Change Log¶

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

The format is based on Keep a Changelog and this project adheres to Semantic Versioning.

Unreleased ¶

Switch build tools to pixi.
Removed zensols.deeplearn dependency.
Upgrade dependencies: torch transformers, datasets, accelerate, peft, trl.
Removed default quantization configuration.
Add LoRA generator configuration.
Separate and configure LoRA adapter output directory.
Add trainer evaluation dataset.
Disable adding end of sentence token in GenerateTask (turn on by setting train_add_eos is True). Now the SFTTrainer does this by default.
InstructTask.apply_chat_template defaults to False. Now the SFTTrainer does this by default.