diff --git a/computility-run.yaml b/computility-run.yaml index 67e274d3..cb8af270 100644 --- a/computility-run.yaml +++ b/computility-run.yaml @@ -13,7 +13,7 @@ command: - '0.92' - --trust-remote-code - -tp - - '2' + - '4' - --max-num-seqs - '2' - --disable-log-requests @@ -35,6 +35,8 @@ command: env: - name: VLLM_ENGINE_ITERATION_TIMEOUT_S value: 3600 + - name: VLLM_ENABLE_EXPERT_PARALLEL + value: 1 - name: BI100_MAX_NUM_SEQS value: 2 - name: BI100_MOE_COREX_DIRECT_ROUTED