HongChuang/ultralytics-ascend
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 3

Hardcode failing GPUs in AMP checks (#17977)

Browse source 
Co-authored-by: UltralyticsAssistant <web@ultralytics.com>
Co-authored-by: Burhan <62214284+Burhan-Q@users.noreply.github.com>
Co-authored-by: Ultralytics Assistant <135830346+UltralyticsAssistant@users.noreply.github.com>
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>
This commit is contained in:
Mohammed Yasin 2024-12-06 05:23:36 +08:00 • committed by GitHub
parent 1f5646634d
commit 5c84099a9d
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 14 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

15
ultralytics/utils/checks.py
View file

@ -669,8 +669,22 @@ def check_amp(model):
from ultralytics.utils.torch_utils import autocast
device = next(model.parameters()).device # get model device
prefix = colorstr("AMP: ")
if device.type in {"cpu", "mps"}:
return False # AMP only used on CUDA devices
else:
# GPUs that have issues with AMP
pattern = re.compile(
r"(nvidia|geforce|quadro|tesla).*?(1660|1650|1630|t400|t550|t600|t1000|t1200|t2000|k40m)", re.IGNORECASE
)
gpu = torch.cuda.get_device_name(device)
if bool(pattern.search(gpu)):
LOGGER.warning(
f"{prefix}checks failed ❌. AMP training on {gpu} GPU may cause "
f"NaN losses or zero-mAP results, so AMP will be disabled during training."
)
return False
def amp_allclose(m, im):
"""All close FP32 vs AMP results."""
@ -683,7 +697,6 @@ def check_amp(model):
return a.shape == b.shape and torch.allclose(a, b.float(), atol=0.5) # close to 0.5 absolute tolerance
im = ASSETS / "bus.jpg" # image to check
prefix = colorstr("AMP: ")
LOGGER.info(f"{prefix}running Automatic Mixed Precision (AMP) checks...")
warning_msg = "Setting 'amp=True'. If you experience zero-mAP or NaN losses you can disable AMP with amp=False."
try: