Skip to content

GitLab

Unverified Commit 1a86583b authored by guoshzhao's avatar guoshzhao Committed by GitHub
Browse files

Benchmarks: Fix bug - Fix bug when set force_fp32 option. (#214) 

**Description**
Fix typo when set force_fp32 option.
parent f9442456
Show whitespace changes
Inline Side-by-side
Showing with 2 additions and 2 deletions
superbench/benchmarks/model_benchmarks/pytorch_base.py
View file @ 1a86583b
...@@ -38,8 +38,8 @@ def _set_force_fp32(self): ...@@ -38,8 +38,8 @@ def _set_force_fp32(self):
On Ampere or newer GPUs, pytorch and tensorflow will use TF32 instead of FP32 by default. On Ampere or newer GPUs, pytorch and tensorflow will use TF32 instead of FP32 by default.
We can disable TF32 execution by setting force_fp32 as True. We can disable TF32 execution by setting force_fp32 as True.
""" """
torch.backends.cuda.matmul.allow_tf32 = self._args.force_fp32 torch.backends.cuda.matmul.allow_tf32 = not self._args.force_fp32
torch.backends.cudnn.allow_tf32 = self._args.force_fp32 torch.backends.cudnn.allow_tf32 = not self._args.force_fp32
def _init_distributed_setting(self): def _init_distributed_setting(self):
"""Initialize the distributed library and bind the worker to GPU. """Initialize the distributed library and bind the worker to GPU.
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment