SeanWangJS opened a new issue, #13211:
URL: https://github.com/apache/tvm/issues/13211
### Expected behavior
The same output value as pytorch function.
### Actual behavior
Random output I guess.
### Environment
OS: Ubuntu 18.04
PyTorch: 1.10.0
TVM: 0.10.0rc0.dev73+g2a2dd9ac5
### Steps to reproduce
```python
from torchvision import ops
import torchvision.transforms.functional as F
import torch
import tvm
from tvm import relay
from tvm.contrib import graph_executor
torch.manual_seed(123)
imgs = torch.randn(1, 3, 14, 14)
rois = torch.FloatTensor([[ 0., 0., 56., 448., 392]])
roi_align = ops.RoIAlign(2, spatial_scale = 14 / 448.0, sampling_ratio = -1,
aligned = True)
out1 = roi_align(imgs, rois)
print("pytorch output: ", out1)
script_roi_align = torch.jit.trace(roi_align, (imgs, rois)).eval()
shape_list = [("xs", imgs.shape), ("rois", rois.shape)]
mod, params = relay.frontend.from_pytorch(script_roi_align, shape_list)
target = tvm.target.Target("llvm", host="llvm")
with tvm.transform.PassContext(opt_level=0):
lib = relay.build(mod, target=target, params = params)
dev = tvm.cpu(0)
model = graph_executor.GraphModule(lib['default'](dev))
model.set_input("xs", imgs)
model.set_input("rois", rois)
model.run()
out3 = model.get_output(0)
print("tvm output: ", out3)
```
The tvm output seems random value at every run, however, Its OK when I set
"aligned" of RoIAlign to False.
### Triage
I'am not sure if current version of TVM is support PyTorch 1.10.0 perfectly
since I noticed that they say now TVM supports PyTorch 1.7 and 1.4.
* needs-triage
--
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]
