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 14, 05:30 UTC
main
fix(v1/kv_cache): resolve async KV transfer bug in cascade attention (#23485)
Ayush Satyam committed about 1 year ago
[Core] Simplify setting new_token_ids in CachedRequestData (#26388)
Nick Hill committed about 1 year ago
Add SwigluOAI implementation for CPUFusedMOE (#26347)
isharif168 committed about 1 year ago
[torchao] safetensors integration (#25969)
liangel-02 committed about 1 year ago
[Misc] Clean up cruft from previous FlashMLA sparse implementation (#26125)
Lucas Wilkinson committed about 1 year ago
Revert #24446 and #26168 (#26332)
Thomas Parnell committed about 1 year ago
[Bugfix] Fix MTP+FlashInfer crash when trtllm kernels are available but disabled (#26361)
Benjamin Chislett committed about 1 year ago
Add more libraries to rlhf.md (#26374)
Michael Goin committed about 1 year ago
[TPU] update TPU benchmark threshold (#25713)
Johnny Yang committed about 1 year ago
[Spec Decode] Enable efficient speculative decoding with FlashInfer-MLA (#25984)
Benjamin Chislett committed about 1 year ago
[V1] Logit processors for rejection sampler (#19482)
Sergei Skvortsov committed about 1 year ago
Morrison Turnansky committed about 1 year ago
[Bugfix] Add missing sink tensor into flash attn cascade attn implementation (#26325)
Pei-Lun Liao committed about 1 year ago
[Bug] Fix Shape Validation for Fallback while Enabling E8M0 for DeepGEMM (#26322)
Wentao Ye committed about 1 year ago
[deepseek] add EP8 FusedMOE config for H200 and B200 (#26331)
Chen Zhang committed about 1 year ago
[CI] Add Qwen3 MoE NVFP4 to Blackwell lm-eval (#26316)
Michael Goin committed about 1 year ago
[ci] Rename `test_mxfp4_moe.py` to `test_ocp_mx_moe.py` (#26364)
fxmarty-amd committed about 1 year ago
Paul Pak committed about 1 year ago
[V0 Deprecation] Remove `VLLM_USE_V1` from tests (#26341)
Cyrus Leung committed about 1 year ago
[Misc] Move `LRUCache` into its own file (#26342)
Cyrus Leung committed about 1 year ago