关联 Issue #25749 [RFC]: Support Prefill Context Parallel (PCP) #43809 [DeepSeekV4][PCP] Enable context-parallel prefill for hybrid KV and MegaMoE #49109 [Core][KV Connector][Mooncake] Support MRV2 prefill PCP in Mooncake #60423 [Bugfix][MoE] Size Humming scratch for PCP #60447 [Bugfix][MoE] Size the permute scratch from its inputs #60946 [Bugfix][PCP] Align MoE routing padding with gathered tokens #1 [DSv4.1][PCP] Compress ratio-2 groups rank-locally (proposal for vllm-project/vllm#59857)
参与讨论