A high-throughput and memory-efficient inference and serving engine for LLMs
main
[ROCm][Perf] Extend QK-norm/RoPE/KV-cache fusion to MRoPE (#50212)
vorapolsiloai committed 29 days ago
[Quark] Remove quark-specific silent online quantization (#51800)
fxmarty-amd committed 29 days ago
[Quantization][Bugfix] Bump humming-kernels to 0.1.16 (#58054)
Jinzhen Lin committed 29 days ago
[Quantization] Enable humming wNaM asymmetric quant (zero_point) with compressed-tensors (#46528)
HDCharles committed 29 days ago
[CPU] Adds support for fp32 attention sinks (#56252)
Ankit-Jaiswal-AMD committed 29 days ago
[Perf][MoE] Skip top-k slots routed to non-local experts in TritonExp… (#58051)
Shuolei Wang committed 29 days ago
[Feature][Frontend] Add DeepSeek-V4 FIM completion rendering (#44229)
jack committed 29 days ago
[DSpark] Support pipeline-parallel targets in aggregated serving (#56956)
Gabriel Wu committed 29 days ago
[Bugfix][Tool Parser] Migrate Granite to the streaming Parser Engine (#49648)
Nikhil Kulkarni committed 29 days ago
[Bugfix] Disable prefix caching for encoder-only before model config hooks (#58287)
Tianyu Guo committed 29 days ago
[ROCm][Test] Check GDN prefill numerics and output ownership (#58091)
Andreas Karatzas committed 29 days ago
[XPU] upgrade to PyTorch 2.14 (#56013)
Yan Ma committed 29 days ago
Gabriel Wu committed 29 days ago
[MRV2] Miscellaneous code cleanup (#57980)
Nick Hill committed 29 days ago
[ROCm][CI] Include Python tooling in ROCm CI artifacts (#58271)
Andreas Karatzas committed 29 days ago
[ROCm][Bugfix] Keep zero MiniMax MXFP8 activation blocks finite (#58089)
Andreas Karatzas committed 29 days ago
[Perf][ROCm][Attention] Narrow the Triton prefill-attention KV tile on RDNA3/RDNA4 (#58225)
lijipeng committed 29 days ago
[CI][Bugfix] Extend groupwise rms_norm scale tolerance to CUDA (#58252)
vllm-agent committed 29 days ago
[Perf][Attention] Avoid CPU-GPU sync in DCP sequence lengths (#58169)
Jiangyun Zhu committed 29 days ago
[ROCm][Bugfix] Register MRV2 sampler JIT warmups (#58092)
Andreas Karatzas committed 29 days ago