A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 7, 19:00 UTC
main
[Perf] Optimize Python Slice for Structured Output using `islice` instead of [:] (#33593)
Wentao Ye committed 9 months ago
Harry Mellor committed 9 months ago
[CI] Remove Duplicated Tests (#35199)
Robert Shaw committed 9 months ago
Integrate flashinfer mm_mxfp8 in ModelOpt MXFP8 (#35053)
danisereb committed 9 months ago
Fix fallback to default tactic (flashinfer autotuner) with trtllm_fp4_block_scale_moe (#35088)
danisereb committed 9 months ago
[CPU][Perf] Accelerate Attention head for s390x using vector intrinsics (#34434)
R3hankhan committed 9 months ago
Fix GLM4 parser tests (#34905)
Robin Nabel committed 9 months ago
[Attn,KV-cache] Use per-head scales in the attention selector (#34281)
Eldar Kurtić committed 9 months ago
[Frontend] Always pass `supported_tasks` to validation (#35186)
Cyrus Leung committed 9 months ago
[Bugfix] Fix failing FunASR processor test (#35111)
Isotr0py committed 9 months ago
[glm-asr] change defaults dummy audio size (#35108)
eustlb committed 9 months ago
[Perf] Optimize pooling model redundant copy, 1.8% throughput improvement (#35127)
Wentao Ye committed 9 months ago
[compile] Save aot compile artifacts atomically. (#35117)
Zhengxu Chen committed 9 months ago
[Feature] Add LoRA tower/connector support for Llama 4 Vision (mllama4) (#35147)
Dor Huri committed 9 months ago
Make voxtral compile friendly (#33959)
Tugsbayasgalan Manlaibaatar committed 9 months ago
[LoRA] Update LoRA expand kernel block_n calculation (#32621)
Xin Yang committed 9 months ago
[Refactor] [1/N] Reorganize kernel abstraction directory (#34055)
BadrBasowid committed 9 months ago
[CI/Build] Remove redundant OpenTelemetry pip install from CI configs (#35032)
Vlad Tiberiu Mihailescu committed 9 months ago
[MM] Allow audio chunking for offline LLM (#34628)
Nicolò Lucchesi committed 9 months ago
[BUGFIX][Qwen3.5] Hardcode `mlp.gate` as not quantizable (#35156)
Vadim Gimpelson committed 9 months ago