mirror of
https://github.com/clearml/clearml
synced 2025-01-31 17:17:00 +00:00
71 lines
2.1 KiB
Python
71 lines
2.1 KiB
Python
import sys
|
|
import random
|
|
|
|
try:
|
|
import numpy as np
|
|
except Exception:
|
|
np = None
|
|
|
|
|
|
def make_deterministic(seed=1337, cudnn_deterministic=False):
|
|
"""
|
|
Ensure deterministic behavior across PyTorch using the provided random seed.
|
|
This function makes sure that torch, numpy and random use the same random seed.
|
|
|
|
When using trains's task, call this function using the task's random seed like so:
|
|
make_deterministic(task.get_random_seed())
|
|
|
|
:param int seed: Seed number
|
|
:param bool cudnn_deterministic: In order to make computations deterministic on your specific platform
|
|
and PyTorch release, set this value to True. torch will only allow those CuDNN algorithms that are
|
|
(believed to be) deterministic. This can have a performance impact (slower execution) depending on your model.
|
|
"""
|
|
seed = int(seed) & 0xFFFFFFFF
|
|
torch = sys.modules.get("torch")
|
|
tf = sys.modules.get("tensorflow")
|
|
|
|
if cudnn_deterministic:
|
|
try:
|
|
torch.backends.cudnn.deterministic = True
|
|
torch.backends.cudnn.benchmark = False
|
|
except Exception:
|
|
pass
|
|
|
|
random.seed(seed)
|
|
|
|
if np is not None:
|
|
np.random.seed(seed)
|
|
|
|
if torch is not None:
|
|
try:
|
|
torch.manual_seed(seed)
|
|
torch.cuda.manual_seed(seed)
|
|
except Exception:
|
|
pass
|
|
|
|
if tf is not None:
|
|
# reset graph state
|
|
try:
|
|
import tensorflow # noqa: F401
|
|
from tensorflow.python.eager.context import _context
|
|
eager_mode_bypass = _context is None
|
|
except Exception:
|
|
eager_mode_bypass = False
|
|
|
|
if not eager_mode_bypass:
|
|
try:
|
|
tf.compat.v1.set_random_seed(seed)
|
|
except Exception:
|
|
try:
|
|
tf.set_random_seed(seed)
|
|
except Exception:
|
|
pass
|
|
|
|
try:
|
|
tf.compat.v1.random.set_random_seed(seed)
|
|
except Exception:
|
|
try:
|
|
tf.random.set_random_seed(seed)
|
|
except Exception:
|
|
pass
|