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 10, 06:30 UTC
main
[Kernel][MoE] fix computation order of MoE weight multiplication and improve flow (#31962)
xuebwang-amd committed 10 months ago
[Model Runner V2] Add support for M-RoPE (#32143)
Woosuk Kwon committed 10 months ago
[Model Runner V2] Minor refactor for logit_bias (#32209)
Woosuk Kwon committed 10 months ago
[BUGFIX] Add missed remaping of the names of fp8 kv-scale (#32199)
Vadim Gimpelson committed 10 months ago
[NIXL][Bugfix] Failure logging overhaul + early metadata free on failure (#32031)
Nicolò Lucchesi committed 10 months ago
[Model Runner V2] Support logit_bias, allowed_token_ids, min_tokens (#32163)
Woosuk Kwon committed 10 months ago
[Misc][BE] Type coverage for vllm/compilation [3/3] (#31748)
Lucas Kabela committed 10 months ago
[Misc] Change log level for batch queue log (#32192)
Nicolò Lucchesi committed 10 months ago
[BugFix] scheduler: Fix ordering preserving of skipped requests (#32173)
Or Ozeri committed 10 months ago
[Misc] Set default torch num threads for input processing (#31879)
Roger Wang committed 10 months ago
[Refactor] EPLB rebalance algo to NumPy (#30697)
Ilya Markov committed 10 months ago
[BugFix] fix FusedMoE.make_expert_params_mapping in EXAONE-MoE (#32196)
Kyungmin Lee committed 10 months ago
[3/N][Attention] Move AttentionMetadata-related code from utils.py to backend.py (#32054)
Matthew Bonanni committed 10 months ago
[Benchmark] Share data between SLA runs (#32184)
Cyrus Leung committed 10 months ago
[Misc][PD] Fix `get_attn_backend` usage in transfer connectors (#31988)
Nicolò Lucchesi committed 10 months ago
[Bugfix] Fix stale SSM state for new Mamba requests scheduled as decode (#32118)
Asaf Joseph Gardin committed 10 months ago
[Model] Standardize pooling heads (#32148)
Cyrus Leung committed 10 months ago
[FIX] Add NO_MUL activation support for modular kernel path (#31528)
danielafrimi committed 10 months ago
[MODEL] New model support for kakaocorp/kanana-1.5-v-3b-instruct (#29384)
Jaehyun An committed 10 months ago
Add K-EXAONE-236B-A23B (#31621)
Kyungmin Lee committed 10 months ago