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 prometheus-cudagraph-pct 分支119ccde424 [Metrics] Add Prometheus counter for CUDA graph iteration mode
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 pr-44891 分支比较 7 提交 »6d32045b96 Address review: use registered envs.VLLM_DISABLE_PUSH_ALLREDUCE41f6a7664a Address review: register VLLM_DISABLE_PUSH_ALLREDUCE in envs.pyd35dfbb2e3 Address review: add architecture-specific threshold selection66eb0286ed Address review: add type annotation and cleanup for push_ar_comm7b879d0a98 Address review: bind test sockets to localhost instead of all interfaces
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 pil_image 分支比较 2 提交 »f96a3cc713 test32c0155774 initial
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 perf/push-allreduce-2buffer 分支比较 9 提交 »fa6b6a83ec Apply pre-commit formatting and use torch.accelerator APIfd44100bb0 Address review: add push_ar to benchmark and backend loggingbc50e5fc2e Address review: use registered envs.VLLM_DISABLE_PUSH_ALLREDUCE625bea3939 Address review: register VLLM_DISABLE_PUSH_ALLREDUCE in envs.py17488a743f Address review: add architecture-specific threshold selection
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 pd_scheduling 分支161010c384 Initial stubs for P/D scheduling changes
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 pcp-alt 分支a97bd607af [Core] Add Prefill Context Parallelism (PCP) and All-to-All DCP communication
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 optimize-prefix-caching-scheduling 分支d5bf492f16 Merge branch 'main' into optimize-prefix-caching-scheduling
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 openai226 分支比较 2 提交 »b7a9ca4ebf 2.24.0e9b140aea6 upgrade openai
- frostbyte_neo 推送了仓库 github-featured/vllm-project--vllm 的 njhill/extensible-kv-cache 分支比较 10 提交 »d7f10f2407 [CI] Stop pinning gpu_memory_utilization in tests that do not need it2b19b5793a [Core] Size the KV cache from measured memory, and use it by defaultf67507803d [Core] Extensible KV cache: make the ROCm path safe58d824e804 [Core] Move extensible KV cache logic out of the central files01585efef1 [ModelRunner V2] Reserve encoder cache budget during V2 memory profiling