diff --git a/computility-run.yaml b/computility-run.yaml index da12e930..cb8af270 100644 --- a/computility-run.yaml +++ b/computility-run.yaml @@ -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 diff --git a/qwen3_6_scripts/patch_ops.sh b/qwen3_6_scripts/patch_ops.sh index f3482a95..642121b4 100755 --- a/qwen3_6_scripts/patch_ops.sh +++ b/qwen3_6_scripts/patch_ops.sh @@ -164,17 +164,6 @@ install_patch_file \ "${VLLM_OVERRIDE_ROOT}/worker/worker.py" \ "${VLLM_ROOT}/worker/worker.py" -build_stage "installing PR #2269 DP/EP-aware speculative decoding overrides" -install_patch_file \ - "${VLLM_OVERRIDE_ROOT}/spec_decode/spec_decode_worker.py" \ - "${VLLM_ROOT}/spec_decode/spec_decode_worker.py" -install_patch_file \ - "${VLLM_OVERRIDE_ROOT}/spec_decode/multi_step_worker.py" \ - "${VLLM_ROOT}/spec_decode/multi_step_worker.py" -install_patch_file \ - "${VLLM_OVERRIDE_ROOT}/spec_decode/metrics.py" \ - "${VLLM_ROOT}/spec_decode/metrics.py" - build_stage "installing hash-pinned CoreX 3.2.3 extensions" bash ./install_prebuilt_corex.sh "${VLLM_ROOT}"