Thrsu opened a new issue, #15394:
URL: https://github.com/apache/tvm/issues/15394
Pad produced inconsistent inference results with PyTorch because numerical
overflow when setting padding values.
### Expected behavior
TVM and PyTorch have the consistent inference results.
### Actual behavior
```
Traceback (most recent call last):
...
np.testing.assert_allclose(torch_outputs, tvm_outputs, rtol=1e-3,
atol=1e-3)
File
"/workplace/software/miniconda3/envs/torch/lib/python3.7/site-packages/numpy/testing/_private/utils.py",
line 1528, in assert_allclose
verbose=verbose, header=header, equal_nan=equal_nan)
File
"/workplace/software/miniconda3/envs/torch/lib/python3.7/site-packages/numpy/testing/_private/utils.py",
line 769, in assert_array_compare
hasval='+inf')
File
"/workplace/software/miniconda3/envs/torch/lib/python3.7/site-packages/numpy/testing/_private/utils.py",
line 740, in func_assert_same_pos
raise AssertionError(msg)
AssertionError:
Not equal to tolerance rtol=0.001, atol=0.001
x and y +inf location mismatch:
x: array([[[[3.477122e+307, 3.477122e+307, 3.477122e+307, ...,
3.477122e+307, 3.477122e+307, 3.477122e+307],
[3.477122e+307, 3.477122e+307, 3.477122e+307, ...,...
y: array([[[[inf, inf, inf, ..., inf, inf, inf],
[inf, inf, inf, ..., inf, inf, inf],
[inf, inf, inf, ..., inf, inf, inf],..
```
### Environment
- TVM version: latest
- PyTorch: 1.13.1+cu117
### Steps to reproduce
```Python
import torch
from tvm import relay
import tvm
import numpy as np
from torch.nn import Module
para_0 = torch.randn([20, 20, 20, 5], dtype=torch.float64)
para_1 = (107, 107)
para_2 = 'constant'
para_3 = 3.4771218336380595e+307
class pad(Module):
def forward(self, *args):
return torch.nn.functional.pad(args[0], para_1,para_2,para_3,)
m = pad().float().eval()
input_data = torch.tensor(para_0)
torch_outputs = m(input_data)
trace = torch.jit.trace(m, input_data)
input_shapes = [('input0', torch.Size([20, 20, 20, 5]))]
mod, params = relay.frontend.from_pytorch(trace, input_shapes)
input_data_np = input_data.numpy()
with tvm.transform.PassContext(opt_level=3):
exe = relay.create_executor('graph', mod=mod, params=params,
device=tvm.device('llvm', 0), target='llvm').evaluate()
input_tvm = {'input0': tvm.nd.array(input_data_np.astype(np.float64))}
tvm_outputs = exe(**input_tvm).asnumpy()
np.testing.assert_allclose(torch_outputs, tvm_outputs, rtol=1e-3, atol=1e-3)
```
### Triage
* needs-triage
* frontend:pytorch
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
