A high-throughput and memory-efficient inference and serving engine for LLMs
main
[Bugfix][V1] Reject encoder-cache hits with mismatched embedding counts (#57696)
jackLei committed 28 days ago
[CPU] Use pre-built triton (#58140)
Li, Jiang committed 28 days ago
[Feature][Frontend] Request JSON body debug logging on `--enable-log-requests` flag (#58163)
Talor Abramovich committed 28 days ago
[Perf] Use Conv3dLayer for MiniMax M3 patch embedding (#58512)
Thien Tran committed 28 days ago
[EPD][Model Loader] Skip language-model checkpoint shards for `--mm-encoder-only` (#58086)
GuorongYe committed 28 days ago
[ROCm][CI] Add the MI355 TurboQuant t3nc mirror (#58432)
Andreas Karatzas committed 28 days ago
[CPU][GDN] Support NIXL DS convolution-state layout (#53300)
Tianmu Li committed 28 days ago
[Compile][CI] Honor Triton cache overrides and add AMD timeout headroom (#58474)
Andreas Karatzas committed 28 days ago
[Rust Frontend] Support `--sse-keep-alive-interval` (#58306)
Bugen Zhao committed 28 days ago
[Rust Frontend] Pass vision preprocessing context for Nemotron-H (#57634)
Biswa Panda committed 28 days ago
[Bugfix][Spec Decode] Separate DSpark width from MTP stage validation (#54631)
lucamotz committed 28 days ago
[Dependency] Upgrade FlashInfer version to 0.7.0 (#58069)
Wei Zhao committed 28 days ago
[Rust Frontend] Accept custom chat roles for HF templates (#58311)
Bugen Zhao committed 28 days ago
[Rust Frontend] Recognize new frontend-owned serve args as unsupported or no-op (#58330)
Bugen Zhao committed 28 days ago
[Bugfix][CPU][MoE] Fix out-of-bounds write and segfault when router weights are fp32 (#56168)
Farzad Abdolhosseini committed 28 days ago
Fix full logprobs in token-in/token-out responses (#58488)
aoshen02 committed 28 days ago
[Bugfix][KV Offload] Retain offload event metadata through batch translation (#57453)
Kapil Arya committed 28 days ago
[Kernel] Resubmit PR 48666 - Gemma4 FP8 KV FA4 head dim 512 backend selection (#53175)
Jhao-Ting Chen committed 28 days ago
[ROCm] Give turboquant boundary layers a layout-compatible backend (#54988)
stefankoncarevic committed 28 days ago
[PCP] Support prefill context parallelism with data parallelism (#57075)
Lucas Wilkinson committed 28 days ago