Fix model unittest CI

Author: huanghengheng

.github/workflows/model-unittest-gpu.yml

@@ -55,7 +55,7 @@ jobs:
         run: |
           container_name=${TASK}-$(date +%Y%m%d-%H%M%S)
           echo "container_name=${container_name}" >> "$GITHUB_ENV"
-          docker run -d -t --gpus all --name ${container_name} --net=host -v /dev/shm:/dev/shm --shm-size=32G \
+          docker run -d -t --gpus all -it --name ${container_name} --net=host -v /dev/shm:/dev/shm --shm-size=32G \
             -v $work_dir/../../..:$work_dir/../../.. \
             -v $work_dir:/workspace \
             -v /home/.cache/pip:/home/.cache/pip \
@@ -117,6 +117,7 @@ jobs:
           echo "Check models:"
           ls -l ./models
           echo "Test Start"
+          hostname
           timeout 30m bash scripts/regression/ci_model_unittest.sh ${paddle_whl}
           '
           

scripts/regression/ci_model_unittest.sh

@@ -109,6 +109,9 @@ if [[ ${FLAGS_enable_CI} == "true" ]] || [[ ${FLAGS_enable_CE} == "true" ]];then
     echo "Check docker Cuda Version"
     nvcc -V  
     cat /usr/local/cuda/version.txt
+    echo "Check nvidia-smi"
+    nvidia-smi
+    python -c "import paddle; print(paddle.device.get_device_count())"
     PYTHONPATH=$(pwd) \
     COVERAGE_SOURCE=paddleformers \
     python -m pytest -s -v ${model_unittest_path} > ${log_path}/model_unittest.log 2>&1