[CI/Build] Avoid CUDA initialization (#8534)
This commit is contained in:
@@ -48,7 +48,7 @@ WTYPE_ZEROPOINTS = [
|
||||
# `is_quant_method_supported` conflates kernels with quantization methods
|
||||
# an assumption which is breaking down as quantizations methods can have
|
||||
# have kernels and some kernels support multiple quantization methods.
|
||||
IS_SUPPORTED_BY_GPU = current_platform.get_device_capability()[0] >= 9
|
||||
IS_SUPPORTED_BY_GPU = current_platform.has_device_capability(90)
|
||||
|
||||
|
||||
def rand_data(shape, dtype=torch.float16):
|
||||
|
||||
Reference in New Issue
Block a user