Highlights
- Qwen 3 Support!
- Updated Llama.cpp fixes installation issues for some users and now runs most recent models (including Gemma 3 and Qwen 3)
- Mac specific GPU metrics on Computer page supplied by MacMon library
Many bug fixes including:
- fixed inaccurate token count in interact tab
- cloud-hosted apps now automatically look for API server on the origin machine, not localhost
- filter plugins to those supported by server architecture
- option to delete models from Hugging Face cache when removing from Transformer Lab
- times show up correctly in train/eval tabs when server and app are in different timezones
What's Changed
- Show Dataset Name on Data card if available by @deep1401 in #414
- Have the Cloud App look for the API on the current server by @aliasaria in #396
- Add MacMon Specific Metrics by @deep1401 in #400
- Add filtering based on the supports field for interact in loader plugins by @deep1401 in #399
- Fix Token Count in Interact tab by @deep1401 in #397
- Replace workflow Run display with a table by @aliasaria in #401
- Auto update plugins by enabling a setting in the Settings tab by @deep1401 in #391
- Improve the look of workflow nodes by @aliasaria in #402
- Fix start time to not show future times by @deep1401 in #404
- Fix the app restart on crash by @deep1401 in #405
- Add graph for VRAM usage in header for MLX by @deep1401 in #411
- Delete models from HF cache by @deep1401 in #410
Full Changelog (part 1): v0.14.1...v0.14.2
Full Changelog (part 2): v0.14.2...v0.15.0