PyCharm运行代码报DNN library is not found错误，终端执行正常

问题背景 #

执行pip install waymo-open-dataset-tf-2-6-0 --user安装Waymo数据集后，TensorFlow被自动从2.11降级到2.6。恢复TensorFlow 2.11版本后，代码在终端运行完全正常，但在PyCharm中执行时触发如下报错：

11/Mar/23 13:07:49 - utils.aux_funcs - ERROR - Exception encountered when calling layer 'conv2d' (type Conv2D).

{{function_node __wrapped__Conv2D_device_/job:localhost/replica:0/task:0/device:GPU:0}} DNN library is not found. [Op:Conv2D]

Call arguments received by layer 'conv2d' (type Conv2D):
  • inputs=tf.Tensor(shape=(64, 128, 128, 1), dtype=float32)
Traceback (most recent call last):
  File "/home/USER/projects/QANetV2/qanet/tf_train.py", line 87, in <module>
    trained_model = train_model(
  File "/home/USER/projects/QANetV2/qanet/tensor_flow/utils/tf_utils.py", line 484, in train_model
    model.fit(
  File "/home/USER/anaconda3/envs/qanet/lib/python3.9/site-packages/keras/utils/traceback_utils.py", line 70, in error_handler
    raise e.with_traceback(filtered_tb) from None
  File "/home/USER/projects/QANetV2/qanet/tensor_flow/custom/tf_models.py", line 208, in call
    return self.model(inputs)
tensorflow.python.framework.errors_impl.UnimplementedError: Exception encountered when calling layer 'conv2d' (type Conv2D).

{{function_node __wrapped__Conv2D_device_/job:localhost/replica:0/task:0/device:GPU:0}} DNN library is not found. [Op:Conv2D]

Call arguments received by layer 'conv2d' (type Conv2D):
  • inputs=tf.Tensor(shape=(64, 128, 128, 1), dtype=float32)

已做排查：

• 确认终端与PyCharm使用的解释器完全一致
• 尝试创建新的conda环境，问题仍未解决
• 怀疑是PyCharm缓存导致的差异

解决方案 #

1. 清理PyCharm缓存并重启 #

• 点击顶部菜单栏File -> Invalidate Caches...
• 勾选Clear file system cache and local history，点击Invalidate and Restart
• 重启后等待PyCharm完成项目索引重建

2. 同步PyCharm的环境变量 #

• 打开Run/Debug Configurations（右上角运行按钮旁下拉框选择Edit Configurations）
• 进入目标运行配置的Environment variables标签页
• 点击Load environment variables from shell，直接导入终端的环境变量配置
• 保存配置后重新运行代码

3. 刷新PyCharm的包索引 #

• 打开File -> Settings -> Project: [你的项目名] -> Python Interpreter
• 点击解释器列表旁的刷新图标🔄，等待PyCharm重新扫描已安装包
• 确认TensorFlow 2.11及cudatoolkit、cudnn等依赖包状态正常

4. 验证GPU设备识别 #

• 在代码开头添加以下片段，分别在终端和PyCharm中运行，对比输出：

import tensorflow as tf
print("Available GPUs:", tf.config.list_physical_devices('GPU'))

• 如果PyCharm无法识别GPU，检查运行配置中是否添加了CUDA_VISIBLE_DEVICES=0环境变量（GPU编号根据实际情况调整）

5. 通过PyCharm重新安装TensorFlow #

• 在Python Interpreter界面找到TensorFlow，点击减号(-)卸载
• 点击加号(+)搜索tensorflow==2.11.0并安装
• 全程通过PyCharm的包管理界面操作，确保PyCharm能正确追踪包的安装路径

内容的提问来源于stack exchange，提问作者Michael