🐛 Describe the bug
This was found in opinfo tests.
Repro:
import torch
from torch import tensor
dtype = torch.float16
def f(add, a, b):
return torch.addmm(add, a, b)
data = []
data.append((
tensor([-7.2227], dtype=torch.float16),
tensor([[ 0.7109, -4.8516], [-0.6484, 2.6719]], dtype=torch.float16),
tensor([[-3.6816, -7.9023, -3.7266], [1.0547, 4.1719, -1.1328]], dtype=torch.float16),
))
use_trace = False
for (add, a, b) in data:
x = torch.addmm(add, a, b)
if use_trace:
# here, torch.jit.trace throws a warning about output not being correct
sf = torch.jit.trace(f, [add, a, b])
y = torch.addmm(add, a, b)
else:
sf = torch.jit.script(f)
y = sf(add, a, b)
print(sf.graph)
print("X: ", x)
print("Y: ", y)
print(x-y)
assert((x==y).all().item())
print()Output:
graph(%add.1 : Tensor,
%a.1 : Tensor,
%b.1 : Tensor):
%8 : int = prim::Constant[value=1]()
%10 : Tensor = aten::addmm(%add.1, %a.1, %b.1, %8, %8) # /data/sandcastle/boxes/fbsource/fbcode/buck-out/opt/gen/scripts/dberard/matmul#link-tree/scripts/dberard/matmul.py:7:11
return (%10)
X: tensor([[-14.9609, -33.0625, -4.3789],
[ -2.0176, 9.0469, -7.8320]], dtype=torch.float16)
Y: tensor([[-14.9531, -33.0625, -4.3750],
[ -2.0195, 9.0625, -7.8359]], dtype=torch.float16)
tensor([[-0.0078, 0.0000, -0.0039],
[ 0.0020, -0.0156, 0.0039]], dtype=torch.float16)
Traceback (most recent call last):
File "/path/matmul.py", line 32, in <module>
assert((x==y).all().item())
AssertionError
Versions
(non-gpu devserver)
Collecting environment information...
PyTorch version: 1.11.0a0+gitad36af8
Is debug build: True
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A
OS: CentOS Stream 8 (x86_64)
GCC version: (GCC) 8.5.0 20210514 (Red Hat 8.5.0-3)
Clang version: Could not collect
CMake version: version 3.19.6
Libc version: glibc-2.28
Python version: 3.9.6 (default, Aug 18 2021, 19:38:01) [GCC 7.5.0] (64-bit runtime)
Python platform: Linux-5.6.13-0_fbk19_hardened_6064_gabfd136bb69a-x86_64-with-glibc2.28
Is CUDA available: False
CUDA runtime version: No CUDA
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A
Versions of relevant libraries:
[pip3] mypy==0.812
[pip3] mypy-extensions==0.4.3
[pip3] numpy==1.20.0
[pip3] torch==1.11.0a0+gitfcc5577
[conda] blas 1.0 mkl
[conda] mkl 2021.3.0 h06a4308_520
[conda] mkl-include 2021.3.0 h06a4308_520
[conda] mkl-service 2.4.0 py39h7f8727e_0
[conda] mkl_fft 1.3.0 py39h42c9631_2
[conda] mkl_random 1.2.2 py39h51133e4_0
[conda] mypy 0.910 pyhd3eb1b0_0
[conda] mypy_extensions 0.4.3 py39h06a4308_0
[conda] numpy 1.20.3 py39hf144106_0
[conda] numpy-base 1.20.3 py39h74d4b33_0
[conda] torch 1.11.0a0+git33ab075 pypi_0 pypi
🐛 Describe the bug
This was found in opinfo tests.
Repro:
Output:
Versions
(non-gpu devserver)