Hunan University · Changsha
Working on LLM inference system optimization, currently KV cache transfer and RDMA for disaggregated serving.
Upstream
- vLLM #65593 Make EC connector control-plane ZMQ heartbeats take effect
- vLLM #65575 Fail Mooncake pulls from a lost prefill worker
- vLLM #64193 Decoder-renewed KV leases for the Mooncake connector
- vLLM #62478 Faster Mooncake transfer descriptor build
- SGLang #47100 Fail parked prefill requests when their decode goes away
