We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
There was an error while loading. Please reload this page.
1 parent ed6ae1e commit e1eefa4Copy full SHA for e1eefa4
vllm/model_executor/layers/batch_invariant.py
@@ -852,5 +852,6 @@ def init_batch_invariance():
852
enable_batch_invariant_mode()
853
854
# Disable TF32 for batch invariance - it causes non-deterministic rounding
855
- torch.backends.cuda.matmul.allow_tf32 = False
856
- torch.backends.cudnn.allow_tf32 = False
+ torch.backends.cuda.matmul.fp32_precision = "ieee"
+ torch.backends.cudnn.conv.fp32_precision = "ieee"
857
+ torch.backends.cudnn.rnn.fp32_precision = "ieee"
0 commit comments