A high-throughput and memory-efficient inference and serving engine for LLMs
main
[Bugfix] Fix Qwen3-Omni crash on video with no audio track when use_audio_in_video=True (#48420)
Ryan Hamby committed 3 months ago
[Bugfix][Hybrid] Fix cross-block race on num_accepted in MRv2 align prefix cache (#50432)
Francesco Fusco committed 3 months ago
fix: NVFP4 quantization out_dtype should match model dtype, not torch default (#48861)
Chris Fontes committed 3 months ago
feat(frontend): session id plumbing into requests (#48048)
Karen Chung committed 3 months ago
[ModelRunnerV2] Fix scalar Mamba state update with int32 mappings (#50327)
Varun Vinayak Shenoy committed 3 months ago
[ROCm] Enable AITER and FP8 inference on GFX120x (#43615)
skysnow2001 committed 3 months ago
[Bugfix][Model Runner V2] Restore multimodal draft capability detection (#50417)
Raphaël Rialland committed 3 months ago
[MRV2] Enable routed-experts capture (#50721)
aoshen02 committed 3 months ago
[ROCm][Test] Fix AITER MXFP4 oracle contract (#50728)
Andreas Karatzas committed 3 months ago
[Bugfix][Frontend] Reject empty gRPC stop strings (#50746)
zcxGGmu committed 3 months ago
[Hybrid] Stage the postprocess inputs with a single loop over the request list (#48120)
Francesco Fusco committed 3 months ago
[Bugfix] Default Gemma3 Model intermediate_tensors to None (#50777)
Taneem Ibrahim committed 3 months ago
[Bugfix] Emit a valid media type from encode_{audio,image,video}_url (#49056)
Vineeth Sai Varikuntla committed 3 months ago
[ROCm][Kimi-K3] aiter moe environment variable cleanup (#50582)
Hongxia Yang committed 3 months ago
[Bugfix] Validate NIXL speculative config compatibility (#49230)
Tzu-Ling Kan committed 3 months ago
[Kernel] Skip fully masked key blocks in windowed Triton prefill (#50776)
Almog Tavor committed 3 months ago
[CI][ROCm] Export Helion benchmark script in test artifacts (#50726)
Andreas Karatzas committed 3 months ago
[Bugfix] Shard UniformTypeKVCacheSpecs block table width under DCP (#50823)
drakosha committed 3 months ago
[Kimi-K3] Add option to shard the shared expert instead of replicating (#50656)
Tyler Michael Smith committed 3 months ago
[Misc] Remove deprecated calculate_kv_scales runtime KV scale calculation (#49389)
wangxiyuan committed 3 months ago