0.1.8

Stability and documentation

• New version of Documentation is live, thanks to @tushar2407.
• Updated contribution guide.
• Stability improvements and bugfixing.
• int8 generation now works even better.

0.1.7

Evaluation and documentation, CPU support

• Evaluation of finetuned models, thanks to @tushar2407.
• Llama2 model support and enchantments of Generic models.
• Batch size support for both evaluation and generation.
• CPU inference and finetuning.
• Documentation and examples, thanks to @tushar2407.

0.1.6

INT4 models

• INT4 Generic model now supports large variety of open source models finetuning in INT4, thanks to @tushar2407.
• All INT4 finetunings are now much stable and work on large variety of GPUs.
• Support of flags for models loading, thanks to @msinha251.
• Documentation enchantment and bug fix, thanks to @tushar2407.

BaseModel bug fix

• Changing BaseModel to work with LoRA after previous release, thanks to @tushar2407

0.1.4

New models and logging

• Generic model to support large variety of open source models, thanks to @tushar2407
• Falcon model support as a latest open source model release, thanks to @tushar2407
• Documentation enchantment, again, big thanks to @tushar2407
• Logger integration and refactoring, thanks to @Georgehe4

0.1.3

Bugs and fixes

• Fixed issues with INT8 and INT4 model saving and loading after fine-tuning
• Dataset from different sources saving fixes
• Hub loading progress bar fixing, replaced old model formats
• Added generic Wandb integration, thanks to #180 by @cnbeining

Release 0.1.2

Bugs and fixes

• Improved default generation parameters for all models
• Fixed issue of both INT8 and INT4 generation like #167. Added correct model weights mapping for INT4
• Bumped transformers version to 4.28.1
• Fix hub recursive downloads

v0.1.1

Bug fix

• There was an issue with printing parameters of GPT-2 model. Exception handling is added.

v0.1.0

INT4 generation added

• While we already supported INT4 fine-tuning through Docker, it wasn't complete. This release allows you to perform INT4 fine-tuning and INT4 generation as part of the xTuring pipeline