Skip to content

Gate set_detect_anomaly behind STYLETTS2_DETECT_ANOMALY env var #3

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
shreyaskarnik wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Gate set_detect_anomaly behind STYLETTS2_DETECT_ANOMALY env var #3

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 8 additions & 1 deletion train_second.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,13 +1,20 @@
# load packages
import copy
import os
import random
import yaml
import time
from munch import Munch
import numpy as np
import torch

torch.autograd.set_detect_anomaly(True)
# Anomaly detection is a debugging tool — it traces every op in the autograd
# graph to surface NaN/inf sources, but it makes backward 5–10× slower, holds
# extra memory (causes OOMs that wouldn't otherwise happen), and on some
# single-GPU + bf16 setups it deadlocks the first backward pass entirely.
# Off by default; opt in via STYLETTS2_DETECT_ANOMALY=1 when diagnosing.
if os.environ.get("STYLETTS2_DETECT_ANOMALY") == "1":
torch.autograd.set_detect_anomaly(True)

if getattr(torch, "_original_load", None) is None:
torch._original_load = torch.load
Expand Down