A high-throughput and memory-efficient inference and serving engine for LLMs
main
xx-thomas committed 5 months ago
[DSV4][Minor] Fix supported KV cache dtypes (#44892)
Woosuk Kwon committed 5 months ago
[EP] Enable DBO with NIXL EP (#45275)
Itay Alroy committed 5 months ago
[EP] Query NIXL EP top-k index dtype (#45298)
Itay Alroy committed 5 months ago
nixl_ep: Skip post-receive quantization for NVFP4 (#45606)
Itay Alroy committed 5 months ago
llx committed 5 months ago
[Frontend] Replace legacy Gemma4 parsers with engine-based implementation (#45588)
Ben Browning committed 5 months ago
Fix parallel_tool_calls: null treated as false instead of default true (#44955)
Zang Peiyu committed 5 months ago
[Model Runner V2][Bugfix] Fix MRV2 LoRA warmup (#35536)
Jee Jee Li committed 5 months ago
[Frontend] Skip structural tags for auto tool_choice without strict mode (#45600)
Flora Feng committed 5 months ago
[Perf] Optimize DSv4 prefill chunk planning, 4.0% E2E Throughput Improvement (#45061)
Wentao Ye committed 5 months ago
[Bugfix] Chat Completions Harmony Refactor Clean up (#45464)
yzong-rh committed 5 months ago
[KV Offloading] Implement `reset_cache` for `TieringOffloadingManager` (#44541)
Ronen Schaffer committed 5 months ago
[Model] Add MiniMax M3 support (#45381)
youkaichao committed 5 months ago
[AMD][Bugfix][Quantization] Honor fused-name match in is_layer_skipped (#43981)
RoyWang committed 5 months ago
[Chore] Consolidate reasoning/tool parser attributes into unified Parser in chat serving (#45548)
Flora Feng committed 5 months ago
[Bugfix] Two-phase KV allocation for cross-group prefix cache hits (supersedes #33775) (#44409)
Saddss committed 5 months ago
Mike G committed 5 months ago
Fix the E8M0 scale computation in the MXFP4 (W4A4) MOE CUTLASS kernel (#43557)
Xin He committed 5 months ago
(security) Enforce audio upload size limit before full file materialization (#45510)
Juan Pérez de Algaba committed 5 months ago