Highlights:
- Llama Trainer has been renamed to Hugging Face TRL trainer and now supports many model architectures
- You can now access a web-based version of the app at the root endpoint of the API (e.g. localhost:8338)
- Support for a number of new models including ServiceNow-AI/Apriel-5B-Instruct (in gallery)
- Plugins now have their own virtual environment (making dependency management easier going forward)
Numerous bug fixes including:
- fix for “DB lock” issues on system with slower I/O
- fix auto-indexing for documents uploaded to rag folder
github log
- Show error when downloading unauthorized HF model. by @dadmobile in #363
- Simplify command to publish webapp zip file on release. by @dadmobile in #362
- Add useEffect to select the first available workflow by @aliasaria in #367
- Add support for GPT 4.1 models by @deep1401 in #366
- Create a tar.gz file instead of a zip when building the app by @dadmobile in #368
Full Changelog: v0.13.1...v0.13.2