What's Changed
- other: fix circular import error by @rebel-seinpark in #556
- model: Moe custom op args by @rebel-thkim in #557
- other: skip some pytest which cannot fix in compiler by @rebel-kblee in #561
- model(whisper): allocate extra block for prefill padding (vllm) by @rebel-eunji in #560
- dependency: Update dependency rebel-compiler to 0.10.4.dev201+g84f77eb1.prod by @rebel-thkim in #539
- model: lower minimum flash attention seq len and partition length by @rebel-thkim in #569
- model: vlm's max seq len by @rebel-thkim in #565
- other(test): fix disallowed flash-attn tests after lowered minimums (#569) by @rebel-jongho in #571
- dependency: Update dependency rebel-compiler to 0.10.4.dev321+gc1b63c2d.prod by @rebel-thkim in #568
- other: attention_validation with variable by @rebel-thkim in #577
- release: v0.10.4 by @rebel-thkim in #578
Full Changelog: v0.10.3...v0.10.4