-                output_attentions=False,
-                output_hidden_states=False,
-                output_scores=False,
-            )
-            streamer = TextStreamer(tokenizer)
-            generated = model.generate(
-                inputs=batch["input_ids"].to(cfg.device),
-                generation_config=generation_config,
-                streamer=streamer,
-            )
-        print("=" * 40)
-        print(tokenizer.decode(generated["sequences"].cpu().tolist()[0]))
-
-
-def choose_config(path: Path):
-    yaml_files = list(path.glob("*.yml"))
-
-    if not yaml_files:
-        raise ValueError(
-            "No YAML config files found in the specified directory. Are you using a .yml extension?"
-        )
-
-    if len(yaml_files) == 1:
-        print(f"Using default YAML file '{yaml_files[0]}'")
-        return yaml_files[0]
-
-    print("Choose a YAML file:")
-    for idx, file in enumerate(yaml_files):
-        print(f"{idx + 1}. {file}")
-
-    chosen_file = None
-    while chosen_file is None:
-        try:
-            choice = int(input("Enter the number of your choice: "))
-            if 1 <= choice <= len(yaml_files):
-                chosen_file = yaml_files[choice - 1]
-            else:
-                print("Invalid choice. Please choose a number from the list.")
-        except ValueError:
-            print("Invalid input. Please enter a number.")
-
-    return chosen_file
-
-
-def check_not_in(list1: List[str], list2: Union[Dict[str, Any], List[str]]) -> bool:
-    return not any(el in list2 for el in list1)
-
-
-def load_cfg(config: Path = Path("examples/"), **kwargs):
-    if Path(config).is_dir():
-        config = choose_config(config)
-
-    # load the config from the yaml file
-    with open(config, encoding="utf-8") as file:
-        cfg: DictDefault = DictDefault(yaml.safe_load(file))
-    cfg.axolotl_config_path = config
-    # if there are any options passed in the cli, if it is something that seems valid from the yaml,
-    # then overwrite the value
-    cfg_keys = cfg.keys()
-    for k, _ in kwargs.items():
-        # if not strict, allow writing to cfg even if it's not in the yml already
-        if k in cfg_keys or not cfg.strict:
-            # handle booleans
-            if isinstance(cfg[k], bool):
-                cfg[k] = bool(kwargs[k])
-            else:
-                cfg[k] = kwargs[k]
-
-    validate_config(cfg)
-
-    normalize_config(cfg)
-
-    setup_wandb_env_vars(cfg)
-    return cfg
-
-
-def load_datasets(
-    *,
-    cfg: DictDefault,
-    cli_args: TrainerCliArgs,
-) -> TrainDatasetMeta:
-    tokenizer = load_tokenizer(cfg)
-
-    train_dataset, eval_dataset, total_num_steps, prompters = prepare_dataset(
-        cfg, tokenizer
-    )
-
-    if cli_args.debug or cfg.debug:
-        LOG.info("check_dataset_labels...")
-        check_dataset_labels(
-            train_dataset.select(
-                [
-                    random.randrange(0, len(train_dataset) - 1)  # nosec
-                    for _ in range(cli_args.debug_num_examples)
-                ]
-            ),
-            tokenizer,
-            num_examples=cli_args.debug_num_examples,
-            text_only=cli_args.debug_text_only,
-        )
-
-        LOG.info("printing prompters...")
-        for prompter in prompters:
-            LOG.info(prompter)
-
-    return TrainDatasetMeta(
-        train_dataset=train_dataset,
-        eval_dataset=eval_dataset,
-        total_num_steps=total_num_steps,
-    )
-
-
-def check_accelerate_default_config():
-    if Path(config_args.default_yaml_config_file).exists():
-        LOG.warning(
-            f"accelerate config file found at {config_args.default_yaml_config_file}. This can lead to unexpected errors"
-        )
-
-
-def check_user_token():
-    # Verify if token is valid
-    api = HfApi()
-    try:
-        user_info = api.whoami()
-        return bool(user_info)
-    except LocalTokenNotFoundError:
-        LOG.warning(
-            "Error verifying HuggingFace token. Remember to log in using `huggingface-cli login` and get your access token from https://huggingface.co/settings/tokens if you want to use gated models or datasets."
-        )
-        return False

src/axolotl/cli/inference.py

@@ -1,28 +0,0 @@
-"""
-CLI to run inference on a trained model
-"""
-from pathlib import Path
-
-import fire
-import transformers
-
-from axolotl.cli import do_inference, load_cfg, print_axolotl_text_art
-from axolotl.common.cli import TrainerCliArgs
-
-
-def do_cli(config: Path = Path("examples/"), **kwargs):
-    # pylint: disable=duplicate-code
-    print_axolotl_text_art()
-    parsed_cfg = load_cfg(config, **kwargs)
-    parsed_cfg.sample_packing = False
-    parser = transformers.HfArgumentParser((TrainerCliArgs))
-    parsed_cli_args, _ = parser.parse_args_into_dataclasses(
-        return_remaining_strings=True
-    )
-    parsed_cli_args.inference = True
-
-    do_inference(cfg=parsed_cfg, cli_args=parsed_cli_args)
-
-
-if __name__ == "__main__":
-    fire.Fire(do_cli)

src/axolotl/cli/merge_lora.py

@@ -1,27 +0,0 @@
-"""
-CLI to run merge a trained LoRA into a base model
-"""
-from pathlib import Path
-
-import fire
-import transformers
-
-from axolotl.cli import do_merge_lora, load_cfg, print_axolotl_text_art
-from axolotl.common.cli import TrainerCliArgs
-
-
-def do_cli(config: Path = Path("examples/"), **kwargs):
-    # pylint: disable=duplicate-code
-    print_axolotl_text_art()
-    parser = transformers.HfArgumentParser((TrainerCliArgs))
-    parsed_cli_args, _ = parser.parse_args_into_dataclasses(
-        return_remaining_strings=True
-    )
-    parsed_cli_args.merge_lora = True
-    parsed_cfg = load_cfg(config, merge_lora=True, **kwargs)
-
-    do_merge_lora(cfg=parsed_cfg, cli_args=parsed_cli_args)
-
-
-if __name__ == "__main__":
-    fire.Fire(do_cli)

src/axolotl/cli/preprocess.py

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

src/axolotl/cli/shard.py

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

src/axolotl/cli/train.py

@@ -1,38 +0,0 @@
-"""
-CLI to run training on a model
-"""
-import logging
-from pathlib import Path
-
-import fire
-import transformers
-
-from axolotl.cli import (
-    check_accelerate_default_config,
-    check_user_token,
-    load_cfg,
-    load_datasets,
-    print_axolotl_text_art,
-)
-from axolotl.common.cli import TrainerCliArgs
-from axolotl.train import train
-
-LOG = logging.getLogger("axolotl.cli.train")
-
-
-def do_cli(config: Path = Path("examples/"), **kwargs):
-    # pylint: disable=duplicate-code
-    print_axolotl_text_art()
-    parsed_cfg = load_cfg(config, **kwargs)
-    check_accelerate_default_config()
-    check_user_token()
-    parser = transformers.HfArgumentParser((TrainerCliArgs))
-    parsed_cli_args, _ = parser.parse_args_into_dataclasses(
-        return_remaining_strings=True
-    )
-    dataset_meta = load_datasets(cfg=parsed_cfg, cli_args=parsed_cli_args)
-    train(cfg=parsed_cfg, cli_args=parsed_cli_args, dataset_meta=dataset_meta)
-
-
-if __name__ == "__main__":
-    fire.Fire(do_cli)

src/axolotl/common/cli.py

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

src/axolotl/common/const.py

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

src/axolotl/core/trainer_builder.py

@@ -1,711 +0,0 @@
-"""
-Builder for the training args and trainer
-"""
-
-import abc
-import importlib
-import logging
-import math
-import os
-import sys
-from abc import abstractmethod
-from dataclasses import dataclass, field
-from functools import partial
-from pathlib import Path
-from typing import Optional, Union
-
-import tensor_parallel as tp
-import torch
-import transformers
-from datasets import Dataset
-from torch.optim.lr_scheduler import OneCycleLR
-from torch.utils.data import DataLoader, DistributedSampler, SequentialSampler
-from transformers import EarlyStoppingCallback, Trainer, TrainingArguments
-from transformers.trainer_pt_utils import SequentialDistributedSampler
-
-from axolotl.monkeypatch.relora import ReLoRACallback, ReLoRAScheduler
-from axolotl.utils.callbacks import (
-    EvalFirstStepCallback,
-    GPUStatsCallback,
-    SaveAxolotlConfigtoWandBCallback,
-    SaveBetterTransformerModelCallback,
-    bench_eval_callback_factory,
-    log_prediction_callback_factory,
-)
-from axolotl.utils.collators import DataCollatorForSeq2Seq
-from axolotl.utils.dataloader import MultipackDistributedDataloader
-from axolotl.utils.distributed import is_distributed
-from axolotl.utils.schedulers import get_cosine_schedule_with_quadratic_warmup
-
-try:
-    import torch._dynamo  # pylint: disable=ungrouped-imports
-except ImportError:
-    pass
-
-LOG = logging.getLogger("axolotl.core.trainer_builder")
-
-
-@dataclass
-class AxolotlTrainingArguments(TrainingArguments):
-    """
-    Extend the base TrainingArguments for axolotl helpers
-    """
-
-    lr_quadratic_warmup: bool = field(
-        default=False,
-        metadata={"help": "Use quadratic warmup for cosine scheduling."},
-    )
-    sample_packing: bool = field(
-        default=False,
-        metadata={"help": "Use sample packing for efficient training."},
-    )
-    eval_sample_packing: Optional[bool] = field(
-        default=None,
-        metadata={"help": "Use sample packing for efficient evals."},
-    )
-    sample_packing_efficiency: float = field(
-        default=1.0,
-        metadata={"help": "Sample packing efficiency for calculating batch length."},
-    )
-    max_seq_length: int = field(
-        default=2048,
-        metadata={"help": "The maximum sequence length the model can handle"},
-    )
-    sample_packing_seq_len_multiplier: int = field(
-        default=1,
-        metadata={"help": "the multiplier for the max len for packed sequences"},
-    )
-    relora_steps: Optional[int] = field(
-        default=None,
-        metadata={"help": "how often to reset for ReLoRA"},
-    )
-    relora_warmup_steps: Optional[int] = field(
-        default=None,
-        metadata={"help": "how many warmup steps to take after reset for ReLoRA"},
-    )
-    bench_split: Optional[str] = field(
-        default="eval", metadata={"help": "The benchmark split to run on"}
-    )
-    bench_dataset: Optional[str] = field(
-        default="pharaouk/dharma-1/dharma_1_mini.json",
-        metadata={
-            "help": "Benchmark dataset to use: options are `mmlu-zs`, `mmlu-fs`, or the full path to the dataset file"
-        },
-    )
-    do_bench_eval: Optional[bool] = field(
-        default=False, metadata={"help": "Whether to run the Benchmark evaluation."}
-    )
-    max_bench_samples: Optional[int] = field(
-        default=None,
-        metadata={
-            "help": "If set, only evaluates on `max_bench_samples` of the benchmark dataset."
-        },
-    )
-    bench_source_max_len: int = field(
-        default=2048, metadata={"help": "Maximum source sequence length for bench."}
-    )
-    tensor_parallel: bool = field(
-        default=False, metadata={"help": "Use tensor parallelism to train"}
-    )
-
-
-class AxolotlTrainer(Trainer):
-    """
-    Extend the base Trainer for axolotl helpers
-    """
-
-    args = None  # type: AxolotlTrainingArguments
-
-    def __init__(self, *args, num_epochs=1, bench_data_collator=None, **kwargs):
-        self.num_epochs = num_epochs
-        self.bench_data_collator = bench_data_collator
-        super().__init__(*args, **kwargs)
-
-    def create_scheduler(
-        self, num_training_steps: int, optimizer: torch.optim.Optimizer = None
-    ):
-        """
-        Setup the scheduler. The optimizer of the trainer must have been set up either before this method is called or
-        passed as an argument.
-
-        Args:
-            num_training_steps (int): The number of training steps to do.
-            optimizer (torch.optim.Optimizer): The training optimizer
-        """
-
-        # fmt: off
-        if self.lr_scheduler is None:  # type: ignore  # pylint: disable=access-member-before-definition
-            # fmt: on
-            if (
-                self.args.lr_scheduler_type == "cosine"
-                and self.args.lr_quadratic_warmup is True
-            ):
-                self.lr_scheduler = get_cosine_schedule_with_quadratic_warmup(  # pylint: disable=attribute-defined-outside-init
-                    optimizer,
-                    num_warmup_steps=self.args.get_warmup_steps(num_training_steps),
-                    num_training_steps=num_training_steps,
-                )
-            else:
-                return super().create_scheduler(num_training_steps, optimizer)
-        return self.lr_scheduler
-
-    def _get_train_sampler(self) -> Optional[torch.utils.data.Sampler]:
-        if self.args.world_size > 1 and self.args.sample_packing:
-            return DistributedSampler(
-                self.train_dataset,
-                num_replicas=self.args.world_size,
-                rank=self.args.process_index,
-                seed=self.args.seed,
-            )
-        return super()._get_train_sampler()
-
-    def _get_eval_sampler(
-        self, eval_dataset: Dataset
-    ) -> Optional[torch.utils.data.Sampler]:
-        if (
-            self.args.world_size > 1
-            and self.args.sample_packing
-            and self.args.eval_sample_packing is not False
-        ):
-            return SequentialDistributedSampler(
-                eval_dataset,
-                num_replicas=self.args.world_size,
-                rank=self.args.process_index,
-                batch_size=self.args.per_device_eval_batch_size,
-            )
-        return super()._get_eval_sampler(eval_dataset)
-
-    def get_train_dataloader(self) -> Union[DataLoader, MultipackDistributedDataloader]:
-        if self.args.sample_packing:
-            train_sampler = self._get_train_sampler()
-            return self.accelerator.prepare(
-                MultipackDistributedDataloader(
-                    self.train_dataset,
-                    batch_size=self._train_batch_size,
-                    seq_max_length=self.args.max_seq_length,
-                    collate_fn=self.data_collator,
-                    sampler=train_sampler,
-                    packing_efficiency_estimate=self.args.sample_packing_efficiency,
-                    sample_packing_seq_len_multiplier=self.args.sample_packing_seq_len_multiplier,
-                    device_count=int(os.environ.get("WORLD_SIZE", 1)),
-                    num_epochs=self.num_epochs,
-                )
-            )
-        return super().get_train_dataloader()
-
-    def get_eval_dataloader(
-        self, eval_dataset: Optional[Dataset] = None
-    ) -> Union[DataLoader, MultipackDistributedDataloader]:
-        if self.args.sample_packing and self.args.eval_sample_packing is not False:
-            eval_dataset = (
-                eval_dataset if eval_dataset is not None else self.eval_dataset
-            )
-
-            eval_sampler = self._get_eval_sampler(eval_dataset)
-            return self.accelerator.prepare(
-                MultipackDistributedDataloader(
-                    eval_dataset,
-                    batch_size=self.args.eval_batch_size,
-                    seq_max_length=self.args.max_seq_length,
-                    collate_fn=self.data_collator,
-                    sampler=eval_sampler,
-                    packing_efficiency_estimate=self.args.sample_packing_efficiency,
-                    sample_packing_seq_len_multiplier=self.args.eval_batch_size,
-                    device_count=int(os.environ.get("WORLD_SIZE", 1)),
-                    num_epochs=self.num_epochs,
-                )
-            )
-        return super().get_eval_dataloader(eval_dataset)
-
-    def _get_bench_sampler(
-        self, bench_dataset: Dataset
-    ) -> Optional[torch.utils.data.Sampler]:
-        if self.args.world_size <= 1:
-            return SequentialSampler(bench_dataset)
-        return None
-
-    def get_bench_dataloader(
-        self,
-        bench_dataset: Dataset,
-    ) -> Union[DataLoader, MultipackDistributedDataloader]:
-        dataloader_params = {
-            "batch_size": self.args.eval_batch_size,
-            "collate_fn": self.bench_data_collator,
-            "num_workers": self.args.dataloader_num_workers,
-            "pin_memory": self.args.dataloader_pin_memory,
-        }
-
-        if not isinstance(bench_dataset, torch.utils.data.IterableDataset):
-            dataloader_params["sampler"] = self._get_bench_sampler(bench_dataset)
-            dataloader_params["drop_last"] = self.args.dataloader_drop_last
-
-        return DataLoader(bench_dataset, **dataloader_params)
-        # return self.accelerator.prepare(DataLoader(bench_dataset, **dataloader_params))
-
-    def compute_loss(self, model, inputs, return_outputs=False):
-        # use one's weighted cross entropy loss calc
-        # if self.args.sample_packing:
-        #     labels = inputs.pop("labels")
-        #     outputs = model(**inputs)
-        #     loss = trainer_weighted_loss(outputs, labels, shift_labels=True)
-        #     return (loss, outputs) if return_outputs else loss
-        return super().compute_loss(model, inputs, return_outputs=return_outputs)
-
-    def _wrap_model(self, model, training=True, dataloader=None):
-        if self.args.tensor_parallel:
-            model = tp.tensor_parallel(model, distributed=is_distributed())
-            model.hf_device_map = tp.infer_sharded_device_map(model)
-        else:
-            model = super()._wrap_model(model, training=training, dataloader=dataloader)
-        return model
-
-
-class OneCycleLRSchedulerTrainer(AxolotlTrainer):
-    """
-    Trainer subclass that uses the OneCycleLR scheduler
-    """
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        self.lr_scheduler = None
-
-    def create_scheduler(
-        self,
-        num_training_steps: int,
-        optimizer: Optional[torch.optim.Optimizer] = None,
-    ):
-        optimizer = self.optimizer if optimizer is None else optimizer
-        num_warmup_steps = self.args.get_warmup_steps(num_training_steps)
-        pct_start = num_warmup_steps / num_training_steps
-
-        self.lr_scheduler = OneCycleLR(
-            optimizer,
-            max_lr=self.args.learning_rate,
-            total_steps=num_training_steps,
-            pct_start=pct_start,
-            div_factor=6,
-        )
-
-        return self.lr_scheduler
-
-
-class ReLoRATrainer(AxolotlTrainer):
-    """
-    Trainer subclass that uses the OneCycleLR scheduler
-    """
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        self.lr_scheduler = None
-
-    def create_scheduler(
-        self,
-        num_training_steps: int,
-        optimizer: Optional[torch.optim.Optimizer] = None,
-    ):
-        optimizer = self.optimizer if optimizer is None else optimizer
-        lr_scheduler = super().create_scheduler(num_training_steps, optimizer)
-
-        if self.args.relora_steps:
-            warmup_steps = (
-                self.args.relora_warmup_steps if self.args.relora_warmup_steps else 10
-            )
-            self.lr_scheduler = ReLoRAScheduler(
-                optimizer,
-                lr_scheduler,
-                self.args.relora_steps,
-                warmup_steps,
-            )
-        else:
-            self.lr_scheduler = lr_scheduler
-
-        return self.lr_scheduler
-
-
-class TrainerBuilderBase(abc.ABC):
-    """
-    Base class for trainer builder
-    """
-
-    _train_dataset = None
-    _eval_dataset = None
-
-    def __init__(self, cfg, model, tokenizer):
-        self.cfg = cfg
-        self.model = model
-        self.tokenizer = tokenizer
-
-    @property
-    def train_dataset(self):
-        return self._train_dataset
-
-    @train_dataset.setter
-    def train_dataset(self, dataset):
-        self._train_dataset = dataset
-
-    @property
-    def eval_dataset(self):
-        return self._eval_dataset
-
-    @eval_dataset.setter
-    def eval_dataset(self, dataset):
-        self._eval_dataset = dataset
-
-    @abstractmethod
-    def build(self, total_num_steps):
-        pass
-
-    @abstractmethod
-    def get_callbacks(self):
-        pass
-
-    @abstractmethod
-    def get_post_trainer_create_callbacks(self, trainer):
-        """
-        Callbacks added after the trainer is created, usually b/c these need access to the trainer
-        """
-
-
-class HFCausalTrainerBuilder(TrainerBuilderBase):
-    """
-    Build the HuggingFace training args/trainer for Causal models
-    """
-
-    def hook_pre_create_training_args(self, training_arguments_kwargs):
-        # TODO
-        return training_arguments_kwargs
-
-    def hook_post_create_training_args(self, training_arguments):
-        # TODO
-        return training_arguments
-
-    def hook_pre_create_trainer(self, trainer_kwargs, trainer_cls):
-        # TODO
-        return trainer_kwargs, trainer_cls
-
-    def hook_post_create_trainer(self, trainer):
-        if self.cfg.tensor_parallel:
-            trainer.model = trainer.accelerator.prepare_model(
-                trainer.model, device_placement=True
-            )
-        return trainer
-
-    def get_callbacks(self):
-        callbacks = []
-        callbacks.append(GPUStatsCallback(self.cfg))
-        callbacks.append(EvalFirstStepCallback)
-
-        if self.cfg.relora_steps:
-            callbacks.append(ReLoRACallback(self.cfg))
-
-        if (
-            hasattr(self.model, "use_bettertransformer")
-            and self.model.use_bettertransformer is True
-        ):
-            callbacks.append(SaveBetterTransformerModelCallback)
-
-        if self.cfg.use_wandb:
-            callbacks.append(
-                SaveAxolotlConfigtoWandBCallback(self.cfg.axolotl_config_path)
-            )
-
-        return callbacks
-
-    def get_post_trainer_create_callbacks(self, trainer):
-        callbacks = []
-        if self.cfg.use_wandb and self.cfg.eval_table_size > 0:
-            LogPredictionCallback = log_prediction_callback_factory(
-                trainer, self.tokenizer
-            )
-            callbacks.append(LogPredictionCallback(self.cfg))
-
-        if self.cfg.do_bench_eval:
-            callbacks.append(bench_eval_callback_factory(trainer, self.tokenizer))
-
-        if self.cfg.early_stopping_patience:
-            early_stop_cb = EarlyStoppingCallback(
-                self.cfg.early_stopping_patience,
-            )
-            callbacks.append(early_stop_cb)
-
-        return callbacks
-
-    def _get_trainer_cls(self):
-        if self.cfg.lr_scheduler == "one_cycle" and (
-            self.cfg.fsdp or self.cfg.adapter == "qlora"
-        ):
-            return OneCycleLRSchedulerTrainer
-        if self.cfg.relora_steps:
-            return ReLoRATrainer
-        return AxolotlTrainer
-
-    def build(self, total_num_steps):
-        warmup_steps = (
-            self.cfg.warmup_steps
-            if self.cfg.warmup_steps is not None
-            else min(int(0.03 * total_num_steps), 100)
-        )
-        logging_steps = (
-            self.cfg.logging_steps
-            if self.cfg.logging_steps is not None
-            else max(min(int(0.005 * total_num_steps), 10), 1)
-        )
-
-        training_arguments_kwargs = {}
-        if self.cfg.bf16 == "full":
-            training_arguments_kwargs["bf16_full_eval"] = True
-        else:
-            training_arguments_kwargs["bf16"] = self.cfg.bf16
-        training_arguments_kwargs["fp16"] = (
-            self.cfg.fp16 and not self.cfg.bf16
-        ) or False
-        training_arguments_kwargs["tf32"] = self.cfg.tf32
-        training_arguments_kwargs["warmup_steps"] = warmup_steps
-        training_arguments_kwargs["logging_steps"] = logging_steps
-
-        if self.cfg.seed:
-            training_arguments_kwargs["seed"] = self.cfg.seed
-
-        if self.cfg.gradient_checkpointing:
-            training_arguments_kwargs[
-                "gradient_checkpointing"
-            ] = self.cfg.gradient_checkpointing
-        if self.cfg.fsdp:
-            training_arguments_kwargs["fsdp"] = self.cfg.fsdp
-            if self.cfg.fsdp_config:
-                training_arguments_kwargs["fsdp_config"] = dict(self.cfg.fsdp_config)
-
-        # deepspeed
-        if self.cfg.deepspeed:
-            training_arguments_kwargs["deepspeed"] = self.cfg.deepspeed
-
-        if self.cfg.lr_quadratic_warmup is not None:
-            training_arguments_kwargs[
-                "lr_quadratic_warmup"
-            ] = self.cfg.lr_quadratic_warmup
-
-        if self.cfg.adam_beta1:
-            training_arguments_kwargs["adam_beta1"] = self.cfg.adam_beta1
-        if self.cfg.adam_beta2:
-            training_arguments_kwargs["adam_beta2"] = self.cfg.adam_beta2
-        if self.cfg.adam_epsilon:
-            training_arguments_kwargs["adam_epsilon"] = self.cfg.adam_epsilon
-        if self.cfg.max_grad_norm:
-            training_arguments_kwargs["max_grad_norm"] = self.cfg.max_grad_norm
-
-        if self.cfg.hub_model_id:
-            training_arguments_kwargs["hub_model_id"] = self.cfg.hub_model_id
-            training_arguments_kwargs["push_to_hub"] = True
-            training_arguments_kwargs["hub_private_repo"] = True
-
-            if self.cfg.hub_strategy:
-                training_arguments_kwargs["hub_strategy"] = self.cfg.hub_strategy
-
-        if self.cfg.save_safetensors:
-            training_arguments_kwargs["save_safetensors"] = self.cfg.save_safetensors
-
-        if self.cfg.sample_packing_eff_est:
-            training_arguments_kwargs[
-                "sample_packing_efficiency"
-            ] = self.cfg.sample_packing_eff_est
-
-        if self.cfg.eval_steps:
-            training_arguments_kwargs["evaluation_strategy"] = "steps"
-            training_arguments_kwargs["eval_steps"] = self.cfg.eval_steps
-        elif self.cfg.evaluation_strategy:
-            training_arguments_kwargs[
-                "evaluation_strategy"
-            ] = self.cfg.evaluation_strategy
-        elif self.cfg.val_set_size == 0:
-            # no eval set, so don't eval
-            training_arguments_kwargs["evaluation_strategy"] = "no"
-        else:
-            # we have an eval set, but no steps defined, default to use epoch
-            training_arguments_kwargs["evaluation_strategy"] = "epoch"
-
-        if self.cfg.save_steps:
-            training_arguments_kwargs["save_strategy"] = "steps"
-            training_arguments_kwargs["save_steps"] = self.cfg.save_steps
-        elif self.cfg.save_strategy:
-            training_arguments_kwargs["save_strategy"] = self.cfg.save_strategy
-        else:
-            # default to saving each epoch if not defined
-            training_arguments_kwargs["save_strategy"] = "epoch"
-
-        if self.cfg.do_bench_eval:
-            training_arguments_kwargs["do_bench_eval"] = self.cfg.do_bench_eval
-            if self.cfg.bench_dataset:
-                training_arguments_kwargs["bench_dataset"] = self.cfg.bench_dataset
-        if self.cfg.metric_for_best_model:
-            training_arguments_kwargs[
-                "metric_for_best_model"
-            ] = self.cfg.metric_for_best_model
-        if self.cfg.greater_is_better:
-            training_arguments_kwargs["greater_is_better"] = self.cfg.greater_is_better
-
-        if self.cfg.torch_compile:
-            if torch.__version__ < "2.1.0":  # pylint: disable=protected-access
-                LOG.warning("torch>=2.1.0 required for torch_compile to work properly")
-            elif torch._dynamo:  # pylint: disable=protected-access
-                torch._dynamo.config.suppress_errors = (  # pylint: disable=protected-access
-                    True
-                )
-                training_arguments_kwargs["torch_compile"] = self.cfg.torch_compile
-                if self.cfg.torch_compile_backend:
-                    training_arguments_kwargs[
-                        "torch_compile_backend"
-                    ] = self.cfg.torch_compile_backend
-
-        # DDP Config
-        if self.cfg.ddp_timeout:
-            training_arguments_kwargs["ddp_timeout"] = self.cfg.ddp_timeout
-        # see https://pytorch.org/docs/stable/generated/torch.nn.parallel.DistributedDataParallel.html
-        if self.cfg.ddp_bucket_cap_mb:
-            training_arguments_kwargs["ddp_bucket_cap_mb"] = self.cfg.ddp_bucket_cap_mb
-        if self.cfg.ddp_broadcast_buffers is not None:
-            training_arguments_kwargs[
-                "ddp_broadcast_buffers"
-            ] = self.cfg.ddp_broadcast_buffers
-
-        # these are all the "standard" kwargs that are def used
-        training_arguments_kwargs["max_steps"] = (
-            total_num_steps if self.cfg.max_steps else -1
-        )
-        training_arguments_kwargs["max_seq_length"] = self.cfg.sequence_len
-        training_arguments_kwargs[
-            "per_device_train_batch_size"
-        ] = self.cfg.micro_batch_size
-        training_arguments_kwargs[
-            "per_device_eval_batch_size"
-        ] = self.cfg.eval_batch_size
-        training_arguments_kwargs[
-            "gradient_accumulation_steps"
-        ] = self.cfg.gradient_accumulation_steps
-        training_arguments_kwargs[
-            "eval_accumulation_steps"
-        ] = self.cfg.gradient_accumulation_steps
-        training_arguments_kwargs["num_train_epochs"] = self.cfg.num_epochs
-        training_arguments_kwargs["learning_rate"] = self.cfg.learning_rate
-        training_arguments_kwargs["output_dir"] = self.cfg.output_dir
-        training_arguments_kwargs["save_total_limit"] = (
-            self.cfg.save_total_limit if self.cfg.save_total_limit else 4
-        )
-        training_arguments_kwargs["load_best_model_at_end"] = (
-            (
-                self.cfg.load_best_model_at_end is not False
-                or self.cfg.early_stopping_patience
-            )
-            and self.cfg.val_set_size > 0
-            and self.cfg.save_steps
-            and self.cfg.eval_steps
-            and self.cfg.save_steps % self.cfg.eval_steps == 0
-        ) or False
-        training_arguments_kwargs["ddp_find_unused_parameters"] = (
-            False if self.cfg.ddp else None
-        )
-        training_arguments_kwargs["group_by_length"] = self.cfg.group_by_length
-        training_arguments_kwargs["report_to"] = "wandb" if self.cfg.use_wandb else None
-        training_arguments_kwargs["run_name"] = (
-            self.cfg.wandb_run_id if self.cfg.use_wandb else None
-        )
-        training_arguments_kwargs["optim"] = (
-            self.cfg.optimizer if self.cfg.optimizer else "adamw_hf"
-        )
-        training_arguments_kwargs["lr_scheduler_type"] = (
-            self.cfg.lr_scheduler
-            if self.cfg.lr_scheduler
-            and self.cfg.lr_scheduler not in ("one_cycle", "log_sweep")
-            else "cosine"
-        )
-        training_arguments_kwargs["weight_decay"] = (
-            self.cfg.weight_decay if self.cfg.weight_decay is not None else 0.0
-        )
-        training_arguments_kwargs["sample_packing"] = (
-            self.cfg.sample_packing if self.cfg.sample_packing else False
-        )
-        training_arguments_kwargs["eval_sample_packing"] = (
-            self.cfg.sample_packing if self.cfg.sample_packing else False
-        )
-        training_arguments_kwargs[
-            "sample_packing_seq_len_multiplier"
-        ] = self.cfg.micro_batch_size
-        training_arguments_kwargs["relora_steps"] = self.cfg.relora_steps
-        training_arguments_kwargs["relora_warmup_steps"] = self.cfg.relora_warmup_steps
-        training_arguments_kwargs["tensor_parallel"] = self.cfg.tensor_parallel is True
-
-        training_arguments_kwargs = self.hook_pre_create_training_args(
-            training_arguments_kwargs
-        )
-        training_args = (
-            AxolotlTrainingArguments(  # pylint: disable=unexpected-keyword-arg
-                **training_arguments_kwargs,
-            )
-        )
-        training_args = self.hook_post_create_training_args(training_args)
-        trainer_kwargs = {}
-
-        if self.cfg.optimizer == "adamw_anyprecision":
-            if Path(self.cfg.torchdistx_path).exists():
-                sys.path.append(self.cfg.torchdistx_path)
-                importlib.import_module("torchdistx")
-
-        data_collator_kwargs = {
-            "padding": True,  # True/"longest" is the default
-        }
-        if self.cfg.pad_to_sequence_len:
-            data_collator_kwargs["pad_to_multiple_of"] = 64 * math.ceil(
-                self.cfg.sequence_len / 64
-            )
-        else:
-            # A100 is best at 64, while others at 8. Let's use the larger so we don't have to check
-            # https://docs.nvidia.com/deeplearning/performance/dl-performance-matrix-multiplication/index.html
-            data_collator_kwargs["pad_to_multiple_of"] = 64
-
-        if self.cfg.is_llama_derived_model and self.cfg.landmark_attention:
-            from axolotl.monkeypatch.llama_landmark_attn import (
-                add_mem_tokens,
-                get_mem_id,
-                set_model_mem_id,
-            )
-
-            set_model_mem_id(self.model, self.tokenizer)
-
-            LOG.info("Adding landmark attention tokens to dataset")
-
-            for dataset in [self.train_dataset, self.eval_dataset]:
-                dataset = dataset.map(
-                    partial(
-                        add_mem_tokens, mem_freq=50, mem_id=get_mem_id(self.tokenizer)
-                    ),
-                    batched=False,
-                    num_proc=32,
-                )
-
-        trainer_cls = self._get_trainer_cls()
-        trainer_kwargs, trainer_cls = self.hook_pre_create_trainer(
-            trainer_kwargs, trainer_cls
-        )
-        trainer = trainer_cls(
-            model=self.model,
-            train_dataset=self.train_dataset,
-            eval_dataset=self.eval_dataset,
-            args=training_args,
-            data_collator=DataCollatorForSeq2Seq(
-                self.tokenizer,
-                return_tensors="pt",
-                **data_collator_kwargs,
-            ),
-            bench_data_collator=transformers.DataCollatorForSeq2Seq(
-                self.tokenizer,
-                return_tensors="pt",
-                **data_collator_kwargs,
-            ),
-            callbacks=self.get_callbacks(),
-            num_epochs=self.cfg.num_epochs,
-            **trainer_kwargs,
-        )
-        trainer = self.hook_post_create_trainer(trainer)
-        for callback in self.get_post_trainer_create_callbacks(trainer):
-            trainer.add_callback(callback)
-
-        return trainer

src/axolotl/datasets.py

@@ -22,7 +22,7 @@ class TokenizedPromptDataset(Dataset):
     """
     Dataset that returns tokenized prompts from a stream of text files.
         Args:
-            prompt_tokenizer (PromptTokenizingStrategy): The prompt tokenizing method for processing the data.
+            prompt_tokenizer (PromptTokenizingStrategy): The prompt tokenizing method for proccessing the data.
             dataset (dataset.Dataset): Dataset with text files.
     """
 
@@ -38,15 +38,10 @@ class TokenizedPromptDataset(Dataset):
     def process(self, dataset):
         features = dataset.features.keys()
         num_proc = min(64, os.cpu_count())
-        map_kwargs = {}
-        if self.prompt_tokenizer.supports_batched:
-            map_kwargs["batched"] = True
-            map_kwargs["batch_size"] = 100
         return dataset.map(
             self.prompt_tokenizer.tokenize_prompt,
             num_proc=num_proc,
             remove_columns=features,
-            **map_kwargs,
         )
 
 
@@ -55,7 +50,7 @@ class ConstantLengthDataset(IterableDataset):
     """
     Iterable dataset that returns constant length chunks of tokens from stream of text files.
         Args:
-            tokenizer (Tokenizer): The processor used for processing the data.
+            tokenizer (Tokenizer): The processor used for proccessing the data.
             dataset (dataset.Dataset): Dataset with text files.
             seq_length (int): Length of token sequences to return.
     """

src/axolotl/logging_config.py

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

src/axolotl/models/phi/__init__.py

@@ -1,6 +0,0 @@
-"""
-MixFormers model architecture used for phi models
-"""
-
-from .configuration_mixformer_sequential import MixFormerSequentialConfig  # noqa
-from .modeling_mixformer_sequential import MixFormerSequentialForCausalLM  # noqa

src/axolotl/models/phi/configuration_mixformer_sequential.py

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

src/axolotl/models/phi/modeling_mixformer_sequential.py

@@ -1,930 +0,0 @@
-# pylint: skip-file
-
-# Copyright (c) Microsoft Corporation.
-# Licensed under the MIT license.
-
-# BSD 3-Clause License
-#
-# Copyright (c) 2022, Tri Dao, trid@cs.stanford.edu.
-# All rights reserved.
-#
-# Redistribution and use in source and binary forms, with or without
-# modification, are permitted provided that the following conditions are met:
-#
-# * Redistributions of source code must retain the above copyright notice, this
-#   list of conditions and the following disclaimer.
-#
-# * Redistributions in binary form must reproduce the above copyright notice,
-#   this list of conditions and the following disclaimer in the documentation
-#   and/or other materials provided with the distribution.
-#
-# * Neither the name of the copyright holder nor the names of its
-#   contributors may be used to endorse or promote products derived from
-#   this software without specific prior written permission.
-#
-# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
-# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
-# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
-# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
-# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
-# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
-# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
-# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
-# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-from __future__ import annotations
-
-import copy
-import inspect
-from dataclasses import dataclass, field
-from typing import Any, Dict, Optional, Tuple
-
-import torch
-import torch.nn as nn
-from einops import rearrange
-from flash_attn.flash_attn_interface import (
-    flash_attn_kvpacked_func,
-    flash_attn_qkvpacked_func,
-    flash_attn_varlen_qkvpacked_func,
-)
-from transformers import PretrainedConfig, PreTrainedModel
-from transformers.activations import ACT2FN
-from transformers.modeling_outputs import CausalLMOutputWithPast
-
-from ...monkeypatch.utils import get_cu_seqlens_from_pos_ids
-from .configuration_mixformer_sequential import MixFormerSequentialConfig
-
-
-@dataclass
-class InferenceParams:
-    """Inference parameters that are passed to the main model in order
-    to efficienly calculate and store the context during inference.
-    Adapted from https://github.com/Dao-AILab/flash-attention."""
-
-    max_sequence_len: int
-    max_batch_size: int
-    sequence_len_offset: int = 0
-    batch_size_offset: int = 0
-    key_value_memory_dict: dict = field(default_factory=dict)
-    fused_ft_kernel: bool = False
-    lengths_per_sample: Optional[torch.Tensor] = None
-
-
-class Embedding(nn.Module):
-    """Token embedding with dropout."""
-
-    def __init__(self, config: PretrainedConfig) -> None:
-        super().__init__()
-
-        self.wte = nn.Embedding(config.vocab_size, config.n_embd)
-        self.drop = nn.Dropout(config.embd_pdrop)
-
-    def forward(self, input_ids: torch.LongTensor) -> torch.FloatTensor:
-        input_shape = input_ids.size()
-        input_ids = input_ids.view(-1, input_shape[-1])
-
-        hidden_states = self.wte(input_ids)
-        hidden_states = self.drop(hidden_states)
-
-        return hidden_states
-
-
-class RotaryEmbedding(nn.Module):
-    """PyTorch implementation of `flash-attn` RotaryEmbedding layer.
-    Adapted from https://github.com/Dao-AILab/flash-attention."""
-
-    def __init__(
-        self,
-        dim: int,
-        base: Optional[int] = 10000,
-        scale_base: Optional[float] = None,
-        device: Optional[str] = None,
-        **kwargs,
-    ) -> None:
-        super().__init__()
-
-        if scale_base is not None:
-            raise NotImplementedError
-
-        # Generate and save the inverse frequency buffer (non-trainable)
-        self.dim = dim
-        self.base = base
-        self.scale_base = scale_base
-        self.device = device
-
-        inv_freq = 1.0 / (
-            base ** (torch.arange(0, dim, 2, device=device, dtype=torch.float32) / dim)
-        )
-        self.register_buffer("inv_freq", inv_freq)
-
-        scale = (
-            (torch.arange(0, dim, 2, device=device, dtype=torch.float32) + 0.4 * dim)
-            / (1.4 * dim)
-            if scale_base is not None
-            else None
-        )
-        self.register_buffer("scale", scale)
-
-        self._seq_len_cached = 0
-        self._cos_cached = None
-        self._sin_cached = None
-        self._cos_k_cached = None
-        self._sin_k_cached = None
-
-    def _update_cos_sin_cache(
-        self, x: torch.FloatTensor, seqlen_offset: Optional[int] = 0
-    ) -> None:
-        # Reset the tables if the sequence length has changed,
-        # or if we're on a new device (possibly due to tracing for instance)
-        seqlen = x.shape[1] + seqlen_offset
-
-        # Re-generate the inverse frequency buffer if it's not fp32
-        # (for instance if model.half() was called)
-        if self.inv_freq.dtype != "torch.float32":
-            self.inv_freq = 1.0 / (
-                self.base
-                ** (
-                    torch.arange(
-                        0, self.dim, 2, device=self.device, dtype=torch.float32
-                    )
-                    / self.dim
-                )
-            )
-
-        if (
-            seqlen > self._seq_len_cached
-            or self._cos_cached.device != x.device
-            or self._cos_cached.dtype != x.dtype
-        ):
-            self._seq_len_cached = seqlen
-            t = torch.arange(seqlen, device=x.device, dtype=torch.float32)
-
-            # Don't do einsum, it converts fp32 to fp16
-            # freqs = torch.einsum("i,j->ij", t, self.inv_freq)
-            freqs = torch.outer(
-                t, self.inv_freq.to(device=t.device, dtype=torch.float32)
-            )
-            if self.scale is None:
-                self._cos_cached = torch.cos(freqs).to(x.dtype)
-                self._sin_cached = torch.sin(freqs).to(x.dtype)
-            else:
-                power = (
-                    torch.arange(
-                        seqlen, dtype=self.scale.dtype, device=self.scale.device
-                    )
-                    - seqlen // 2
-                ) / self.scale_base
-                scale = self.scale.to(device=power.device) ** rearrange(
-                    power, "s -> s 1"
-                )
-
-                # We want the multiplication by scale to happen in fp32
-                self._cos_cached = (torch.cos(freqs) * scale).to(x.dtype)
-                self._sin_cached = (torch.sin(freqs) * scale).to(x.dtype)
-                self._cos_k_cached = (torch.cos(freqs) / scale).to(x.dtype)
-                self._sin_k_cached = (torch.sin(freqs) / scale).to(x.dtype)
-
-    def apply_rotary_emb_qkv(
-        self,
-        qkv: torch.FloatTensor,
-        sin: torch.FloatTensor,
-        cos: torch.FloatTensor,
-        sin_k: Optional[torch.FloatTensor] = None,
-        cos_k: Optional[torch.FloatTensor] = None,
-    ) -> torch.FloatTensor:
-        _, seqlen, three, _, headdim = qkv.shape
-        assert three == 3
-
-        rotary_seqlen, rotary_dim = cos.shape
-        rotary_dim *= 2
-        assert rotary_dim <= headdim
-        assert seqlen <= rotary_seqlen
-
-        cos_k = cos if cos_k is None else cos_k
-        sin_k = sin if sin_k is None else sin_k
-        assert (
-            sin.shape == cos_k.shape == sin_k.shape == (rotary_seqlen, rotary_dim // 2)
-        )
-
-        q_rot = qkv[:, :, 0, :, :rotary_dim]
-        q_pass = qkv[:, :, 0, :, rotary_dim:]
-
-        k_rot = qkv[:, :, 1, :, :rotary_dim]
-        k_pass = qkv[:, :, 1, :, rotary_dim:]
-
-        # Splits the queries and keys in half
-        q1, q2 = q_rot.chunk(2, dim=-1)
-        k1, k2 = k_rot.chunk(2, dim=-1)
-        c, s = rearrange(cos[:seqlen], "s d -> s 1 d"), rearrange(
-            sin[:seqlen], "s d -> s 1 d"
-        )
-
-        # Casts to fp32 are necessary to prevent fp16 overflow issues
-        q1, q2, k1, k2, c, s = [
-            t.to(dtype=torch.float32) for t in [q1, q2, k1, k2, c, s]
-        ]
-
-        # Computes the new keys and queries, recasting to original dtype
-        q_rot = torch.cat([q1 * c - q2 * s, q1 * s + q2 * c], axis=-1).to(qkv.dtype)
-
-        k_rot = torch.cat([k1 * c - k2 * s, k1 * s + k2 * c], axis=-1).to(qkv.dtype)
-
-        return torch.cat(
-            [
-                torch.cat([q_rot, q_pass], axis=-1).unsqueeze(2),
-                torch.cat([k_rot, k_pass], axis=-1).unsqueeze(2),
-                qkv[:, :, 2:3, :, :],
-            ],
-            axis=2,
-        )
-
-    def forward(
-        self, qkv: torch.Tensor, seqlen_offset: int = 0
-    ) -> Tuple[torch.Tensor, torch.Tensor]:
-        """Perform the forward pass.
-
-        Args:
-            qkv: Query, key and value tensors of shape (batch, seqlen, nheads, headdim) or (batch, seqlen, 3, nheads, headdim).
-            seqlen_offset: Used in generation where the passed `qkv` is only the last token in the batch.
-
-        Returns:
-            New `qkv` and the cached sinusoids.
-
-        """
-
-        self._update_cos_sin_cache(qkv, seqlen_offset)
-
-        return self.apply_rotary_emb_qkv(
-            qkv, self._sin_cached[seqlen_offset:], self._cos_cached[seqlen_offset:]
-        )
-
-
-def _update_kv_cache(kv, inference_params, layer_idx):
-    """kv: (batch_size, seqlen, 2, nheads, head_dim) or (batch_size, 1, 2, nheads, head_dim)
-    Adapted from https://github.com/Dao-AILab/flash-attention."""
-    # Pre-allocate memory for key-values for inference.
-    num_heads, head_dim = kv.shape[-2:]
-    if layer_idx not in inference_params.key_value_memory_dict:
-        kv_cache = torch.empty(
-            inference_params.max_batch_size,
-            inference_params.max_sequence_len,
-            2,
-            num_heads,
-            head_dim,
-            dtype=kv.dtype,
-            device=kv.device,
-        )
-        inference_params.key_value_memory_dict[layer_idx] = kv_cache
-    else:
-        kv_cache = inference_params.key_value_memory_dict[layer_idx]
-
-    # Adjust key and value for inference
-    batch_start = inference_params.batch_size_offset
-    batch_end = batch_start + kv.shape[0]
-    sequence_start = inference_params.sequence_len_offset
-    sequence_end = sequence_start + kv.shape[1]
-    assert batch_end <= (
-        kv_cache.shape[0] if kv_cache is not None else v_cache.shape[0]  # noqa
-    )
-    assert sequence_end <= (
-        kv_cache.shape[1] if kv_cache is not None else v_cache.shape[2]  # noqa
-    )
-
-    assert kv_cache is not None
-    kv_cache[batch_start:batch_end, sequence_start:sequence_end, ...] = kv
-    kv = kv_cache[batch_start:batch_end, :sequence_end, ...]
-    return kv
-
-
-class MLP(nn.Module):
-    """Multi-Layer Perceptron.
-
-    Reference:
-        Attention Is All You Need.
-        https://arxiv.org/pdf/1706.03762.pdf.
-
-    """
-
-    def __init__(
-        self,
-        config: PretrainedConfig,
-        n_inner: Optional[int] = None,
-        act_fn: Optional[str] = None,
-    ) -> None:
-        super().__init__()
-
-        act_fn = config.activation_function if act_fn is None else act_fn
-        assert act_fn in ACT2FN.keys(), f"`act_fn` must be one of: {ACT2FN.keys()}."
-
-        n_inner = getattr(config, "n_inner", None) if n_inner is None else n_inner
-        n_inner = n_inner if n_inner is not None else 4 * config.n_embd
-
-        self.fc1 = nn.Linear(config.n_embd, n_inner)
-        self.fc2 = nn.Linear(n_inner, config.n_embd)
-        self.act = ACT2FN[act_fn]
-
-    def _load_from_state_dict(
-        self,
-        state_dict,
-        prefix,
-        local_metadata,
-        strict,
-        missing_keys,
-        unexpected_keys,
-        error_msgs,
-    ):
-        old_keys = [
-            prefix + "fc_in.weight",
-            prefix + "fc_out.weight",
-            prefix + "fc_in.bias",
-            prefix + "fc_out.bias",
-        ]
-        new_keys = [
-            prefix + "fc1.weight",
-            prefix + "fc2.weight",
-            prefix + "fc1.bias",
-            prefix + "fc2.bias",
-        ]
-
-        if all(k in state_dict for k in old_keys) and not all(
-            k in state_dict for k in new_keys
-        ):
-            # Older version of `MLP` saved with different key names.
-            for old_key, new_key in zip(old_keys, new_keys):
-                state_dict[new_key] = state_dict.pop(old_key)
-
-        return super()._load_from_state_dict(
-            state_dict,
-            prefix,
-            local_metadata,
-            strict,
-            missing_keys,
-            unexpected_keys,
-            error_msgs,
-        )
-
-    def forward(self, hidden_states: torch.FloatTensor) -> torch.FloatTensor:
-        hidden_states = self.fc1(hidden_states)
-        hidden_states = self.act(hidden_states)
-        hidden_states = self.fc2(hidden_states)
-
-        return hidden_states
-
-
-class FusedMLP(nn.Module):
-    """Fused Multi-Layer Perceptron from `flash-attn`.
-
-    Reference:
-        https://github.com/HazyResearch/flash-attention/blob/main/flash_attn/ops/fused_dense.py.
-
-    """
-
-    def __init__(
-        self,
-        config: PretrainedConfig,
-        n_inner: Optional[int] = None,
-        act_fn: Optional[str] = None,
-        raise_on_missing: bool = False,
-    ) -> None:
-        super().__init__()
-
-        act_fn = config.activation_function if act_fn is None else act_fn
-        assert act_fn in ACT2FN.keys(), f"`act_fn` must be one of: {ACT2FN.keys()}."
-
-        n_inner = getattr(config, "n_inner", None) if n_inner is None else n_inner
-        n_inner = n_inner if n_inner is not None else 4 * config.n_embd
-
-        gelu_activations = ["gelu_new", "gelu_fast", "gelu_approx"]  # noqa
-        activation = "gelu_approx" if act_fn in gelu_activations else "relu"  # noqa
-
-        self.mlp = MLP(config, n_inner=n_inner, act_fn=act_fn)
-
-    def forward(self, hidden_states: torch.FloatTensor) -> torch.FloatTensor:
-        return self.mlp(hidden_states)
-
-
-class SelfAttention(nn.Module):
-    """Implement the scaled dot product attention with softmax.
-    Adapted from https://github.com/Dao-AILab/flash-attention.
-    Arguments
-    ---------
-        softmax_scale: The temperature to use for the softmax attention.
-                      (default: 1/sqrt(d_keys) where d_keys is computed at
-                      runtime)
-        attention_dropout: The dropout rate to apply to the attention
-                           (default: 0.0)
-    """
-
-    def __init__(self, causal=False, softmax_scale=None, attention_dropout=0.0):
-        super().__init__()
-        self.causal = causal
-        self.softmax_scale = softmax_scale
-        self.drop = nn.Dropout(attention_dropout)
-
-    def forward(
-        self, qkv, causal=None, key_padding_mask=None, cu_seqlens=None, max_seqlen=None
-    ):
-        """Implements the multihead softmax attention.
-        Arguments
-        ---------
-            qkv: The tensor containing the query, key, and value. (B, S, 3, H, D)
-            causal: if passed, will override self.causal
-            key_padding_mask: boolean mask to apply to the attention weights. True means to keep,
-                False means to mask out. (B, S)
-        """
-        causal = self.causal if causal is None else causal
-        if cu_seqlens is not None:
-            return flash_attn_varlen_qkvpacked_func(
-                qkv.squeeze(0),
-                cu_seqlens,
-                max_seqlen,
-                dropout_p=self.drop.p,
-                softmax_scale=self.softmax_scale,
-                causal=causal,
-            )
-        else:
-            return flash_attn_qkvpacked_func(
-                qkv,
-                dropout_p=self.drop.p,
-                softmax_scale=self.softmax_scale,
-                causal=causal,
-            )
-
-
-class CrossAttention(nn.Module):
-    """Implement the scaled dot product attention with softmax.
-    Adapted from https://github.com/Dao-AILab/flash-attention.
-    Arguments
-    ---------
-        softmax_scale: The temperature to use for the softmax attention.
-                      (default: 1/sqrt(d_keys) where d_keys is computed at
-                      runtime)
-        attention_dropout: The dropout rate to apply to the attention
-                           (default: 0.0)
-    """
-
-    def __init__(self, causal=False, softmax_scale=None, attention_dropout=0.0):
-        super().__init__()
-        self.causal = causal
-        self.softmax_scale = softmax_scale
-        self.drop = nn.Dropout(attention_dropout)
-
-    def forward(self, q, kv, causal=None, key_padding_mask=None):
-        """Implements the multihead softmax attention.
-        Arguments
-        ---------
-            q: The tensor containing the query. (B, Sq, H, D)
-            kv: The tensor containing the key and value. (B, Sk, 2, H, D)
-            causal: if passed, will override self.causal
-            key_padding_mask: boolean mask to apply to the attention weights. True means to keep,
-                False means to mask out. (B, Sk)
-        """
-        causal = self.causal if causal is None else causal
-        return flash_attn_kvpacked_func(
-            q,
-            kv,
-            dropout_p=self.drop.p,
-            softmax_scale=self.softmax_scale,
-            causal=causal,
-        )
-
-
-def find_mha_dims(
-    config: PretrainedConfig,
-    n_head: Optional[int] = None,
-    head_dim: Optional[int] = None,
-) -> Tuple[int, int]:
-    """Validate and return the number of heads and head dimension for multi-head attention.
-
-    Args:
-        config: Model configuration.
-        n_head: Number of heads.
-        head_dim: Head dimension.
-
-    Returns:
-        Number of heads and head dimension.
-
-    """
-
-    assert all(
-        hasattr(config, attr) for attr in ["n_embd", "n_head"]
-    ), "`config` must have `n_embd` and `n_head` attributes."
-
-    if head_dim is None:
-        assert (
-            config.n_embd % config.n_head == 0
-        ), f"Hidden size ({config.n_embd}) must be divisible by the number of heads ({config.n_head})."
-
-    if n_head is None and head_dim is None:
-        head_dim = config.n_embd // config.n_head
-        n_head = config.n_head
-    elif n_head is None or head_dim is None:
-        raise ValueError("`n_head` and `head_dim` must be both specified or `None`.")
-
-    return n_head, head_dim
-
-
-class MHA(nn.Module):
-    """Multi-head attention layer.
-    Adapted from https://github.com/Dao-AILab/flash-attention."""
-
-    def __init__(
-        self,
-        config: PretrainedConfig,
-        rotary_dim: Optional[int] = None,
-        n_head: Optional[int] = None,
-        head_dim: Optional[int] = None,
-        bias: Optional[bool] = True,
-        dropout: Optional[float] = 0.0,
-        softmax_scale: Optional[float] = None,
-        causal: Optional[bool] = True,
-        layer_idx: Optional[int] = None,
-        rotary_emb_scale_base: Optional[float] = None,
-        return_residual: Optional[bool] = False,
-        checkpointing: Optional[bool] = False,
-        device: Optional[str] = None,
-        dtype: Optional[torch.dtype] = None,
-        fused_dense: Optional[bool] = True,
-        flash_attn: Optional[bool] = True,
-        cutlass_attn: Optional[bool] = False,
-        flash_rotary: Optional[bool] = True,
-        raise_on_missing: Optional[bool] = False,
-    ) -> None:
-        super().__init__()
-
-        factory_kwargs = {"device": device, "dtype": dtype}
-        n_head, head_dim = find_mha_dims(config, n_head, head_dim)
-
-        self.hidden_size = config.n_embd
-        self.n_head = n_head
-        self.head_dim = head_dim
-        self.op_size = n_head * head_dim
-
-        self.causal = causal
-        self.layer_idx = layer_idx
-        self.rotary_emb_dim = (
-            rotary_dim if rotary_dim is not None else getattr(config, "rotary_dim", 0)
-        )
-        self.fused_dense = fused_dense
-        self.flash_attn = flash_attn
-        self.cutlass_attn = cutlass_attn
-        self.flash_rotary = flash_rotary
-        self.return_residual = return_residual
-        self.checkpointing = checkpointing
-
-        if self.rotary_emb_dim > 0:
-            rotary_kwargs = {"device": device}
-            if rotary_emb_scale_base is not None and rotary_emb_scale_base > 0.0:
-                rotary_kwargs["scale_base"] = rotary_emb_scale_base
-
-            self.rotary_emb = RotaryEmbedding(self.rotary_emb_dim, **rotary_kwargs)
-        else:
-            pass
-
-        self.Wqkv = nn.Linear(
-            self.hidden_size, 3 * self.op_size, bias=bias, **factory_kwargs
-        )
-        self.out_proj = nn.Linear(
-            self.op_size, self.hidden_size, bias=bias, **factory_kwargs
-        )
-
-        self.inner_attn = SelfAttention(
-            causal=causal, softmax_scale=softmax_scale, attention_dropout=dropout
-        )
-        self.inner_cross_attn = CrossAttention(
-            causal=causal, softmax_scale=softmax_scale, attention_dropout=dropout
-        )
-
-    def _update_kv_cache(
-        self, kv: torch.FloatTensor, inference_params: InferenceParams
-    ) -> None:
-        """kv: (batch_size, seqlen, 2, nheads, head_dim) or (batch_size, 1, 2, nheads, head_dim)
-        Adapted from https://github.com/Dao-AILab/flash-attention."""
-
-        assert (
-            self.layer_idx is not None
-        ), "Generation requires layer_idx in the constructor"
-
-        return _update_kv_cache(kv, inference_params, self.layer_idx)
-
-    def forward(
-        self,
-        x: torch.FloatTensor,
-        x_kv: Optional[torch.FloatTensor] = None,
-        key_padding_mask: Optional[torch.BoolTensor] = None,
-        cu_seqlens: Optional[torch.LongTensor] = None,
-        max_seqlen: Optional[int] = None,
-        mixer_subset: Optional[torch.LongTensor] = None,
-        past_cache: Optional[InferenceParams] = None,
-        **kwargs,
-    ) -> Tuple[torch.FloatTensor, torch.FloatTensor]:
-        """Perform the forward pass.
-
-        Args:
-            x: (batch, seqlen, hidden_dim) (where hidden_dim = num heads * head dim) if
-                cu_seqlens is None and max_seqlen is None, else (total, hidden_dim) where total
-                is the is the sum of the sequence lengths in the batch.
-            x_kv: (batch, seqlen, hidden_dim), only applicable for cross-attention. If None, use x.
-            key_padding_mask: boolean mask, True means to keep, False means to mask out.
-                (batch, seqlen). Only applicable when not using FlashAttention.
-            cu_seqlens: (batch_size + 1,), dtype torch.int32. The cumulative sequence lengths
-                of the sequences in the batch, used to index into x. Only applicable when using
-                FlashAttention.
-            max_seqlen: int. Maximum sequence length in the batch.
-            mixer_subset: for cross-attention only. If not None, will take a subset of x
-                before applying the query projection. Useful for e.g., ViT where we only care
-                about the CLS token in the last layer.
-            past_cache: For generation only.
-
-        Returns:
-            (batch, seqlen, hidden_dim) if cu_seqlens is None and max_seqlen is None,
-                else (total, hidden_dim) where total is the is the sum of the sequence lengths
-                in the batch.
-
-        """
-
-        if cu_seqlens is not None:
-            assert max_seqlen is not None
-            assert key_padding_mask is None
-            assert self.flash_attn
-            # assert self.rotary_emb_dim == 0
-
-        if key_padding_mask is not None:
-            assert cu_seqlens is None
-            assert max_seqlen is None
-            assert not self.flash_attn
-
-        if past_cache is not None:
-            assert key_padding_mask is None
-            assert cu_seqlens is None and max_seqlen is None
-
-        attn_kwargs = {"key_padding_mask": key_padding_mask}
-
-        assert x_kv is None and mixer_subset is None
-
-        qkv = self.Wqkv(x)
-        qkv = rearrange(
-            qkv, "... (three h d) -> ... three h d", three=3, d=self.head_dim
-        )
-
-        if past_cache is None:
-            if self.rotary_emb_dim > 0:
-                qkv = self.rotary_emb(qkv)
-                context = self.inner_attn(
-                    qkv, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen, **attn_kwargs
-                )
-
-        else:
-            if self.rotary_emb_dim > 0:
-                qkv = self.rotary_emb(qkv, seqlen_offset=past_cache.sequence_len_offset)
-            q = qkv[:, :, 0]
-            kv = self._update_kv_cache(qkv[:, :, 1:], past_cache)
-            # If we're processing the prompt, causal=None (use self.causal).
-            # If we're decoding, then causal=False.
-            causal = None if past_cache.sequence_len_offset == 0 else False
-            context = self.inner_cross_attn(q, kv, causal=causal)
-
-        out = rearrange(context, "... h d -> ... (h d)")
-        out = self.out_proj(out)
-
-        return out if not self.return_residual else (out, x)
-
-
-class ParallelBlock(nn.Module):
-    """Parallel block.
-
-    This block applies parallel mixer and MLP layers to the input (used in GPT-J and CodeGen).
-
-    """
-
-    def __init__(
-        self,
-        config: PretrainedConfig,
-        mixer: Optional[Dict[str, Any]] = None,
-        mlp: Optional[Dict[str, Any]] = None,
-        block_idx: Optional[int] = None,
-    ) -> None:
-        super().__init__()
-
-        self.ln = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
-        self.resid_dropout = nn.Dropout(config.resid_pdrop)
-        self.block_idx = block_idx
-
-        self.mixer = MHA(config, layer_idx=block_idx)
-        self.mlp = MLP(config)
-
-    def forward(
-        self,
-        hidden_states: torch.FloatTensor,
-        past_cache: Optional[torch.FloatTensor] = None,
-        cu_seqlens: Optional[torch.LongTensor] = None,
-        max_seqlen: Optional[int] = None,
-    ) -> torch.FloatTensor:
-        residual = hidden_states
-        hidden_states = self.ln(hidden_states)
-
-        attn_outputs = self.mixer(
-            hidden_states,
-            past_cache=past_cache,
-            cu_seqlens=cu_seqlens,
-            max_seqlen=max_seqlen,
-        )
-        if isinstance(attn_outputs, tuple):
-            attn_outputs = attn_outputs[0]
-
-        attn_outputs = self.resid_dropout(attn_outputs)
-        feed_forward_hidden_states = self.resid_dropout(self.mlp(hidden_states))
-
-        hidden_states = attn_outputs + feed_forward_hidden_states + residual
-
-        return hidden_states
-
-
-class CausalLMHead(nn.Module):
-    """Causal Language Modeling head.
-
-    Reference:
-        Improving Language Understanding by Generative Pre-Training.
-        https://cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf.
-
-    """
-
-    def __init__(self, config: PretrainedConfig) -> None:
-        super().__init__()
-
-        self.ln = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
-        self.linear = nn.Linear(config.n_embd, config.vocab_size)
-
-    def forward(self, hidden_states: torch.FloatTensor) -> torch.FloatTensor:
-        hidden_states = self.ln(hidden_states)
-        logits = self.linear(hidden_states).to(torch.float32)
-
-        return logits
-
-
-class CausalLMLoss(nn.Module):
-    """Causal Language Modeling loss.
-
-    Reference:
-        Improving Language Understanding by Generative Pre-Training.
-        https://cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf.
-
-    """
-
-    def __init__(self, shift_labels: Optional[bool] = True) -> None:
-        super().__init__()
-
-        self.shift_labels = shift_labels
-        self.loss_fct = nn.CrossEntropyLoss()
-
-    def forward(
-        self, logits: torch.FloatTensor, labels: torch.LongTensor
-    ) -> torch.FloatTensor:
-        if self.shift_labels:
-            logits = logits[..., :-1, :].contiguous()
-            labels = labels[..., 1:].contiguous()
-
-        loss = self.loss_fct(logits.view(-1, logits.size(-1)), labels.view(-1))
-
-        return loss
-
-
-class MixFormerSequentialPreTrainedModel(PreTrainedModel):
-    """MixFormer (sequential for DeepSpeed) pre-trained model."""
-
-    config_class = MixFormerSequentialConfig
-    base_model_prefix = "transformer"
-    supports_gradient_checkpointing = True
-
-    def __init__(self, *inputs, **kwargs) -> None:
-        super().__init__(*inputs, **kwargs)
-
-    def prepare_inputs_for_generation(
-        self, input_ids, past_key_values=None, **kwargs
-    ) -> Dict[str, Any]:
-        if "use_cache" in kwargs and not kwargs["use_cache"]:
-            return {"input_ids": input_ids}
-
-        if past_key_values is None or not (
-            isinstance(past_key_values, InferenceParams)
-        ):
-            past_key_values = InferenceParams(
-                max_batch_size=input_ids.shape[0],
-                max_sequence_len=self.config.n_positions,
-                sequence_len_offset=0,
-                batch_size_offset=0,
-                fused_ft_kernel=False,
-                key_value_memory_dict={},
-            )
-        else:
-            # assume past_key_values has cached all but last token in input_ids
-            past_key_values.sequence_len_offset = len(input_ids[0]) - 1
-            input_ids = input_ids[:, -1].unsqueeze(-1)
-
-        return {"input_ids": input_ids, "past_key_values": past_key_values, **kwargs}
-
-
-class PackedSequential(nn.Sequential):
-    def forward(
-        self,
-        input,
-        cu_seqlens: Optional[torch.LongTensor] = None,
-        max_seqlen: Optional[int] = None,
-    ):
-        for module in self:
-            sig = inspect.signature(module.forward)
-            if "cu_seqlens" in sig.parameters:
-                input = module(input, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen)
-            else:
-                input = module(input)
-        return input
-
-
-class MixFormerSequentialForCausalLM(MixFormerSequentialPreTrainedModel):
-    """MixFormer (sequential for DeepSpeed) for Causal Language Modeling."""
-
-    _keys_to_ignore_on_load_missing = [""]
-    _keys_to_ignore_on_load_unexpected = [
-        r"layers\.\d+\.mlp.(fc_in|fc_out)\.(weight|bias)"
-    ]
-    _no_split_modules = ["ParallelBlock"]
-
-    def __init__(self, config: MixFormerSequentialConfig) -> None:
-        super().__init__(config)
-
-        modules = [Embedding(config)]
-        block_config = config.architecture
-
-        if not isinstance(block_config, list):
-            block_config = [block_config for _ in range(config.n_layer)]
-
-        if config.n_layer != len(block_config):
-            config.n_layer = len(block_config)
-
-        for block_idx, block in enumerate(block_config):
-            # `block_cls` with `legacy` value is for backward compatibility
-            # `path` key is for backward compatibility
-            block = copy.deepcopy(block) or {"block_cls": "parallel"}
-            block.pop("path", None) or block.pop("block_cls", None)
-
-            block["block_idx"] = block_idx
-            modules.append(ParallelBlock(config, **block))
-
-        modules.append(CausalLMHead(config))
-
-        self.layers = PackedSequential(*modules)
-        self.loss = CausalLMLoss()
-
-        self.post_init()
-
-    def get_input_embeddings(self) -> nn.Embedding:
-        return self.layers[0].wte
-
-    def set_input_embeddings(self, new_embeddings: nn.Embedding) -> None:
-        self.layers[0].wte = new_embeddings
-
-    def get_output_embeddings(self) -> nn.Linear:
-        return self.layers[-1].linear
-
-    def set_output_embeddings(self, new_embeddings: nn.Linear) -> None:
-        self.layers[-1].linear = new_embeddings
-
-    def forward(
-        self,
-        input_ids: torch.LongTensor,
-        labels: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[torch.FloatTensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        **kwargs,
-    ) -> CausalLMOutputWithPast:
-        cu_seqlens: Optional[torch.LongTensor] = None
-        max_seqlen: Optional[int] = None
-        if position_ids is not None:
-            batch_size, seq_length = input_ids.shape
-            position_ids = position_ids.view(-1, seq_length).long()
-            cu_seqlens, max_seqlen = get_cu_seqlens_from_pos_ids(position_ids)
-            cu_seqlens = cu_seqlens.squeeze()
-
-        if not past_key_values:
-            lm_logits = self.layers(
-                input_ids, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen
-            )
-        else:
-            hidden_layer = self.layers[0](input_ids)
-            for module in self.layers[1:-1]:
-                hidden_layer = module(
-                    hidden_layer,
-                    past_cache=past_key_values,
-                    cu_seqlens=cu_seqlens,
-                    max_seqlen=max_seqlen,
-                )
-            lm_logits = self.layers[-1](hidden_layer)
-
-        loss = None
-        if labels is not None:
-            loss = self.loss(lm_logits, labels)
-
-        return CausalLMOutputWithPast(
-            loss=loss, logits=lm_logits, past_key_values=past_key_values
-        )

src/axolotl/monkeypatch/btlm_attn_hijack_flash.py

@@ -1,66 +0,0 @@
-"""
-Flash attention monkey patch for cerebras btlm model
-"""
-
-import importlib
-import logging
-from typing import Optional, Tuple
-
-import torch
-from accelerate import init_empty_weights
-from flash_attn.flash_attn_interface import flash_attn_func
-from transformers import AutoConfig, AutoModelForCausalLM
-
-LOG = logging.getLogger("axolotl")
-
-
-def replace_btlm_attn_with_flash_attn(model_name="cerebras/btlm-3b-8k-base"):
-    # this is a wonky hack to get the remotely loaded module
-    model_config = AutoConfig.from_pretrained(model_name, trust_remote_code=True)
-    # we need to load the model here in order for modeling_btlm to be available
-    with init_empty_weights():
-        AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True)
-    module_name = model_config.__class__.__module__.replace(
-        ".configuration_btlm", ".modeling_btlm"
-    )
-    modeling_btlm = importlib.import_module(module_name)
-    modeling_btlm.BTLMAttention._attn = (  # pylint: disable=protected-access
-        flashattn_attn
-    )
-
-
-def flashattn_attn(
-    self,
-    query: torch.Tensor,
-    key: Optional[torch.Tensor] = None,
-    value: Optional[torch.Tensor] = None,
-    attention_mask: Optional[torch.Tensor] = None,  # pylint: disable=unused-argument
-    head_mask: Optional[torch.Tensor] = None,
-    position_bias: Optional[torch.Tensor] = None,  # pylint: disable=unused-argument
-) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
-    softmax_scale = (
-        1 / (key.size(-1) ** self.attn_scale_power) if self.scale_attn_weights else None
-    )
-
-    query = query.permute(0, 2, 1, 3)
-    key = key.permute(0, 2, 1, 3)
-    value = value.permute(0, 2, 1, 3)
-
-    # Perform Flash attention
-    attn_output = flash_attn_func(
-        query,
-        key,
-        value,
-        dropout_p=0.0,  # Assuming you have this attribute
-        softmax_scale=softmax_scale,  # Set this if you have specific scaling in mind
-        causal=not self.is_cross_attention,  # Assuming you have this attribute
-        return_attn_probs=False,  # Set this based on your needs
-    )
-
-    # Optional: Apply head mask if it's not None
-    if head_mask is not None:
-        attn_output *= head_mask
-
-    attn_output = attn_output.permute(0, 2, 1, 3)
-
-    return attn_output, None  # We don't have explicit attn_weights in Flash attention

src/axolotl/monkeypatch/fastchat_conversation_turns.py

@@ -1,174 +0,0 @@
-"""
-monkeypatch to add a get_turns method
-"""
-
-import logging
-from typing import Generator, Tuple
-
-from fastchat.conversation import SeparatorStyle
-
-LOG = logging.getLogger("axolotl.monkeypatch.fastchat_conversation_turns")
-
-
-def get_prompt(self) -> str:
-    ret = ""
-    for role, msg in self.get_turns():
-        ret += role + msg
-    return ret
-
-
-def get_turns(  # pylint: disable=too-many-return-statements
-    self,
-) -> Generator[Tuple[str, str], None, None]:
-    """Get the prompt for generation."""
-    system_prompt = self.system_template.format(system_message=self.system_message)
-    if self.sep_style == SeparatorStyle.ADD_COLON_SINGLE:
-        yield "", system_prompt + self.sep
-        for role, message in self.messages:
-            if message:
-                yield role + ": ", message + self.sep
-            else:
-                yield role + ":", ""
-        return
-    if self.sep_style == SeparatorStyle.ADD_COLON_TWO:
-        seps = [self.sep, self.sep2]
-        yield "", system_prompt + seps[0]
-        for i, (role, message) in enumerate(self.messages):
-            if message:
-                yield role + ": ", message + seps[i % 2]
-            else:
-                yield role + ":", ""
-        return
-    if self.sep_style == SeparatorStyle.ADD_COLON_SPACE_SINGLE:
-        yield "", system_prompt + self.sep
-        for role, message in self.messages:
-            if message:
-                yield role + ": ", message + self.sep
-            else:
-                yield role + ": ", ""  # must be end with a space
-        return
-    if self.sep_style == SeparatorStyle.ADD_NEW_LINE_SINGLE:
-        yield "", "" if system_prompt == "" else system_prompt + self.sep
-        for role, message in self.messages:
-            if message:
-                yield role + "\n", message + self.sep
-            else:
-                yield role + "\n", ""
-        return
-    if self.sep_style == SeparatorStyle.NO_COLON_SINGLE:
-        yield "", system_prompt
-        for role, message in self.messages:
-            if message:
-                yield role, message + self.sep
-            else:
-                yield role, ""
-        return
-    if self.sep_style == SeparatorStyle.NO_COLON_TWO:
-        seps = [self.sep, self.sep2]
-        yield "", system_prompt
-        for i, (role, message) in enumerate(self.messages):
-            if message:
-                yield role, message + seps[i % 2]
-            else:
-                yield role, ""
-        return
-    if self.sep_style == SeparatorStyle.RWKV:
-        yield "", system_prompt
-        for i, (role, message) in enumerate(self.messages):
-            if message:
-                yield role + ": ", message.replace("\r\n", "\n").replace(
-                    "\n\n", "\n"
-                ) + "\n\n"
-            else:
-                yield role + ":", ""
-        return
-    if self.sep_style == SeparatorStyle.LLAMA2:
-        seps = [self.sep, self.sep2]
-        if self.system_message:
-            yield "", system_prompt
-        else:
-            yield "", "[INST] "
-        for i, (role, message) in enumerate(self.messages[1:]):
-            if message:
-                yield role + " ", message + seps[i % 2]
-            else:
-                yield role, ""
-        return
-    if self.sep_style == SeparatorStyle.CHATGLM:
-        # source: https://huggingface.co/THUDM/chatglm-6b/blob/1d240ba371910e9282298d4592532d7f0f3e9f3e/modeling_chatglm.py#L1302-L1308
-        # source2: https://huggingface.co/THUDM/chatglm2-6b/blob/e186c891cf64310ac66ef10a87e6635fa6c2a579/modeling_chatglm.py#L926
-        round_add_n = 1 if self.name == "chatglm2" else 0
-        if system_prompt:
-            yield "", system_prompt + self.sep
-
-        for i, (role, message) in enumerate(self.messages):
-            if i % 2 == 0:
-                yield "", f"[Round {i//2 + round_add_n}]{self.sep}"
-
-            if message:
-                yield f"{role}：", f"{message}{self.sep}"
-            else:
-                yield f"{role}：", ""
-        return
-    if self.sep_style == SeparatorStyle.CHATML:
-        yield "", "" if system_prompt == "" else system_prompt + self.sep + "\n"
-        for role, message in self.messages:
-            if message:
-                yield role + "\n", message + self.sep + "\n"
-            else:
-                yield role + "\n", ""
-        return
-    if self.sep_style == SeparatorStyle.CHATINTERN:
-        # source: https://huggingface.co/internlm/internlm-chat-7b-8k/blob/bd546fa984b4b0b86958f56bf37f94aa75ab8831/modeling_internlm.py#L771
-        seps = [self.sep, self.sep2]
-        yield "", system_prompt
-        for i, (role, message) in enumerate(self.messages):
-            prefix = "<s>" if i % 2 == 0 else ""
-            if message:
-                yield prefix + role + ":", message + seps[i % 2] + "\n"
-            else:
-                yield role + ":", ""
-        return
-    if self.sep_style == SeparatorStyle.DOLLY:
-        seps = [self.sep, self.sep2]
-        yield "", system_prompt
-        for i, (role, message) in enumerate(self.messages):
-            if message:
-                suffix = "\n\n" if i % 2 == 1 else ""
-                yield role + ":\n", message + seps[i % 2] + suffix
-            else:
-                yield role + ":\n", ""
-        return
-    if self.sep_style == SeparatorStyle.PHOENIX:
-        yield "", system_prompt
-        for role, message in self.messages:
-            if message:
-                yield role + ": ", "<s>" + message + "</s>"
-            else:
-                yield role + ": " + "<s>", ""
-        return
-    if self.sep_style == SeparatorStyle.ROBIN:
-        yield "", system_prompt + self.sep
-        for role, message in self.messages:
-            if message:
-                yield role + ":\n", message + self.sep
-            else:
-                yield role + ":\n", ""
-        return
-    if self.sep_style == SeparatorStyle.FALCON_CHAT:
-        if self.system_message:
-            yield "", system_prompt + self.sep
-        for role, message in self.messages:
-            if message:
-                yield role + ": ", message + self.sep
-            else:
-                yield role + ":", ""
-    else:
-        raise ValueError(f"Invalid style: {self.sep_style}")
-
-
-def add_get_turns_to_conversation():
-    import fastchat.conversation
-
-    fastchat.conversation.Conversation.get_turns = get_turns
-    fastchat.conversation.Conversation.get_prompt = get_prompt

src/axolotl/monkeypatch/llama_attn_hijack_flash.py

@@ -2,202 +2,142 @@
 
 # copied from https://github.com/lm-sys/FastChat/blob/main/fastchat/train/llama_flash_attn_monkey_patch.py
 
-import logging
-import warnings
-from functools import partial
-from typing import List, Optional, Tuple, Union
+from typing import Optional, Tuple
 
 import torch
-import torch.nn.functional as F
 import transformers
 from einops import rearrange
 from flash_attn.bert_padding import pad_input, unpad_input
-from transformers.modeling_outputs import BaseModelOutputWithPast
-from transformers.models.llama.modeling_llama import LlamaAttention
-from transformers.models.llama.modeling_llama import (
-    LlamaDecoderLayer as OriginalLlamaDecoderLayer,
-)
-from transformers.models.llama.modeling_llama import (
-    LlamaMLP,
-    apply_rotary_pos_emb,
-    repeat_kv,
-)
-from xformers.ops import SwiGLU
-
-from axolotl.monkeypatch.utils import get_cu_seqlens_from_pos_ids, set_module_name
 
 try:
-    from flash_attn.flash_attn_interface import (  # pylint: disable=ungrouped-imports
-        flash_attn_kvpacked_func,
-        flash_attn_varlen_kvpacked_func,
-        flash_attn_varlen_qkvpacked_func,
-    )
+    from flash_attn.flash_attn_interface import flash_attn_varlen_qkvpacked_func
 except ImportError:
-    from flash_attn.flash_attn_interface import (
-        flash_attn_unpadded_kvpacked_func as flash_attn_varlen_kvpacked_func,
-    )
     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
 
-LOG = logging.getLogger("axolotl")
+from axolotl.monkeypatch.utils import get_cu_seqlens_from_pos_ids
 
 
-def replace_llama_mlp_with_swiglu(model):
-    for name, module in model.named_modules():
-        if isinstance(module, LlamaMLP):
-            mlp = FusedMLP(
-                module.config, module.gate_proj, module.up_proj, module.down_proj
-            )
-            set_module_name(model, name, mlp)
+def forward(
+    self,
+    hidden_states: torch.Tensor,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.Tensor] = 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]]]:
+    """Input shape: Batch x Time x Channel
 
+    attention_mask: [bsz, q_len]
+    """
+    # pylint: disable=duplicate-code
+    bsz, q_len, _ = hidden_states.size()
 
-def replace_llama_qkv_with_fused(model):
-    for name, module in model.named_modules():
-        if isinstance(module, LlamaAttention):
-            qkv = FusedAttention(
-                module.config,
-                module.q_proj,
-                module.k_proj,
-                module.v_proj,
-                module.o_proj,
-            )
-            set_module_name(model, name, qkv)
-
-
-def replace_llama_attn_with_flash_attn(
-    packed: Optional[bool] = False,
-    cross_entropy: Optional[bool] = False,
-    rms_norm: Optional[bool] = False,
-):
-    transformers.models.llama.modeling_llama.LlamaModel._prepare_decoder_attention_mask = (  # pylint: disable=protected-access
-        _prepare_decoder_attention_mask
+    query_states = (
+        self.q_proj(hidden_states)
+        .view(bsz, q_len, self.num_heads, self.head_dim)
+        .transpose(1, 2)
     )
-    transformers.models.llama.modeling_llama.LlamaAttention.forward = flashattn_forward
-    if packed:
-        transformers.models.llama.modeling_llama.LlamaDecoderLayer = LlamaDecoderLayer
-        transformers.models.llama.modeling_llama.LlamaModel.forward = (
-            llama_model_forward
+    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)
+    )
+    # [bsz, q_len, nh, hd]
+    # [bsz, nh, q_len, hd]
+
+    kv_seq_len = key_states.shape[-2]
+    assert past_key_value is None, "past_key_value is not supported"
+
+    cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+    query_states, key_states = apply_rotary_pos_emb(
+        query_states, key_states, cos, sin, position_ids
+    )
+    # [bsz, nh, t, hd]
+    assert not output_attentions, "output_attentions is not supported"
+    assert not use_cache, "use_cache is not supported"
+
+    # Flash attention codes from
+    # https://github.com/HazyResearch/flash-attention/blob/main/flash_attn/flash_attention.py
+
+    # transform the data into the format required by flash attention
+    qkv = torch.stack(
+        [query_states, key_states, value_states], dim=2
+    )  # [bsz, nh, 3, q_len, hd]
+    qkv = qkv.transpose(1, 3)  # [bsz, q_len, 3, nh, hd]
+    # We have disabled _prepare_decoder_attention_mask in LlamaModel
+    # the attention_mask should be the same as the key_padding_mask
+    key_padding_mask = attention_mask
+
+    if key_padding_mask is None:
+        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,
+        )
+        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 position_ids.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,
+        )
+        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,
+            ),
+            "b s (h d) -> b s h d",
+            h=nheads,
         )
 
-    # skip only if explicitly disabled
-    if cross_entropy:
-        try:
-            from flash_attn.losses.cross_entropy import CrossEntropyLoss
-
-            LOG.info("patching with flash_attn.losses.cross_entropy")
-            transformers.models.llama.modeling_llama.CrossEntropyLoss = partial(
-                CrossEntropyLoss, inplace_backward=True
-            )
-        except ImportError:
-            LOG.info(
-                "optimized flash-attention CrossEntropyLoss not found (run `pip install 'git+https://github.com/Dao-AILab/flash-attention.git#egg=xentropy_cuda_lib&subdirectory=csrc/xentropy'`)"
-            )
-
-    # skip only if explicitly disabled
-    if rms_norm:
-        try:
-            from flash_attn.ops.rms_norm import RMSNorm
-
-            class LlamaRMSNorm(RMSNorm):
-                """Patched LLamaRMSNorm"""
-
-                def __init__(self, hidden_size, eps=1e-6):
-                    super().__init__(hidden_size, eps=eps)
-
-            LOG.info("patching with flash_attn.ops.rms_norm")
-            transformers.models.llama.modeling_llama.LlamaRMSNorm = LlamaRMSNorm
-        except ImportError:
-            LOG.info(
-                "optimized flash-attention RMSNorm not found (run `pip install 'git+https://github.com/Dao-AILab/flash-attention.git#egg=dropout_layer_norm&subdirectory=csrc/layer_norm'`)"
-            )
-
-
-class FusedAttention(LlamaAttention):
-    """
-    Fused QKV Attention layer for incrementally improved training efficiency
-    """
-
-    def __init__(
-        self,
-        config,
-        q: torch.nn.Linear,  # pylint: disable=invalid-name
-        k: torch.nn.Linear,  # pylint: disable=invalid-name
-        v: torch.nn.Linear,  # pylint: disable=invalid-name
-        o: torch.nn.Linear,  # pylint: disable=invalid-name
-    ):
-        super().__init__(config)
-        self.config = config
-        self.init_device = next(iter(q.state_dict().values())).device
-
-        # define equivalent fused qkv projection
-        self.out_features: List[int] = [q.out_features, k.out_features, v.out_features]
-        self.qkv_proj = torch.nn.Linear(
-            q.in_features, sum(self.out_features), device=self.init_device, bias=False
-        )
-        self.o_proj = o
-
-        # overwrite initialized weights with pretrained weights
-        self.qkv_proj.weight.data = torch.cat(
-            (q.weight.data, k.weight.data, v.weight.data), dim=0
-        )
-
-    def _post_training(self, model, name):
-        q_proj, k_proj, v_proj = torch.split(
-            self.qkv_proj.weight.data, self.out_features, dim=0
-        )
-
-        new_attn = LlamaAttention(self.config)
-        new_attn.q_proj.weight.data = q_proj
-        new_attn.k_proj.weight.data = k_proj
-        new_attn.v_proj.weight.data = v_proj
-        new_attn.o_proj.weight.data = self.o_proj.weight.data
-
-        set_module_name(model, name, new_attn)
-
-
-class FusedMLP(torch.nn.Module):
-    """
-    Fused MLP layer for incrementally improved training efficiency
-    """
-
-    def __init__(
-        self,
-        config,
-        gate_proj: torch.nn.Linear,
-        up_proj: torch.nn.Linear,
-        down_proj: torch.nn.Linear,
-    ):
-        super().__init__()
-        self.config = config
-        self.swiglu = SwiGLU(
-            in_features=config.hidden_size,
-            hidden_features=config.intermediate_size,
-            bias=False,
-            _pack_weights=True,
-        )
-        # overwrite initialized weights with pretrained weights
-        self.swiglu.w12.weight.data = torch.cat(
-            (gate_proj.weight.data, up_proj.weight.data), dim=0
-        )
-        self.swiglu.w3.weight.data = down_proj.weight.data
-
-    def _post_training(self, model, name):
-        w1, w2 = torch.split(  # pylint: disable=invalid-name
-            self.swiglu.w12.weight.data, self.config.intermediate_size, dim=0
-        )
-
-        # Assign the split weights back to the original layers
-        new_mlp = LlamaMLP(self.config)
-        new_mlp.gate_proj.weight.data = w1
-        new_mlp.up_proj.weight.data = w2
-        new_mlp.down_proj.weight.data = self.swiglu.w3.weight.data
-
-        set_module_name(model, name, new_mlp)
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:  # pylint: disable=invalid-name
-        return self.swiglu(x)
+    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
@@ -213,562 +153,8 @@ def _prepare_decoder_attention_mask(
     return attention_mask
 
 
-def flashattn_forward(
-    self,
-    hidden_states: torch.Tensor,
-    attention_mask: Optional[torch.Tensor] = None,
-    position_ids: Optional[torch.Tensor] = None,
-    past_key_value: Optional[Tuple[torch.Tensor]] = None,
-    output_attentions: bool = False,
-    use_cache: bool = False,
-    padding_mask: Optional[torch.LongTensor] = None,  # pylint: disable=unused-argument
-    cu_seqlens: Optional[torch.Tensor] = None,
-    max_seqlen: Optional[torch.Tensor] = None,
-) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
-    """Input shape: Batch x Time x Channel
-
-    attention_mask: [bsz, q_len]
-    """
-    # 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:
-        if isinstance(self, FusedAttention):
-            query_states, key_states, value_states = self.qkv_proj(hidden_states).split(
-                self.out_features, dim=-1
-            )
-        else:
-            query_states = self.q_proj(hidden_states)
-            key_states = self.k_proj(hidden_states)
-            value_states = self.v_proj(hidden_states)
-
-    query_states = 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)
-    # [bsz, q_len, nh, hd]
-    # [bsz, nh, q_len, hd]
-
-    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 = apply_rotary_pos_emb(
-        query_states, key_states, cos, sin, position_ids
+def replace_llama_attn_with_flash_attn():
+    transformers.models.llama.modeling_llama.LlamaModel._prepare_decoder_attention_mask = (  # pylint: disable=protected-access
+        _prepare_decoder_attention_mask
     )
-    # [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 = repeat_kv(key_states, self.num_key_value_groups)
-    value_states = repeat_kv(value_states, self.num_key_value_groups)
-
-    if output_attentions:
-        warnings.warn(
-            "Output attentions is not supported for patched `LlamaAttention`, returning `None` instead."
-        )
-
-    #
-    # flash-attn v2 start
-    #
-
-    if self.training:
-        # during training q,k,v always have same seqlen
-        assert key_states.shape == query_states.shape
-        is_causal = True
-    else:
-        # turn off FA causal mask after first inference autoregressive iteration
-        # only on first autoregressive step q,k,v have same seqlen
-        is_causal = key_states.shape == query_states.shape
-
-    if cu_seqlens is not None and max_seqlen is not None and cu_seqlens.dim() == 1:
-        # special handling using sample packing
-        qkv = torch.stack(
-            [query_states, key_states, value_states], dim=2
-        )  # [bsz, nh, 3, q_len, hd]
-        qkv = qkv.transpose(1, 3)  # [bsz, q_len, 3, nh, hd]
-        qkv = rearrange(qkv, "b s ... -> (b s) ...")
-
-        output = flash_attn_varlen_qkvpacked_func(
-            qkv, cu_seqlens, max_seqlen, 0.0, softmax_scale=None, causal=True
-        )
-        output = rearrange(output, "(b s) ... -> b s ...", b=bsz)
-    elif query_states.shape == key_states.shape:
-        query_states = query_states.transpose(1, 2)
-        key_states = key_states.transpose(1, 2)
-        value_states = value_states.transpose(1, 2)
-        qkv_unpad, cu_seqlens_q, max_seqlen_q, _, output_pad_fn = generate_qkv(
-            query_states,
-            key_states,
-            value_states,
-            qkvpacked=True,
-            # We have disabled _prepare_decoder_attention_mask in LlamaModel
-            # the attention_mask should be the same as the key_padding_mask
-            key_padding_mask=attention_mask,
-            query_padding_mask=attention_mask[:, -query_states.size(1) :]
-            if attention_mask is not None
-            else None,
-        )
-        output_unpad = flash_attn_varlen_qkvpacked_func(
-            qkv_unpad,
-            cu_seqlens_q,
-            max_seqlen_q,
-            0.0,
-            softmax_scale=None,
-            causal=is_causal,
-        )
-        output = output_pad_fn(output_unpad)
-    else:
-        query_states = query_states.transpose(1, 2)
-        key_states = key_states.transpose(1, 2)
-        value_states = value_states.transpose(1, 2)
-        if attention_mask is None or attention_mask.all().item():
-            output = flash_attn_kvpacked_func(
-                query_states,
-                torch.stack([key_states, value_states], 2),
-                causal=is_causal,
-            )
-        else:
-            (  # pylint: disable=unbalanced-tuple-unpacking
-                q_unpad,
-                kv_unpad,
-                cu_seqlens_q,
-                cu_seqlens_k,
-                max_seqlen_q,
-                max_seqlen_k,
-                _,
-                _,
-                output_pad_fn,
-            ) = generate_qkv(
-                query_states,
-                key_states,
-                value_states,
-                kvpacked=True,
-                key_padding_mask=attention_mask,
-                query_padding_mask=attention_mask[:, -query_states.size(1) :]
-                if attention_mask is not None
-                else None,
-            )
-            if q_unpad.dtype != kv_unpad.dtype:
-                kv_unpad = kv_unpad.to(q_unpad.dtype)
-            output_unpad = flash_attn_varlen_kvpacked_func(
-                q_unpad,
-                kv_unpad,
-                cu_seqlens_q,
-                cu_seqlens_k,
-                max_seqlen_q,
-                max_seqlen_k,
-                0.0,
-                softmax_scale=None,
-                causal=is_causal,
-            )
-            output = output_pad_fn(output_unpad)
-
-    attn_output = output
-    if attn_output.size() != (bsz, q_len, self.num_heads, self.head_dim):
-        raise ValueError(
-            f"`attn_output` should be of size {(bsz, q_len, self.num_heads, self.head_dim)}, but is"
-            f" {attn_output.size()}"
-        )
-    attn_output = rearrange(attn_output, "b s h d -> b s (h d)")
-
-    #
-    # flash-attn v2 end
-    #
-
-    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, None, past_key_value
-
-
-# based on https://github.com/Dao-AILab/flash-attention/blob/364a5b/tests/test_flash_attn.py#L38
-def generate_qkv(
-    q,
-    k,
-    v,
-    query_padding_mask=None,
-    key_padding_mask=None,
-    kvpacked=False,
-    qkvpacked=False,
-):  # pylint: disable=invalid-name,unnecessary-lambda-assignment
-    """
-    Arguments:
-        q: (batch_size, seqlen_q, nheads, d)
-        k: (batch_size, seqlen_k, nheads_k, d)
-        v: (batch_size, seqlen_k, nheads_k, d)
-        query_padding_mask: (batch_size, seqlen), bool
-        key_padding_mask: (batch_size, seqlen), bool
-    """
-    assert not (kvpacked and qkvpacked)
-    batch_size, seqlen_q, nheads, d = q.shape
-    _, seqlen_k, nheads_k, _ = k.shape
-    assert k.shape == (batch_size, seqlen_k, nheads_k, d)
-    assert v.shape == (batch_size, seqlen_k, nheads_k, d)
-
-    if query_padding_mask is not None:
-        q_unpad, indices_q, cu_seqlens_q, max_seqlen_q = unpad_input(
-            q, query_padding_mask
-        )
-
-        output_pad_fn = lambda output_unpad: pad_input(  # noqa: E731
-            output_unpad, indices_q, batch_size, seqlen_q
-        )
-
-    else:
-        q_unpad = rearrange(q, "b s h d -> (b s) h d")
-        cu_seqlens_q = torch.arange(
-            0,
-            (batch_size + 1) * seqlen_q,
-            step=seqlen_q,
-            dtype=torch.int32,
-            device=q_unpad.device,
-        )
-        max_seqlen_q = seqlen_q
-
-        output_pad_fn = lambda output_unpad: rearrange(  # noqa: E731
-            output_unpad, "(b s) h d -> b s h d", b=batch_size
-        )
-
-    if key_padding_mask is not None:
-        k_unpad, _, cu_seqlens_k, max_seqlen_k = unpad_input(k, key_padding_mask)
-        v_unpad, _, _, _ = unpad_input(v, key_padding_mask)
-    else:
-        k_unpad = rearrange(k, "b s h d -> (b s) h d")
-        v_unpad = rearrange(v, "b s h d -> (b s) h d")
-        cu_seqlens_k = torch.arange(
-            0,
-            (batch_size + 1) * seqlen_k,
-            step=seqlen_k,
-            dtype=torch.int32,
-            device=k_unpad.device,
-        )
-        max_seqlen_k = seqlen_k
-
-    if qkvpacked:
-        assert nheads == nheads_k
-        qkv_unpad = torch.stack([q_unpad, k_unpad, v_unpad], dim=1)
-        qkv = torch.stack([q, k, v], dim=2)
-        return (qkv_unpad, cu_seqlens_q, max_seqlen_q, qkv, output_pad_fn)
-
-    if kvpacked:
-        kv_unpad = torch.stack([k_unpad, v_unpad], dim=1)
-        kv = torch.stack([k, v], dim=2)
-        return (
-            q_unpad,
-            kv_unpad,
-            cu_seqlens_q,
-            cu_seqlens_k,
-            max_seqlen_q,
-            max_seqlen_k,
-            q,
-            kv,
-            output_pad_fn,
-        )
-
-    return (
-        q_unpad,
-        k_unpad,
-        v_unpad,
-        cu_seqlens_q,
-        cu_seqlens_k,
-        max_seqlen_q,
-        max_seqlen_k,
-        q,
-        k,
-        v,
-        output_pad_fn,
-    )
-
-
-def llama_model_forward(
-    self,
-    input_ids: torch.LongTensor = None,
-    attention_mask: Optional[torch.Tensor] = None,
-    position_ids: Optional[torch.LongTensor] = None,
-    past_key_values: Optional[List[torch.FloatTensor]] = None,
-    inputs_embeds: Optional[torch.FloatTensor] = None,
-    use_cache: Optional[bool] = None,
-    output_attentions: Optional[bool] = None,
-    output_hidden_states: Optional[bool] = None,
-    return_dict: Optional[bool] = None,
-) -> Union[Tuple, BaseModelOutputWithPast]:
-    output_attentions = (
-        output_attentions
-        if output_attentions is not None
-        else self.config.output_attentions
-    )
-    output_hidden_states = (
-        output_hidden_states
-        if output_hidden_states is not None
-        else self.config.output_hidden_states
-    )
-    use_cache = use_cache if use_cache is not None else self.config.use_cache
-
-    return_dict = (
-        return_dict if return_dict is not None else self.config.use_return_dict
-    )
-
-    # retrieve input_ids and inputs_embeds
-    if input_ids is not None and inputs_embeds is not None:
-        raise ValueError(
-            "You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time"
-        )
-    if input_ids is not None:
-        batch_size, seq_length = input_ids.shape
-    elif inputs_embeds is not None:
-        batch_size, seq_length, _ = inputs_embeds.shape
-    else:
-        raise ValueError(
-            "You have to specify either decoder_input_ids or decoder_inputs_embeds"
-        )
-
-    seq_length_with_past = seq_length
-    past_key_values_length = 0
-
-    if past_key_values is not None:
-        past_key_values_length = past_key_values[0][0].shape[2]
-        seq_length_with_past = seq_length_with_past + past_key_values_length
-
-    cu_seqlens = None
-    max_seqlen = None
-    if position_ids is None:
-        device = input_ids.device if input_ids is not None else inputs_embeds.device
-        position_ids = torch.arange(
-            past_key_values_length,
-            seq_length + past_key_values_length,
-            dtype=torch.long,
-            device=device,
-        )
-        position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
-    else:
-        position_ids = position_ids.view(-1, seq_length).long()
-        cu_seqlens, max_seqlen = get_cu_seqlens_from_pos_ids(position_ids)
-        cu_seqlens = cu_seqlens.squeeze()
-
-    if inputs_embeds is None:
-        inputs_embeds = self.embed_tokens(input_ids)
-    # embed positions
-    if attention_mask is None:
-        attention_mask = torch.ones(
-            (batch_size, seq_length_with_past),
-            dtype=torch.bool,
-            device=inputs_embeds.device,
-        )
-        padding_mask = None
-    else:
-        if 0 in attention_mask:
-            padding_mask = attention_mask
-        else:
-            padding_mask = None
-
-    attention_mask = (
-        self._prepare_decoder_attention_mask(  # pylint: disable=protected-access
-            attention_mask,
-            (batch_size, seq_length),
-            inputs_embeds,
-            past_key_values_length,
-        )
-    )
-
-    hidden_states = inputs_embeds
-
-    if self.gradient_checkpointing and self.training:
-        if use_cache:
-            transformers.logger.warning_once(
-                "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
-            )
-            use_cache = False
-
-    # decoder layers
-    all_hidden_states = () if output_hidden_states else None
-    all_self_attns = () if output_attentions else None
-    next_decoder_cache = () if use_cache else None
-
-    for idx, decoder_layer in enumerate(self.layers):
-        if output_hidden_states:
-            all_hidden_states += (hidden_states,)
-
-        past_key_value = past_key_values[idx] if past_key_values is not None else None
-
-        if self.gradient_checkpointing and self.training:
-
-            def create_custom_forward(module):
-                def custom_forward(*inputs):
-                    # None for past_key_value
-                    return module(
-                        *inputs,
-                    )
-
-                return custom_forward
-
-            layer_outputs = torch.utils.checkpoint.checkpoint(
-                create_custom_forward(decoder_layer),
-                hidden_states,
-                attention_mask,
-                position_ids,
-                past_key_value,
-                output_attentions,
-                None,
-                padding_mask,
-                cu_seqlens,
-                max_seqlen,
-            )
-        else:
-            layer_outputs = decoder_layer(
-                hidden_states,
-                attention_mask=attention_mask,
-                position_ids=position_ids,
-                past_key_value=past_key_value,
-                output_attentions=output_attentions,
-                use_cache=use_cache,
-                padding_mask=padding_mask,
-                cu_seqlens=cu_seqlens,
-                max_seqlen=max_seqlen,
-            )
-
-        hidden_states = layer_outputs[0]
-
-        if use_cache:
-            next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
-
-        if output_attentions:
-            all_self_attns += (layer_outputs[1],)
-
-    hidden_states = self.norm(hidden_states)
-
-    # add hidden states from the last decoder layer
-    if output_hidden_states:
-        all_hidden_states += (hidden_states,)
-
-    next_cache = next_decoder_cache if use_cache else None
-    if not return_dict:
-        return tuple(
-            v
-            for v in [hidden_states, next_cache, all_hidden_states, all_self_attns]
-            if v is not None
-        )
-    return BaseModelOutputWithPast(
-        last_hidden_state=hidden_states,
-        past_key_values=next_cache,
-        hidden_states=all_hidden_states,
-        attentions=all_self_attns,
-    )
-
-
-class LlamaDecoderLayer(OriginalLlamaDecoderLayer):
-    """
-    patched version of LlamaDecoderLayer to pass through the precalculated cu_seqlens
-    """
-
-    def 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: Optional[bool] = False,
-        use_cache: Optional[bool] = False,
-        padding_mask: Optional[torch.LongTensor] = None,
-        cu_seqlens: Optional[torch.Tensor] = None,
-        max_seqlen: Optional[torch.Tensor] = None,
-    ) -> Tuple[
-        torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]
-    ]:
-        """
-        Args:
-            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
-            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
-                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
-            output_attentions (`bool`, *optional*):
-                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
-                returned tensors for more detail.
-            use_cache (`bool`, *optional*):
-                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
-                (see `past_key_values`).
-            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
-            cu_seqlens (`torch.Tensor`, *optional*) cumulative sequence len when packing
-        """
-
-        residual = hidden_states
-
-        hidden_states = self.input_layernorm(hidden_states)
-
-        # Self Attention
-        hidden_states, self_attn_weights, present_key_value = self.self_attn(
-            hidden_states=hidden_states,
-            attention_mask=attention_mask,
-            position_ids=position_ids,
-            past_key_value=past_key_value,
-            output_attentions=output_attentions,
-            use_cache=use_cache,
-            padding_mask=padding_mask,
-            cu_seqlens=cu_seqlens,
-            max_seqlen=max_seqlen,
-        )
-        hidden_states = residual + hidden_states
-
-        # Fully Connected
-        residual = hidden_states
-        hidden_states = self.post_attention_layernorm(hidden_states)
-        hidden_states = self.mlp(hidden_states)
-        hidden_states = residual + hidden_states
-
-        outputs = (hidden_states,)
-
-        if output_attentions:
-            outputs += (self_attn_weights,)
-
-        if use_cache:
-            outputs += (present_key_value,)
-
-        return outputs
+    transformers.models.llama.modeling_llama.LlamaAttention.forward = forward

src/axolotl/monkeypatch/llama_attn_hijack_sdp.py

@@ -1,140 +0,0 @@
-"""
-Patched LlamaAttention to use torch.nn.functional.scaled_dot_product_attention
-"""
-
-import warnings
-from typing import Optional, Tuple
-
-import torch
-import torch.nn.functional as F
-import transformers.models.llama.modeling_llama
-from transformers.models.llama.modeling_llama import apply_rotary_pos_emb, repeat_kv
-
-
-def hijack_llama_sdp_attention():
-    transformers.models.llama.modeling_llama.LlamaAttention.forward = (
-        sdp_attention_forward
-    )
-
-
-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()
-
-    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)
-    # [bsz, q_len, nh, hd]
-    # [bsz, nh, q_len, hd]
-
-    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 = 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 = repeat_kv(key_states, self.num_key_value_groups)
-    value_states = repeat_kv(value_states, self.num_key_value_groups)
-
-    if output_attentions:
-        warnings.warn(
-            "Output attentions is not supported for patched `LlamaAttention`, returning `None` instead."
-        )
-
-    #
-    # sdp-attn start
-    #
-
-    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,
-        )
-
-    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)
-
-    #
-    # sdp-attn end
-    #
-
-    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, None, past_key_value

src/axolotl/monkeypatch/llama_attn_hijack_xformers.py

@@ -3,13 +3,13 @@ Directly copied the code from https://raw.githubusercontent.com/oobabooga/text-g
 """
 
 import logging
-import warnings
+import math
 from typing import Optional, Tuple
 
 import torch
 import torch.nn.functional as F
 import transformers.models.llama.modeling_llama
-from transformers.models.llama.modeling_llama import apply_rotary_pos_emb, repeat_kv
+from torch import nn
 
 try:
     import xformers.ops
@@ -21,6 +21,12 @@ 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,
@@ -75,15 +81,15 @@ def xformers_forward(
     value_states = value_states.view(
         bsz, q_len, self.num_key_value_heads, self.head_dim
     ).transpose(1, 2)
-    # [bsz, q_len, nh, hd]
-    # [bsz, nh, q_len, hd]
 
     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 = apply_rotary_pos_emb(
+    (
+        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]
@@ -96,50 +102,74 @@ def xformers_forward(
     past_key_value = (key_states, value_states) if use_cache else None
 
     # repeat k/v heads if n_kv_heads < n_heads
-    key_states = repeat_kv(key_states, self.num_key_value_groups)
-    value_states = repeat_kv(value_states, self.num_key_value_groups)
+    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
+    )
 
-    if output_attentions:
-        warnings.warn(
-            "Output attentions is not supported for patched `LlamaAttention`, returning `None` instead."
-        )
+    # 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)
 
-    #
-    # xformers-attn start
-    #
-
-    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
-        )
+        # 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:
-        # 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 = 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
 
-    if attn_output.size() != (bsz, q_len, self.num_heads, self.head_dim):
-        raise ValueError(
-            f"`attn_output` should be of size {(bsz, q_len, self.num_heads, self.head_dim)}, but is"
-            f" {attn_output.size()}"
-        )
     attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
 
-    #
-    # xformers-attn end
-    #
-
     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(
@@ -152,4 +182,103 @@ def xformers_forward(
     else:
         attn_output = self.o_proj(attn_output)
 
-    return attn_output, None, past_key_value
+    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/mistral_attn_hijack_flash.py

@@ -1,640 +0,0 @@
-"""Flash attention monkey patch for mistral model"""
-# pylint: disable=duplicate-code
-
-import logging
-from typing import List, Optional, Tuple, Union
-
-import torch
-import transformers
-from einops import rearrange
-from flash_attn.bert_padding import pad_input, unpad_input
-from flash_attn.flash_attn_interface import (  # pylint: disable=ungrouped-imports
-    flash_attn_kvpacked_func,
-    flash_attn_varlen_kvpacked_func,
-    flash_attn_varlen_qkvpacked_func,
-)
-from transformers.modeling_outputs import BaseModelOutputWithPast
-from transformers.models.mistral.modeling_mistral import (
-    MistralAttention as OriginalMistralAttention,
-)
-from transformers.models.mistral.modeling_mistral import (
-    MistralDecoderLayer as OriginalMistralDecoderLayer,
-)
-from transformers.models.mistral.modeling_mistral import apply_rotary_pos_emb, repeat_kv
-
-from axolotl.monkeypatch.utils import get_cu_seqlens_from_pos_ids
-
-LOG = logging.getLogger("axolotl.monkeypatch.mistral")
-
-
-def replace_mistral_attn_with_flash_attn(
-    packed: Optional[bool] = False,
-):
-    transformers.models.mistral.modeling_mistral.MistralModel._prepare_decoder_attention_mask = (  # pylint: disable=protected-access
-        _prepare_decoder_attention_mask
-    )
-    transformers.models.mistral.modeling_mistral.MistralAttention.forward = (
-        flashattn_forward
-    )
-    if packed:
-        transformers.models.mistral.modeling_mistral.MistralDecoderLayer = (
-            MistralDecoderLayer
-        )
-        transformers.models.mistral.modeling_mistral.MistralModel.forward = (
-            mistral_model_forward
-        )
-
-
-@torch.jit.script
-def _make_sliding_window_causal_mask(
-    bsz: int,
-    tgt_len: int,
-    dtype: torch.dtype,
-    device: torch.device,
-    past_key_values_length: int = 0,
-    sliding_window: int = 4096,
-):
-    """
-    Make causal mask used for sliding window attention
-    """
-    tensor = torch.full(
-        (tgt_len, tgt_len),
-        fill_value=1,
-        device=device,
-    )
-    mask = torch.tril(tensor, diagonal=0)
-    # make the mask banded to account for sliding window
-    # NOTE: HF implementation is wrong as of 14-10-2023 for torch.triu, needs +1
-    mask = torch.triu(mask, diagonal=-sliding_window + 1)
-    mask = torch.log(mask).to(dtype)
-
-    if past_key_values_length > 0:
-        mask = torch.cat(
-            [
-                torch.zeros(
-                    tgt_len, past_key_values_length, dtype=dtype, device=device
-                ),
-                mask,
-            ],
-            dim=-1,
-        )
-    return mask[None, None, :, :].expand(
-        bsz, 1, tgt_len, tgt_len + past_key_values_length
-    )
-
-
-# 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,
-    sliding_window,
-):  # pylint: disable=unused-argument
-    # [bsz, seq_len]
-    if attention_mask is None:
-        return attention_mask
-
-    # NOTE: attention mask and sliding masks are only broadcastable in certain scenarios.
-    # Without attention_mask.shape[0] == 1, error will trigger after eval loss but only when wandb is enabled.
-    if input_shape[-1] > 1 and attention_mask.shape[0] == 1:
-        sliding_window_mask = _make_sliding_window_causal_mask(
-            bsz=input_shape[0],
-            tgt_len=input_shape[1],
-            dtype=inputs_embeds.dtype,
-            device=inputs_embeds.device,
-            past_key_values_length=past_key_values_length,
-            sliding_window=sliding_window,
-        )
-        attention_mask = attention_mask + sliding_window_mask
-    else:
-        LOG.info("skipping sliding window mask, not broadcastable with attention mask")
-
-    return attention_mask
-
-
-def flashattn_forward(
-    self: OriginalMistralAttention,
-    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,
-    cu_seqlens: Optional[torch.Tensor] = None,
-    max_seqlen: Optional[torch.Tensor] = None,
-) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
-    bsz, q_len, _ = hidden_states.size()
-
-    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 = apply_rotary_pos_emb(
-        query_states, key_states, cos, sin, position_ids
-    )
-
-    use_sliding_windows = (
-        hasattr(self.config, "sliding_window") is not None
-        and kv_seq_len > self.config.sliding_window
-    )
-
-    if use_sliding_windows:
-        window_size = (self.config.sliding_window, self.config.sliding_window)
-    else:
-        window_size = (-1, -1)
-
-    if past_key_value is not None:
-        # Activate slicing cache only if the config has a value `sliding_windows` attribute
-        if (
-            hasattr(self.config, "sliding_window")
-            and kv_seq_len > self.config.sliding_window
-        ):
-            slicing_tokens = kv_seq_len - self.config.sliding_window
-
-            past_key = past_key_value[0]
-            past_value = past_key_value[1]
-
-            past_key = past_key[:, :, slicing_tokens:, :].contiguous()
-            past_value = past_value[:, :, slicing_tokens:, :].contiguous()
-
-            if past_key.shape[-2] != self.config.sliding_window - 1:
-                raise ValueError(
-                    f"past key much have a shape of (`batch_size, num_heads, self.config.sliding_window-1, head_dim`), got"
-                    f" {past_key.shape}"
-                )
-
-            past_key_value = (past_key, past_value) if use_cache else None
-
-        if past_key_value is not None:
-            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 = repeat_kv(key_states, self.num_key_value_groups)
-    value_states = repeat_kv(value_states, self.num_key_value_groups)
-
-    if self.training:
-        # during training q,k,v always have same seqlen
-        assert key_states.shape == query_states.shape
-        is_causal = True
-    else:
-        # turn off FA causal mask after first inference autoregressive iteration
-        # only on first autoregressive step q,k,v have same seqlen
-        is_causal = key_states.shape == query_states.shape
-
-    if cu_seqlens is not None and max_seqlen is not None and cu_seqlens.dim() == 1:
-        # special handling using sample packing
-        qkv = torch.stack(
-            [query_states, key_states, value_states], dim=2
-        )  # [bsz, nh, 3, q_len, hd]
-        qkv = qkv.transpose(1, 3)  # [bsz, q_len, 3, nh, hd]
-        qkv = rearrange(qkv, "b s ... -> (b s) ...")
-
-        output = flash_attn_varlen_qkvpacked_func(
-            qkv,
-            cu_seqlens,
-            max_seqlen,
-            0.0,
-            softmax_scale=None,
-            causal=True,
-            window_size=window_size,
-        )
-        output = rearrange(output, "(b s) ... -> b s ...", b=bsz)
-    elif query_states.shape == key_states.shape:
-        query_states = query_states.transpose(1, 2)
-        key_states = key_states.transpose(1, 2)
-        value_states = value_states.transpose(1, 2)
-        qkv_unpad, cu_seqlens_q, max_seqlen_q, _, output_pad_fn = generate_qkv(
-            query_states,
-            key_states,
-            value_states,
-            qkvpacked=True,
-            # We have disabled _prepare_decoder_attention_mask in LlamaModel
-            # the attention_mask should be the same as the key_padding_mask
-            key_padding_mask=attention_mask,
-            query_padding_mask=attention_mask[:, -query_states.size(1) :]
-            if attention_mask is not None
-            else None,
-        )
-        output_unpad = flash_attn_varlen_qkvpacked_func(
-            qkv_unpad,
-            cu_seqlens_q,
-            max_seqlen_q,
-            0.0,
-            softmax_scale=None,
-            causal=is_causal,
-            window_size=window_size,
-        )
-        output = output_pad_fn(output_unpad)
-    else:
-        query_states = query_states.transpose(1, 2)
-        key_states = key_states.transpose(1, 2)
-        value_states = value_states.transpose(1, 2)
-        if attention_mask is None or attention_mask.all().item():
-            output = flash_attn_kvpacked_func(
-                query_states,
-                torch.stack([key_states, value_states], 2),
-                causal=is_causal,
-                window_size=window_size,
-            )
-        else:
-            (  # pylint: disable=unbalanced-tuple-unpacking
-                q_unpad,
-                kv_unpad,
-                cu_seqlens_q,
-                cu_seqlens_k,
-                max_seqlen_q,
-                max_seqlen_k,
-                _,
-                _,
-                output_pad_fn,
-            ) = generate_qkv(
-                query_states,
-                key_states,
-                value_states,
-                kvpacked=True,
-                key_padding_mask=attention_mask,
-                query_padding_mask=attention_mask[:, -query_states.size(1) :]
-                if attention_mask is not None
-                else None,
-            )
-            if q_unpad.dtype != kv_unpad.dtype:
-                kv_unpad = kv_unpad.to(q_unpad.dtype)
-            output_unpad = flash_attn_varlen_kvpacked_func(
-                q_unpad,
-                kv_unpad,
-                cu_seqlens_q,
-                cu_seqlens_k,
-                max_seqlen_q,
-                max_seqlen_k,
-                0.0,
-                softmax_scale=None,
-                causal=is_causal,
-                window_size=window_size,
-            )
-            output = output_pad_fn(output_unpad)
-
-    attn_output = output
-    if attn_output.size() != (bsz, q_len, self.num_heads, self.head_dim):
-        raise ValueError(
-            f"`attn_output` should be of size {(bsz, q_len, self.num_heads, self.head_dim)}, but is"
-            f" {attn_output.size()}"
-        )
-    attn_output = rearrange(attn_output, "b s h d -> b s (h d)")
-
-    attn_output = self.o_proj(attn_output)
-
-    if not output_attentions:
-        attn_weights = None
-
-    return attn_output, attn_weights, past_key_value
-
-
-# based on https://github.com/Dao-AILab/flash-attention/blob/364a5b/tests/test_flash_attn.py#L38
-def generate_qkv(
-    q,
-    k,
-    v,
-    query_padding_mask=None,
-    key_padding_mask=None,
-    kvpacked=False,
-    qkvpacked=False,
-):  # pylint: disable=invalid-name,unnecessary-lambda-assignment
-    """
-    Arguments:
-        q: (batch_size, seqlen_q, nheads, d)
-        k: (batch_size, seqlen_k, nheads_k, d)
-        v: (batch_size, seqlen_k, nheads_k, d)
-        query_padding_mask: (batch_size, seqlen), bool
-        key_padding_mask: (batch_size, seqlen), bool
-    """
-    assert not (kvpacked and qkvpacked)
-    batch_size, seqlen_q, nheads, d = q.shape
-    _, seqlen_k, nheads_k, _ = k.shape
-    assert k.shape == (batch_size, seqlen_k, nheads_k, d)
-    assert v.shape == (batch_size, seqlen_k, nheads_k, d)
-
-    if query_padding_mask is not None:
-        q_unpad, indices_q, cu_seqlens_q, max_seqlen_q = unpad_input(
-            q, query_padding_mask
-        )
-
-        output_pad_fn = lambda output_unpad: pad_input(  # noqa: E731
-            output_unpad, indices_q, batch_size, seqlen_q
-        )
-
-    else:
-        q_unpad = rearrange(q, "b s h d -> (b s) h d")
-        cu_seqlens_q = torch.arange(
-            0,
-            (batch_size + 1) * seqlen_q,
-            step=seqlen_q,
-            dtype=torch.int32,
-            device=q_unpad.device,
-        )
-        max_seqlen_q = seqlen_q
-
-        output_pad_fn = lambda output_unpad: rearrange(  # noqa: E731
-            output_unpad, "(b s) h d -> b s h d", b=batch_size
-        )
-
-    if key_padding_mask is not None:
-        k_unpad, _, cu_seqlens_k, max_seqlen_k = unpad_input(k, key_padding_mask)
-        v_unpad, _, _, _ = unpad_input(v, key_padding_mask)
-    else:
-        k_unpad = rearrange(k, "b s h d -> (b s) h d")
-        v_unpad = rearrange(v, "b s h d -> (b s) h d")
-        cu_seqlens_k = torch.arange(
-            0,
-            (batch_size + 1) * seqlen_k,
-            step=seqlen_k,
-            dtype=torch.int32,
-            device=k_unpad.device,
-        )
-        max_seqlen_k = seqlen_k
-
-    if qkvpacked:
-        assert nheads == nheads_k
-        qkv_unpad = torch.stack([q_unpad, k_unpad, v_unpad], dim=1)
-        qkv = torch.stack([q, k, v], dim=2)
-        return (qkv_unpad, cu_seqlens_q, max_seqlen_q, qkv, output_pad_fn)
-
-    if kvpacked:
-        kv_unpad = torch.stack([k_unpad, v_unpad], dim=1)
-        kv = torch.stack([k, v], dim=2)
-        return (
-            q_unpad,
-            kv_unpad,
-            cu_seqlens_q,
-            cu_seqlens_k,
-            max_seqlen_q,
-            max_seqlen_k,
-            q,
-            kv,
-            output_pad_fn,
-        )
-
-    return (
-        q_unpad,
-        k_unpad,
-        v_unpad,
-        cu_seqlens_q,
-        cu_seqlens_k,
-        max_seqlen_q,
-        max_seqlen_k,
-        q,
-        k,
-        v,
-        output_pad_fn,
-    )
-
-
-def mistral_model_forward(
-    self,
-    input_ids: torch.LongTensor = None,
-    attention_mask: Optional[torch.Tensor] = None,
-    position_ids: Optional[torch.LongTensor] = None,
-    past_key_values: Optional[List[torch.FloatTensor]] = None,
-    inputs_embeds: Optional[torch.FloatTensor] = None,
-    use_cache: Optional[bool] = None,
-    output_attentions: Optional[bool] = None,
-    output_hidden_states: Optional[bool] = None,
-    return_dict: Optional[bool] = None,
-) -> Union[Tuple, BaseModelOutputWithPast]:
-    output_attentions = (
-        output_attentions
-        if output_attentions is not None
-        else self.config.output_attentions
-    )
-    output_hidden_states = (
-        output_hidden_states
-        if output_hidden_states is not None
-        else self.config.output_hidden_states
-    )
-    use_cache = use_cache if use_cache is not None else self.config.use_cache
-
-    return_dict = (
-        return_dict if return_dict is not None else self.config.use_return_dict
-    )
-
-    # retrieve input_ids and inputs_embeds
-    if input_ids is not None and inputs_embeds is not None:
-        raise ValueError(
-            "You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time"
-        )
-    if input_ids is not None:
-        batch_size, seq_length = input_ids.shape
-    elif inputs_embeds is not None:
-        batch_size, seq_length, _ = inputs_embeds.shape
-    else:
-        raise ValueError(
-            "You have to specify either decoder_input_ids or decoder_inputs_embeds"
-        )
-
-    seq_length_with_past = seq_length
-    past_key_values_length = 0
-
-    if past_key_values is not None:
-        past_key_values_length = past_key_values[0][0].shape[2]
-        seq_length_with_past = seq_length_with_past + past_key_values_length
-
-    cu_seqlens = None
-    max_seqlen = None
-    if position_ids is None:
-        device = input_ids.device if input_ids is not None else inputs_embeds.device
-        position_ids = torch.arange(
-            past_key_values_length,
-            seq_length + past_key_values_length,
-            dtype=torch.long,
-            device=device,
-        )
-        position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
-    else:
-        position_ids = position_ids.view(-1, seq_length).long()
-        cu_seqlens, max_seqlen = get_cu_seqlens_from_pos_ids(position_ids)
-        cu_seqlens = cu_seqlens.squeeze()
-
-    if inputs_embeds is None:
-        inputs_embeds = self.embed_tokens(input_ids)
-    # embed positions
-    if attention_mask is None:
-        attention_mask = torch.ones(
-            (batch_size, seq_length_with_past),
-            dtype=torch.bool,
-            device=inputs_embeds.device,
-        )
-    attention_mask = (
-        self._prepare_decoder_attention_mask(  # pylint: disable=protected-access
-            attention_mask,
-            (batch_size, seq_length),
-            inputs_embeds,
-            past_key_values_length,
-            sliding_window=self.config.sliding_window,
-        )
-    )
-
-    hidden_states = inputs_embeds
-
-    if self.gradient_checkpointing and self.training:
-        if use_cache:
-            transformers.logger.warning_once(
-                "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
-            )
-            use_cache = False
-
-    # decoder layers
-    all_hidden_states = () if output_hidden_states else None
-    all_self_attns = () if output_attentions else None
-    next_decoder_cache = () if use_cache else None
-
-    for idx, decoder_layer in enumerate(self.layers):
-        if output_hidden_states:
-            all_hidden_states += (hidden_states,)
-
-        past_key_value = past_key_values[idx] if past_key_values is not None else None
-
-        if self.gradient_checkpointing and self.training:
-
-            def create_custom_forward(module):
-                def custom_forward(*inputs):
-                    # None for past_key_value
-                    return module(*inputs)
-
-                return custom_forward
-
-            layer_outputs = torch.utils.checkpoint.checkpoint(
-                create_custom_forward(decoder_layer),
-                hidden_states,
-                attention_mask,
-                position_ids,
-                past_key_value,
-                output_attentions,
-                None,
-                cu_seqlens,
-                max_seqlen,
-            )
-        else:
-            layer_outputs = decoder_layer(
-                hidden_states,
-                attention_mask=attention_mask,
-                position_ids=position_ids,
-                past_key_value=past_key_value,
-                output_attentions=output_attentions,
-                use_cache=use_cache,
-                cu_seqlens=cu_seqlens,
-                max_seqlen=max_seqlen,
-            )
-
-        hidden_states = layer_outputs[0]
-
-        if use_cache:
-            next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
-
-        if output_attentions:
-            all_self_attns += (layer_outputs[1],)
-
-    hidden_states = self.norm(hidden_states)
-
-    # add hidden states from the last decoder layer
-    if output_hidden_states:
-        all_hidden_states += (hidden_states,)
-
-    next_cache = next_decoder_cache if use_cache else None
-    if not return_dict:
-        return tuple(
-            v
-            for v in [hidden_states, next_cache, all_hidden_states, all_self_attns]
-            if v is not None
-        )
-    return BaseModelOutputWithPast(
-        last_hidden_state=hidden_states,
-        past_key_values=next_cache,
-        hidden_states=all_hidden_states,
-        attentions=all_self_attns,
-    )
-
-
-class MistralDecoderLayer(OriginalMistralDecoderLayer):
-    """
-    patched version of MistralDecoderLayer to pass through the precalculated cu_seqlens
-    """
-
-    def 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: Optional[bool] = False,
-        use_cache: Optional[bool] = False,
-        cu_seqlens: Optional[torch.Tensor] = None,
-        max_seqlen: Optional[torch.Tensor] = None,
-    ) -> Tuple[
-        torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]
-    ]:
-        """
-        Args:
-            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
-            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
-                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
-            output_attentions (`bool`, *optional*):
-                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
-                returned tensors for more detail.
-            use_cache (`bool`, *optional*):
-                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
-                (see `past_key_values`).
-            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
-            cu_seqlens (`torch.Tensor`, *optional*) cumulative sequence len when packing
-        """
-
-        residual = hidden_states
-
-        hidden_states = self.input_layernorm(hidden_states)
-
-        # Self Attention
-        hidden_states, self_attn_weights, present_key_value = self.self_attn(
-            hidden_states=hidden_states,
-            attention_mask=attention_mask,
-            position_ids=position_ids,
-            past_key_value=past_key_value,
-            output_attentions=output_attentions,
-            use_cache=use_cache,
-            cu_seqlens=cu_seqlens,
-            max_seqlen=max_seqlen,
-        )
-        hidden_states = residual + hidden_states
-
-        # Fully Connected
-        residual = hidden_states
-        hidden_states = self.post_attention_layernorm(hidden_states)
-        hidden_states = self.mlp(hidden_states)
-        hidden_states = residual + hidden_states
-
-        outputs = (hidden_states,)
-
-        if output_attentions:
-            outputs += (self_attn_weights,)
-
-        if use_cache:
-            outputs += (present_key_value,)
-
-        return outputs