loss.backward() -> Missing XLA configuration

Calculated losses from models created using pytorch
When you run the following code during propagation: The following error message is troubling me.

loss.backward()

The forward propagation calculation can be performed without any problems.

terminate called after browsing an instance of 'std::runtime_error'
  what():tensorflow/compiler/xla/xla_client/computation_client.cc:280:Missing XLA configuration
Aborted

• pytorch (1.12.0+cu102)
• torchvision(0.13.0+cu102)<- Target models include CNN models that can be installed from torchvision.models
• Google-compute-engine
• GPU (NVIDIA Tesla T4x111.6)<- This code worked in an environment with GPU (11.2) installed, but it doesn't work in the current environment.../In the current environment, the same error occurs when the GPU is not used and the CPU is running.
• TPU is not installed (I don't want to use TPU, I want to use GPU to calculate)

It was a locally working code, and as I mentioned above, it was working in other GPU environments.The environment stopped working when it was updated.

Please help me...