Skip to content

v1.4.0

Choose a tag to compare

View all tags
@brunoasm brunoasm released this 18 Aug 22:10
· 33 commits to main since this release
v1.4.0
1fca263
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Release Notes v1.4.0

New Features

CPU Training Support

  • Add CPU training option: New --cpu/-C flag allows forcing CPU training instead of GPU for environments without CUDA support
  • Minimum batch size control: New --min-batch-size/-B option to set minimum batch size (default: 1) to avoid bugs with small sample sizes
  • Improved device consistency: Enhanced device placement logic ensures all tensors (model, data, pretrained weights) use consistent device allocation
  • Automatic FP16 handling: FP16 training is automatically disabled when using CPU mode

Bug Fixes

Environment & Dependencies

  • Fixed conda environment naming: Corrected conda environment name inconsistencies on macOS systems
  • Updated fastp version: Bumped fastp version in conda environments to prevent varKoder Docker image compatibility issues
  • Fixed documentation: Corrected parameter documentation in query.md to reflect actual command-line parameters (fixes #23)

Code Quality & Maintenance

  • Migrated from setuptools: Modernized package configuration by migrating away from deprecated setuptools usage
  • Enhanced error handling: Improved device consistency and error handling for CPU-only training scenarios

📚 Documentation Updates

  • Enhanced training documentation: Added comprehensive examples and explanations for new CPU training and batch size options
  • Fixed parameter references: Updated query documentation to match actual CLI parameters

Full Changelog: v1.3.1...v1.4.0

Assets 2
Loading