You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The mismatches look like overflowing values (e.g. 50 vs. -206).
Environment
PyTorch version: 1.11.0.dev20211019+cu111
Is debug build: False
CUDA used to build PyTorch: 11.1
ROCM used to build PyTorch: N/A
OS: Ubuntu 20.04.3 LTS (x86_64)
GCC version: (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0
Clang version: Could not collect
CMake version: version 3.21.3
Libc version: glibc-2.31
🐛 Bug
A traced model seems to create overflows when
int8
tensor values are transformed toint32
on the GPU.To Reproduce
Output:
The mismatches look like overflowing values (e.g.
50
vs.-206
).Environment
Reproduced on different GPUs.
Additional information
Enabling
nvfuser
via:yields matching values.
CC @malfet as we've discussed this issue. (Initially I thought it would be ARM-specifc, but that turns out to be wrong)
The text was updated successfully, but these errors were encountered: