nolanliou commented on issue #6268:
URL: https://github.com/apache/incubator-tvm/issues/6268#issuecomment-689542616
@masahi
There are some problems with the type/shape inference when converting the
PyTorch model.
Sample code to reproduce.
```
import numpy as np
import torch
import tvm
from tvm import relay
import tvm.contrib.graph_runtime as runtime
def fn(x):
x /= x.shape[-1]
return x
jit = torch.jit.trace(fn, (torch.ones(4, 5, dtype=torch.int64),))
inputs = torch.randint(low=0, high=10, size=(4, 5), dtype=torch.float32)
print(jit(inputs))
shape_list = [('input', [4, 5])]
mod, params = relay.frontend.pytorch.from_pytorch(jit,
shape_list,
default_dtype="float32")
with tvm.transform.PassContext(opt_level=3):
json, lib, params = relay.build(mod, target="llvm", params=params)
ctx = tvm.cpu(0)
inputs = tvm.nd.array(inputs.cpu().numpy().astype(np.int64), ctx)
runtime = tvm.contrib.graph_runtime.create(json, lib, ctx)
runtime.set_input(**params)
runtime.set_input('input', inputs)
runtime.run()
tvm_result = runtime.get_output(0).asnumpy()
print(tvm_result)
```
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
