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

Unset CUBLAS_WORKSPACE_CONFIG for non-deterministic training and inference (#19138)

Browse source 
Co-authored-by: UltralyticsAssistant <web@ultralytics.com>
This commit is contained in:
Mohammed Yasin 2025-02-10 07:17:08 +08:00 • committed by GitHub
parent 31e96e20d3
commit ff1a04609f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
5 changed files with 16 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

1
ultralytics/utils/__init__.py
View file

@ -128,7 +128,6 @@ torch.set_printoptions(linewidth=320, precision=4, profile="default")
np.set_printoptions(linewidth=320, formatter={"float_kind": "{:11.5g}".format}) # format short g, %precision=5
cv2.setNumThreads(0) # prevent OpenCV from multithreading (incompatible with PyTorch DataLoader)
os.environ["NUMEXPR_MAX_THREADS"] = str(NUM_THREADS) # NumExpr max threads
os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":4096:8" # for deterministic training to avoid CUDA warning
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3" # suppress verbose TF compiler warnings in Colab
os.environ["TORCH_CPP_LOG_LEVEL"] = "ERROR" # suppress "NNPACK.cpp could not initialize NNPACK" warnings
os.environ["KINETO_LOG_LEVEL"] = "5" # suppress verbose PyTorch profiler output when computing FLOPs