[ROCm][Perf] Enable layer-aware CSA2 multi-stream overlap for DeepSeek-V4.1-Flash (#57407)

Shanshan Shen committed 24 days ago

[CI] Split (H200 MIG 35GB) Spec Decode Speculators + MTP into 4 named jobs (#57237)

Thang Nguyen committed 25 days ago

[Perf][Attention] Remove D2H sync from FlashInfer SM90 sparse MLA plan under async scheduling (#58684)

Nicolò Lucchesi committed 25 days ago

[Attention][CPU] Use zentorch SDPA for CPU MLA prefill (#54967)

Rakul-Chauhan committed 25 days ago

[Perf][DSv4.1] Fuse small-batch WO-A with inverse RoPE and MXFP8 quant on SM100/SM103 (#58634)

Juntian Liu committed 25 days ago

[Bugfix] V1: clear stale allowed_token_ids mask in InputBatch.condense (#43931)

Varshith committed 25 days ago

[Bugfix][NIXL] Release a dead peer's NIXL state without waiting for TTL (#50047)

YannikHinteregger committed 25 days ago

[ROCm][Kimi-K3] Make VLLM_ROCM_USE_AITER_MOE_SITUV2 select a4w4/a8w4/a16w4 (#58201)

Hongxia Yang committed 25 days ago

[Bugfix] Disable sequence parallelism / async TP under batch invariance and add a TP regression test (#56377)

Lio Einaudi committed 25 days ago

[Qwen3.8-Flash-Next] Avoid memory fragmentation in QSA indexer logits workspace (#57105)

Thien Tran committed 25 days ago

[mooncake] support CUSTOM_MEM_POOL in vllm (#49300)

void committed 25 days ago

[Metrics][KV Offload] Add Prometheus metrics for SimpleCPUOffloadConnector (#57251)

mevince committed 25 days ago

[Kernel] Bump FlashKDA to keep the recurrent state in fp32 (#58846)

simon-veitner-redhat committed 25 days ago

[PCP][DCP] Support DCP target model with non-DCP Dspark (#56723)

Lucas Wilkinson committed 25 days ago

[Kernel][DSV4.1] Fuse MoE finalize into the TP all-reduce + mHC boundary (#58586)

Yongye Zhu committed 25 days ago

[Security] Harden message sanitization (#58832)

Cyrus Leung committed 25 days ago

[Bugfix][DSV4.1] Avoid host sync in ViT CUDA graph replay metadata (#58499)

Kevin H. Luu committed 25 days ago

[CI] Stabilize batch submission in full CUDA graph tests (#58810)

Andreas Karatzas committed 25 days ago

[GLM5.3 Bug] Fix sparse indexer attn topk backend selection (#58594)

Wentao Ye committed 25 days ago

[CI] Reduce CUDA graph mode test overhead (#58749)

Misha Goin committed 26 days ago