What's Changed
- Flash attention support for Llama by @dvruette in #2277
- Use fixed RNG seed value for all DeepSpeed workers by @andreaskoepf in #2324
- Show model config in the chat UI by @AbdBarho in #2317
- Populate env vars before render layout by @notmd in #2323
- Update configs according to feedback to #2277 by @dvruette in #2325
- Disable Initial Prompt Task for en and es Locales by @hzj5790 in #1849
- stats are shown on the admin by @vivasvan1 in #2330
- inference: use uuid v7 for most of table by @notmd in #2327
- fix deepspeed issue on trainer_rm.py, add crossentropy support by @theblackcat102 in #2321
- Improve scores of small 1.4B reward model.. by @andreaskoepf in #2329
- Refactor inference backend auth and switch to authlib by @olliestanley in #2318
- add dataset counts script by @CloseChoice in #2294
- fix typos by @RainRat in #2264
- Updated nginx config for prod, including streaming headers by @yk in #2239
New Contributors
- @vivasvan1 made their first contribution in #2330
- @RainRat made their first contribution in #2264
Full Changelog: v0.0.1-beta60...v0.0.1-beta61