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 6, 13:15 UTC
main
[ROCm][Quantization] add fp8xfp8 attn support for rocm_aiter_unified_attn (#36927)
Divakar Verma committed 8 months ago
[Kernel] Add non-gated support for NVFP4 CUTLASS MoE (#37320)
Michael Goin committed 8 months ago
[ROCm] Validate block_size for explicitly selected attention backends (#36846)
Andreas Karatzas committed 8 months ago
JartX committed 8 months ago
Make KV connector metadata build overridable via plugin (#37336)
Yong Hoon Shin committed 8 months ago
[Model] Add ColQwen3.5 4.5B support (#36887)
Athrael Soju committed 8 months ago
Ekagra Ranjan committed 8 months ago
[Bugfix] Rescale NVFP4 weight scales to fix BF16 dequant underflow (#34577)
Chao-Ju Chen committed 8 months ago
Dimitrios Bariamis committed 8 months ago
[Perf] Set Flashinfer sparse MLA as default backend for FP8 kv cache (#37252)
Wei Zhao committed 8 months ago
[Deprecation] Deprecate `--calculate-kv-scales` option (#37201)
Michael Goin committed 8 months ago
[Model] Remove unused `handle_oov_mm_token` (#37321)
Cyrus Leung committed 8 months ago
bump compressed-tensors version to 0.14.0.1 (#36988)
Brian Dellabetta committed 8 months ago
[Bug] Fix FlashInfer MNNVL socket collisions under concurrent vLLM jobs (#36674)
Wentao Ye committed 8 months ago
[Torch 2.11] Guard torch._C._cpu attribute checks for forward compatibility (#35673)
Andrey Talman committed 8 months ago
[CI] Split Distributed Tests (4 GPUs) and Kernel MoE tests (#37100)
Avinash Singh committed 8 months ago
[Perf] Optimize top-k search in apply_top_k_top_p_triton sampler (#37225)
Michael Goin committed 8 months ago
[CI] Fix GPU memory leak when RemoteOpenAIServer fails to start in __init__ (#37230)
Andreas Karatzas committed 8 months ago
[Bugfix] Standardize custom HF Processor init (#37289)
Cyrus Leung committed 8 months ago
[BugFix] PyTorch Compilation Tests should error if any test fails (#37300)
Richard Zou committed 8 months ago