[Feature][Frontend] Expose effective attention block size for DCP (#56538)

JulienDarve committed about 1 month ago

[Bugfix][Frontend] Lazy-import model_hosting_container_standards to prevent log suppression (#51366)

wenjinhust committed about 1 month ago

[Parser] Fix: correct parser frontend handling of reasoning end and boundary tokens (#56635)

dev committed about 1 month ago

fix(multimodal): tolerate malformed EXIF metadata during hashing (#56527) (#56576)

Ama Senevirathne committed about 1 month ago

[Bugfix][CPU] Add per-tensor FP8 W8A16 kernel to fix Ministral crash on CPU (#56985)

Li, Jiang committed about 1 month ago

[Bugfix][KV Offload] Track cache recency once per request (#51787)

MINJUN GIL committed about 1 month ago

[ROCm][AITER] Skip AITER norm kernels when flattening to 2D would copy (#55991)

Zheng Gong committed about 1 month ago

[Bugfix] Fix incorrect Mamba block allocation estimate that prevents request admission (#57050)

Wei Zhao committed about 1 month ago

[Kernel] Use Murmur3 RNG for Gumbel sampling (#51367)

freyfwt committed about 1 month ago

[Bugfix][KV Offload] Check cgroup memory before SHM allocation (#54014)

AlexHuang committed about 1 month ago

[Bugfix][Responses] Clean up MCP tool sessions once before closing them (#56988)

shaohuaxi committed about 1 month ago

[ROCm][CI] Fix MLA RoPE fused-kernel tests for TheRock image (#57112)

Matt committed about 1 month ago

[Bugfix] Unreadable `prompt_embeds` payload should be a 400, not a 500 (#55451)

lzhan011 committed about 1 month ago

[Refactor][GLM-5.3-Flash] Move sparse_attn_indexer_kpool into the model folder and split AMD/NVIDIA (#55358)

Jiangyun Zhu committed about 1 month ago

[Bugfix][KV Offload] Restore MTP-retained sliding-window history (#56709)

Liqian Lin committed about 1 month ago

[CI][ROCm] Add opt-in TheRock builds for AMD CI (#56351)

Andreas Karatzas committed about 1 month ago

[Bugfix] Make GPU sync checks safe under torch.compile (#56904)

Kevin H. Luu committed about 1 month ago

[Perf][DSpark] Add KV-only context insertion across V4.1 cache formats (#56441)

0z5a committed about 1 month ago

[LoRA][Nemotron] Add LoRA support for Nemotron VL models (for the language model only) (#56231)

danisereb committed about 1 month ago

[KV Offload] Add per-request `max_load_tokens control` (#55885)

alberto committed about 1 month ago