67lc comments

Repositories
Issues
Comments

Results 2 comments of


                                            67lc

[Installation]: When i build vllm from source with pip install -e ,there is a ninja error: unknown target '_vllm_fa3_C', did you mean '_vllm_fa2_C'.

Thank you for you answer @mmdbhs ,but i want to change C++ and CUDA code.This mode is unsuitable with precompiled wheel.

Support disaggregated prefill ?

**Now,I try to use PD disaggregated with 2 nodes. Every node have 8 V100 with NVlinks** **Commands:** master:`CUDA_VISIBLE_DEVICES=0 python -m lightllm.server.api_server \ --model_dir /share/models/meta-llama/Llama-3.2-1B-Instruct \ --run_mode "pd_master" \ --host 10.0.0.103...