mindspore.device_context.gpu.op_precision.matmul_allow_tf32
- mindspore.device_context.gpu.op_precision.matmul_allow_tf32(value)[source]
Whether to convert FP32 to TF32 for Matmul operators. For detailed information, please refer to CUBLAS_COMPUTE_32F_FAST_TF32.
- Parameters
value (bool) – Whether to convert FP32 to TF32 for Matmul operators. If not configured, the framework defaults to
False
.
Examples
>>> import mindspore as ms >>> ms.device_context.gpu.op_precision.matmul_allow_tf32(True)