12月1月2月3月4月5月6月7月8月9月10月11月
周一周三周五
一年内 264232 次贡献
更少更多
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 mrv2-ci-test 分支95ef052724 Merge branch 'main' into mrv2-ci-test
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 moondream2 分支a7ca0cc47f Merge branch 'main' into moondream2
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 mnnvl_kv_transfer 分支5b4f6d5284 Use cumem allocator for the KV cache by default
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 mla_decode_any_head 分支32e4481626 [Attention] MLA - cutlass decode with unresticted num_heads
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 mla_cuda_graphs 分支比较 10 提交 »0a02744dc8 fix TP984ffddda6 add cuda graph support to triton_mla attention135c404fbb review comments7241acbd64 review comments2b140debbb Update vllm/config.py
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 minus_x 分支f8768f5244 Remove executable flag on a few files
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 minimax-m3-perf 分支比较 10 提交 »8b00f41237 [MiniMax-M3] Retune decode indexer launch grid for occupancy (#130)0586a20184 Add MiniMax-M3 modelopt nvfp4 support (#46380)7a672233eb Bump flashinfer-jit-cache to 0.6.13rc2ea890c8066 remove contiguous constraint for prefill_topk8717eccaf4 Fix non-contiguous topk buffer for MSA sparse GQA prefill (#105)
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 migrate-safe-jobs-to-h200-mig 分支比较 2 提交 »154aa68e32 Fix: keep PYTORCH_NVML_BASED_CUDA_CHECK=1 on MIG35c0231e94 [CI] Migrate 5 verified jobs to h200_18gb MIG + add MIG detection
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 migrate-gpu1-to-h200-18gb-remaining 分支ee83fe043f [CI] Migrate all remaining gpu_1_queue jobs to h200_18gb MIG
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 mgoin/fix-piecewise-capture-attention 分支比较 4 提交 »a6c0906087 [Bugfix] Skip generic attention during the PIECEWISE capture warmup too54f7c3e458 Update cudagraph_utils.py2198bfa09e Update cudagraph_utils.py1ad59bac73 Merge branch 'main' into mgoin/fix-piecewise-capture-attention