debug run

Author: kkscilife

.github/workflows/e2e_test.yaml

@@ -39,7 +39,7 @@ jobs:
           conda activate clusterx 
           conda env list
           unset HTTP_PROXY;unset HTTPS_PROXY;unset http_proxy;unset https_proxy;
-          pytest autotest/test_all.py -m all -n 1 -vv --run_id ${{ github.run_id }}
+          pytest autotest/test_all.py::test_all[qwen3-sft-celoss] -m all -n 1 -vv --run_id ${{ github.run_id }}
 
       - name: Upload Artifacts
         if: ${{ !cancelled() }}

.github/workflows/e2e_test_npu.yaml

@@ -37,7 +37,7 @@ jobs:
         run: |
           source activate npuci
           unset HTTP_PROXY;unset HTTPS_PROXY;unset http_proxy;unset https_proxy;
-          export DEVICE=npu && pytest autotest/test_all.py -m all -n 1 -vv --run_id ${{ github.run_id }}
+          export DEVICE=npu && pytest autotest/test_all.py::test_all[npu-qwen3-sft-ep8] -m all -n 1 -vv --run_id ${{ github.run_id }}
 
       - name: Upload Artifacts
         if: ${{ !cancelled() }}

autotest/config.yaml

@@ -194,6 +194,8 @@ case:
                 config: autotest/config/qwen3_sft_celoss.py
                 output_path: /mnt/shared-storage-user/llmrazor-share/qa-llm-cicd/test_output
             resource:
+                num_nodes: 2
+                cpus_per_task: 80
                 envs:
                     - QWEN3_MOE_PATH=/mnt/shared-storage-user/llmrazor-share/model/Qwen3-30B-A3B
                     - ALPACA_PATH=/mnt/shared-storage-user/llmrazor-share/data/alpaca