What's Changed
- Minimum of torch 2.2.0 during build by @casper-hansen in #649
- post release 1 by @casper-hansen in #650
- Post release 2 - All additional packages goes into extras by @casper-hansen in #653
- Fused attention: Switch to Flash Decoding by @casper-hansen in #656
- Add EXAONE support by @lgai-exaone in #651
- fix exaone import by @casper-hansen in #659
- fix "Expected all tensors to be on the same device" by @casper-hansen in #664
- multi-gpu fix by @casper-hansen in #668
- Fix missing embed_tokens by @casper-hansen in #671
- install hub main branch by @casper-hansen in #672
- pin huggingface_hub by @casper-hansen in #673
- bump to 0.2.7.post3 by @casper-hansen in #674
- improve type hinting and fix use_cache by @casper-hansen in #680
- fix bug when using FSDP by @kaixuanliu in #696
- Added DeepSeek V3 support. by @LagPixelLOL in #688
- add ability to define torch_dtype by @casper-hansen in #697
- automatically load dtype from config by @casper-hansen in #698
- Enable triton on XPU devices by @Egor-Krivov in #695
- fix workflow build by @casper-hansen in #699
- bump to 028 by @casper-hansen in #700
New Contributors
- @lgai-exaone made their first contribution in #651
- @kaixuanliu made their first contribution in #696
- @LagPixelLOL made their first contribution in #688
- @Egor-Krivov made their first contribution in #695
Full Changelog: v0.2.7...v0.2.8
Contributors
Egor-Krivov, kaixuanliu, and 3 other contributors