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 13, 12:00 UTC
main
[Attention] Refactor CUDA attention backend selection logic (#24794)
Matthew Bonanni committed about 1 year ago
[CI] Add mergify rules for `nvidia` label (#28417)
Michael Goin committed about 1 year ago
Multi turn benchmark progress bar for synthetic conversation generation (#28394)
Ido Segev committed about 1 year ago
[Model][Qwen3VL] Slighly speedup `fast_pos_embed_interpolate` (#28434)
Lukas Geiger committed about 1 year ago
[BugFix] 'DeepseekV2Config' object has no attribute 'use_mla'` (#28387)
Fanli Lin committed about 1 year ago
[Bugfix] fix qwen3-next crash (#28202)
Jiangyun Zhu committed about 1 year ago
[EPLB] Refactor balance_packing to use numpy and optimize GPU-CPU transfers in EPLB (#28369)
Sage Moore committed about 1 year ago
[Bugfix] Fix max image size for PaddleOCR-VL (#28442)
Roger Wang committed about 1 year ago
[BugFix] Avoid calling KV connector layer APIs when metadata is unset (#28253)
David Ben-David committed about 1 year ago
[Doc] Sleep mode documentation (#28357)
iAmir97 committed about 1 year ago
[Bugfix] Fix Stream Sync for Shared Expert Overlap (#28430)
Robert Shaw committed about 1 year ago
Zuyi Zhao committed about 1 year ago
[Misc] FlattenLogprobs -> FlatLogprobs (#28335)
Zhuohan Li committed about 1 year ago
Fix rotary embedding benchmark script (#28323)
Xin Yang committed about 1 year ago
Only register rocm_aiter_ops if aiter is found (#28428)
Michael Goin committed about 1 year ago
[Feature] Add env var `VLLM_MOE_USE_DEEP_GEMM` (#28422)
Wentao Ye committed about 1 year ago
[CI/Test Fix] Fix CP tests on Blackwell (#28404)
Lucas Wilkinson committed about 1 year ago
[Feature] Refactor batch invariant fp8 DeepGEMM (#27606)
Wentao Ye committed about 1 year ago
[Test] Remove old non-varlen FA2 test (#28420)
Matthew Bonanni committed about 1 year ago
[Bugfix] Ensure calculated KV scales are applied in attention. (#27232)
Adrian Abeyta committed about 1 year ago