torch.accelerator.device_count

torch.accelerator.device_count()

Return the number of current accelerator available.

Returns:

the number of the current accelerator available.

If there are no available accelerators, return 0.

Return type:

int

Note

This API delegates to the device-specific version of device_count. On CUDA, this API will NOT poison fork if NVML discovery succeeds. Otherwise, it will. For more details, see Poison fork in multiprocessing.