What's Changed
- Fix compile in HunyuanVideo example by @eppaneamd in #556
- Enable AITER in USP code path by @avjves in #560
- feat: Add Apple Silicon (MPS) support by @haqatak in #559
- Fix broken hybrid attention code path by @avjves in #564
- Add possibility to call AttnType.Aiter by @kTorp in #562
- Refactor packages info to remove conflicting logging by @avjves in #571
- Add NPU support for one model in single card by @ChenTaoyu-SJTU in #566
- Fix envs to recognize AMD devices again by @avjves in #577
- Add workaround for default attn_type by @avjves in #578
- Upgrade Flux to new diffusers format by @avjves in #580
- Upgrade Hunyuanvideo to use the new diffusers format by @avjves in #582
- Fix mask handling for batch generation in HunyuanVideo example by @tjkemp in #546
- Add support for Wan2.X I2V models by @avjves in #583
- Add ComfyUI plugin info to README by @feifeibear in #587
- Add joint tensor and KV cache support to USP method by @avjves in #586
- Bump version to 0.4.5 by @feifeibear in #588
New Contributors
- @eppaneamd made their first contribution in #556
- @avjves made their first contribution in #560
- @haqatak made their first contribution in #559
- @kTorp made their first contribution in #562
- @ChenTaoyu-SJTU made their first contribution in #566
- @tjkemp made their first contribution in #546
Full Changelog: 0.4.4...0.4.5