Patch Release v0.2.6

Updated

• Fixed RoPE type missing problem in deepseek-r1-qwen3-8B model
• Update README with Qwen3 model notes and Transformers compatibility details

Release v0.2.5

Added

• Model server support
• Deepseek model support
• Qwen-3 model support
• Langchain integration
• local RAG example

Updated

• Various refactoring and improvements

Release v0.2.4

Fixed

• Source distribution (was missing requirements.txt)

Release v0.2.3

Added

• Evaluation results

Fixed

• Changelog order and date format
• URL in README for PyPI

Release v0.2.2

Added

• Evaluation results

Fixed

• Version numbering

Release v0.2.1

Added

• Initial support for a classical GPTQ model using the MPQLinear layer
• AutoGPTQ information and commands in the repository
• Support for LoRA and GPTQ evaluation
• SFT comparison updates
• Missing comment to the customized trainer class

Fixed

• Issue in GbaSFTTrainer for saving non-GBA models
• Mismatch issue between GPTQ and LoRA
• Bug preventing quant_strategy.json from being saved during SFT

Updated

• README with new AutoGPTQ and GPTQ support information

Release v0.1.0

Added

• Integration with Bitorch Engine
• Full-parameter fine-tuning and PEFT support
• Fast inference capabilities
• Comprehensive evaluation tools and detailed model evaluation results