Skip to content

CHANGELOG.md

Latest commit

 

History

History
147 lines (113 loc) · 5.66 KB

CHANGELOG.md

File metadata and controls

147 lines (113 loc) · 5.66 KB
Raw

Changelog

All notable changes to the Model Quantizer project will be documented in this file.

[0.3.3] - 2025-03-15

Added

  • Added explicit links to GitHub repository and PyPI package in documentation
  • Improved model card template to include better credit to Model Quantizer
  • Enhanced installation instructions with direct links to PyPI

[0.3.2] - 2025-03-15

Fixed

  • Fixed installation issue with run-benchmark command not finding the module
  • Added py_modules to setup.py to include standalone scripts in the package
  • Added fix_installation.py script to help users fix existing installations

[0.3.1] - 2025-03-15

Fixed

  • Fixed GPTQ quantization issue by explicitly installing optimum with GPTQ support (optimum[gptq])
  • Updated installation scripts to ensure proper installation of GPTQ dependencies
  • Improved requirements-all.txt with explicit GPTQ integration for optimum

Added

  • Confirmed compatibility with Python 3.11 and 3.12 on macOS Sonoma 14.2
  • Added detailed platform-specific documentation for macOS, Windows, and Linux
  • Enhanced troubleshooting guide with platform-specific information

[0.3.0] - 2025-03-15

Fixed

  • Fixed dependency installation order issue with gptqmodel requiring torch to be installed first
  • Added installation scripts (install_dependencies.sh and install_dependencies.bat) for reliable dependency installation
  • Updated requirements-all.txt with clear instructions on installation order
  • Improved README.md with detailed installation options and troubleshooting guidance

[0.2.9] - 2025-03-15

Added

  • Created requirements-all.txt for one-shot installation of all dependencies
  • Includes all core, quantization, visualization, and data handling packages
  • Provides a simple way for users to install everything needed with a single command

[0.2.8] - 2025-03-15

Changed

  • Further refined dependencies to absolute minimum required set
  • Moved optimum from core dependencies to gptq extras
  • Improved organization of dependencies in requirements.txt
  • Added detailed comments explaining the purpose of each dependency
  • Grouped GPTQ dependencies together in the extras

[0.2.7] - 2025-03-15

Changed

  • Significantly reduced core dependencies to minimize installation issues
  • Moved non-essential dependencies to optional extras
  • Created new extras: 'viz' for visualization and 'data' for dataset handling
  • Changed gptqmodel dependency to use versions below 2.1.0 to avoid numpy>=2.2.2 requirement
  • Improved requirements.txt with clearer organization and comments
  • Removed torch as a direct dependency to allow more flexible installation

[0.2.6] - 2025-03-15

Fixed

  • Pinned dependency versions to match working Python 3.11 environment
  • Updated torch to version 2.5.1
  • Updated bitsandbytes to version 0.42.0
  • Updated gptqmodel to version 2.1.0
  • Added torchvision and torchaudio as explicit dependencies
  • Added "all" extra in setup.py to install all dependencies at once

[0.2.5] - 2025-03-15

Fixed

  • Added explicit gptqmodel dependency for GPTQ quantization
  • Fixed issue with transformers reporting gptqmodel as available when it's not installed
  • Added proper requirements.txt file with all dependencies
  • Improved dependency management in setup.py

[0.2.4] - 2025-03-15

Fixed

  • Completely redesigned Python 3.12 compatibility for GPTQ quantization
  • Added multiple patching strategies for transformers 4.49.0 compatibility
  • Implemented recursive function scanning to find and patch CUDA checks
  • Added method-level exception handling to bypass GPU requirements on CPU

[0.2.3] - 2025-03-15

Fixed

  • Improved Python 3.12 compatibility for GPTQ quantization
  • Fixed patch targeting for CUDA availability check in transformers
  • Added multiple fallback methods to ensure CPU compatibility

[0.2.2] - 2025-03-15

Fixed

  • Added Python 3.12 compatibility for GPTQ quantization
  • Fixed issue with Optimum's CUDA check in Python 3.12
  • Applied monkey patch to bypass CUDA requirement in newer Python versions

[0.2.1] - 2025-03-15

Fixed

  • Restored GPTQ quantization support for CPU devices
  • Removed incorrect device restriction that was blocking CPU usage for GPTQ
  • Improved error handling with more helpful suggestions

[0.2.0] - 2025-03-15

Added

  • PyPI package support with model-quantizer now available via pip
  • Automatic model card generation when quantizing models
  • Ability to update model cards with benchmark results
  • New command-line tools with consistent naming:
    • model-quantizer: Main tool for quantizing models
    • benchmark-model: Tool for benchmarking models
    • run-benchmark: All-in-one benchmarking solution
    • visualize-benchmark: Tool for creating visual benchmark reports
    • chat-with-model: Interactive testing tool

Changed

  • Renamed benchmark_your_model.py to benchmark_model.py
  • Converted run_benchmark.sh to run_benchmark.py for cross-platform compatibility
  • Updated chat_with_model.py to accept a single model_path parameter
  • Updated setup.py to include necessary dependencies for all command-line tools
  • Consolidated publishing documentation into a single comprehensive guide
  • Improved documentation structure with clearer workflow steps
  • Enhanced examples with more detailed instructions and use cases

Fixed

  • Cross-platform compatibility issues, especially for macOS users
  • Inconsistent command-line interfaces across tools
  • Missing dependencies in setup.py

[0.1.0] - 2025-03-15

Added

  • Initial release of Model Quantizer
  • Support for GPTQ, BitsAndBytes, and AWQ quantization methods
  • Basic benchmarking capabilities
  • Interactive testing via chat interface
  • Documentation for quantizing models
  • Example scripts for common use cases