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 264223 次贡献
更少更多
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 releases/v0.13.0 分支比较 10 提交 »72506c9834 Check for truthy `rope_parameters` not the existence of it (#30983)b2eb84de77 [Bugfix] Remove `tile_size=64` for mm_prefix triton attention (#30973)ac43367ced adds jais 2 support (#30188)30fe765e9f [Fix][FlexAttention] return max logical block index to handle reused blocks (#30915)2c0ee0fde8 [BugFix] Partial revert of #29558 (DeepEP HT + PIECEWISE CG support) (#30910)
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 releases/v0.12.0 分支比较 10 提交 »4fd9d6a85c [Core] Rename PassConfig flags as per RFC #27995 (#29646)a1d627e40f [BugFix] Fix assert in `build_for_cudagraph_capture` (#29893)2f055ec1c1 [Bugfix] Fix incorrect channel order for idefics3 in edge case (#29881)6a6108511f [BUGFIX] Fix regex pattern for Mistral Tool Call (#29918)9057fc2f1b [BUGFIX] llama_4_scaling wrongly passed to DeepseekAttention (#29908)
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 releases/v0.11.2 分支比较 4 提交 »275de34170 [BugFix] Fix false assertion with spec-decode=[2,4,..] and TP>2 (#29036)fa3ffb4365 [BugFix] Ray with multiple nodes (#28873)6d5974369c [BugFix] Fix async-scheduling + FlashAttn MLA (#28990)0ce9990d2c [NVIDIA] Guard SM100 CUTLASS MoE macro to SM100 builds v2 (#28938)
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 releases/v0.11.0 分支比较 10 提交 »b8b302cde4 Update CUDA architecture list in build pipeline for 12.9.1 wheels (#26592)f71952c1c4 [Build/CI] Revert back to Ubuntu 20.04, install python 3.12 with uv (#26103)d1007767c5 [Bugfix] Disable cascade attention with FlashInfer (#26130)c75c2e70d6 [Deepseek v3.2] Support indexer prefill chunking (#25999)9d9a2b77f1 [Small] Prevent bypassing media domain restriction via HTTP redirects (#26035)
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 releases/v0.10.2 分支比较 8 提交 »01efc7ef78 [ci] fix wheel names for arm wheels (#24898)26b999c71a [CI Failure] Fix test_flashinfer_cutlass_mxfp4_mxfp8_fused_moe (#24750)da3fa78dc9 [Compilation Bug] Fix Inductor Graph Output with Shape Issue (#24772)bbb70036cb Enable conversion of multimodal models to pooling tasks (#24451)89da8d9d09 [Qwen3Next] Fixes the cuda graph capture conditions under large batch sizes (#24660) (#24667)
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 releases/v0.10.1 分支比较 6 提交 »1da94e673c Do not use eval() to convert unknown types (#23266)d8b736f913 Limit HTTP header count and size (#23267)3a8708f60a [BugFix] fix CUTLASS MLA full cudagraph (#23200)aab549870d Use Blackwell FlashInfer MXFP4 MoE by default if available (#23008)ba6928cf13 fix: OpenAI SDK compat (ResponseTextConfig) (#23126)
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 releases/v0.10.0 分支bcc0a3cbef fix: do not install lmcache as it overrides torch version