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 3, 13:00 UTC
main
[ROCm] Restore fast top_k_per_row kernels for sparse MLA when topk_tokens=2048 (#42072)
frida-andersson committed 6 months ago
[Bugfix] Fix SM121 (DGX Spark) exclusion from Marlin/CUTLASS FP8 paths (#35568)
Blake Ledden committed 6 months ago
[LMCacheMPConnector] Prioritize importing the lmcache_mp_connector from lmcache (#42596)
chunxiaozheng committed 6 months ago
[ROCm] Widen AITER fused AR RMSNorm 1-stage gate (#42409)
akii96 committed 6 months ago
[Model Runner v2] Support reload weights (sleep mode) (#42673)
Wentao Ye committed 6 months ago
[Model Runner V2] FP32 gumbel sampling. (#41775)
PatchyTIS committed 6 months ago
[ROCm][MLA] FP8 ASM prefill for AITER dense MLA backend on gfx950 (#42509)
Markus Hartikainen committed 6 months ago
[Perf] Set IR Op Priority Once at Worker Init (#42631)
BadrBasowid committed 6 months ago
[Bugfix] Fix DeepGEMM context lens contiguity in MLA indexer (#42135)
Mohammad Miadh Angkad committed 6 months ago
[Perf] Optimize MLA attention `_v_up_proj` bmm by removing additional copy (#42561)
Wentao Ye committed 6 months ago
[Model Runner V2] Fix kv_connector `pre_forward` order (#42676)
Wentao Ye committed 6 months ago
[ROCm] Widen OAI Triton MoE capability range to include gfx12 (RDNA4) (#37826)
L.B.R. committed 6 months ago
[Feat][RL] IPC weight sync optimizations: multigpu support and chunked packed tensors (#37476)
Aaron Hao committed 6 months ago
[Bugfix] DFlash FP8 KV-Cache (#42692)
Benjamin Chislett committed 6 months ago
[Bugfix] Ensure embeding model compilation on CPU (#42709)
Li, Jiang committed 6 months ago
gemma3 multi-gpu bug-fix (#42630)
pmaybank committed 6 months ago
fix: propagate revision/code_revision pins to all artifact boundaries (#42616)
Juan Pérez de Algaba committed 6 months ago
[ROCm][CI] Stage B gating (#42025)
Andreas Karatzas committed 6 months ago
DeepSeekV4-Pro enable cuda graph full and piecewise mode (#42604)
bobofang_amd committed 6 months ago
[Entrypoints] Split the pooling offline API into PoolingOfflineMixin. (#42267)
wang.yuqi committed 6 months ago