42 lines
1.6 KiB
Python
42 lines
1.6 KiB
Python
import torch
|
|
from packaging.version import parse
|
|
|
|
|
|
def configure_torch_backends(allow_tf32: bool = True, deterministic: bool = False):
|
|
"""Configure PyTorch CUDA and cuDNN backends for performance/reproducibility.
|
|
|
|
Args:
|
|
allow_tf32: If True, use TF32 precision for faster computation on Ampere+ GPUs. If
|
|
False, use standard IEEE FP32 precision.
|
|
deterministic: If True, use deterministic algorithms (slower but reproducible).
|
|
If False, allow cuDNN to benchmark and select fastest algorithms.
|
|
|
|
Note:
|
|
TF32 uses reduced precision (10-bit mantissa vs 23-bit for FP32) for internal
|
|
matrix multiplications providing a speedup with minimal impact on accuracy.
|
|
|
|
See https://pytorch.org/docs/stable/notes/cuda.html#tf32-on-ampere for details.
|
|
"""
|
|
torch_version = parse(torch.__version__.split("+")[0]) # Handle e.g., "2.9.0+cu118".
|
|
if torch_version >= parse("2.9.0"):
|
|
_configure_29(allow_tf32)
|
|
else:
|
|
_configure_pre29(allow_tf32)
|
|
|
|
torch.backends.cudnn.benchmark = not deterministic # Find fastest algorithms.
|
|
torch.backends.cudnn.deterministic = deterministic # Ensure reproducibility.
|
|
|
|
|
|
def _configure_29(allow_tf32: bool):
|
|
"""Configure PyTorch CUDA and cuDNN backends for PyTorch 2.9+."""
|
|
# tf32 for performance, ieee for full FP32 accuracy.
|
|
precision = "tf32" if allow_tf32 else "ieee"
|
|
torch.backends.cuda.matmul.fp32_precision = precision
|
|
torch.backends.cudnn.fp32_precision = precision
|
|
|
|
|
|
def _configure_pre29(allow_tf32: bool):
|
|
"""Configure PyTorch CUDA and cuDNN backends for PyTorch <2.9."""
|
|
torch.backends.cuda.matmul.allow_tf32 = allow_tf32
|
|
torch.backends.cudnn.allow_tf32 = allow_tf32
|