ChijinZ commented on issue #16053:
URL: https://github.com/apache/tvm/issues/16053#issuecomment-1798412201
This looks like a very interesting bug. I simplify the code to make it
runable without external input files.
```python
import onnx
import numpy as np
import pickle
from numpy import testing
import tvm
from tvm import relay
import torch
p0 = torch.nn.Parameter(torch.empty([18, 1, 1, 1, 60], dtype=torch.int64),
requires_grad=False)
input_array = torch.round(torch.rand(18, 1, 1, 50, 1,
dtype=torch.float64)).to(torch.int64)
shape_dict = {"x": input_array.shape}
class Model0(torch.nn.Module):
def __init__(self):
super().__init__()
self.const = p0
def forward(self, x):
y = self.const
add = torch.add(x, y)
neg = torch.neg(add)
argmin = neg.argmin(4)
return (argmin)
model_0 = Model0()
torch.onnx.export(model_0, input_array, '0.onnx', verbose=False,
input_names=['x'], output_names=['out'], opset_version=14,
do_constant_folding=False)
class Model1(torch.nn.Module):
def __init__(self):
super().__init__()
self.const = p0
def forward(self, x):
y = self.const
trans_0 = x.transpose(1, 0)
trans_1 = y.transpose(1, 0)
add = torch.add(trans_0, trans_1)
add_trans = add.transpose(1, 0)
neg = torch.neg(add_trans)
argmin = neg.argmin(4)
return (argmin)
model_1 = Model1()
torch.onnx.export(model_1, input_array, '1.onnx', verbose=False,
input_names=['x'], output_names=['out'], opset_version=14,
do_constant_folding=False)
onnx_model_0 = onnx.load('0.onnx')
onnx_model_outputs_0 = [node.name for node in onnx_model_0.graph.output]
mod_0, params_0 = relay.frontend.from_onnx(onnx_model_0, shape_dict,
freeze_params=True)
with tvm.transform.PassContext(opt_level=4):
executor_0 = relay.build_module.create_executor("graph", mod_0,
tvm.cpu(), tvm.target.Target("llvm"), params_0).evaluate()
executor_res_0 = [executor_0(input_array).numpy()]
output_0 = dict(zip(onnx_model_outputs_0, executor_res_0))
onnx_model_1 = onnx.load('1.onnx')
onnx_model_outputs_1 = [node.name for node in onnx_model_1.graph.output]
shape_dict_1 = {"x": input_array.shape}
mod_1, params_1 = relay.frontend.from_onnx(onnx_model_1, shape_dict,
freeze_params=True)
with tvm.transform.PassContext(opt_level=4):
executor_1 = relay.build_module.create_executor("graph", mod_1,
tvm.cpu(), tvm.target.Target("llvm"), params_1).evaluate()
executor_res_1 = [executor_1(input_array).numpy()]
output_1 = dict(zip(onnx_model_outputs_1, executor_res_1))
print('=========================')
try:
testing.assert_allclose(output_0["out"], output_1["out"])
print("tvm_opt_4 does not trigger assertion")
except AssertionError as e:
print("tvm_opt_4 triggers assertion")
print(e)
print('=========================')
mod_0, params_0 = relay.frontend.from_onnx(onnx_model_0, shape_dict,
freeze_params=True)
with tvm.transform.PassContext(opt_level=0):
executor_0 = relay.build_module.create_executor("graph", mod_0,
tvm.cpu(), tvm.target.Target("llvm"), params_0).evaluate()
executor_res_0 = [executor_0(input_array).numpy()]
output_0 = dict(zip(onnx_model_outputs_0, executor_res_0))
mod_1, params_1 = relay.frontend.from_onnx(onnx_model_1, shape_dict,
freeze_params=True)
with tvm.transform.PassContext(opt_level=0):
executor_1 = relay.build_module.create_executor("graph", mod_1,
tvm.cpu(), tvm.target.Target("llvm"), params_1).evaluate()
executor_res_1 = [executor_1(input_array).numpy()]
output_1 = dict(zip(onnx_model_outputs_1, executor_res_1))
print('=========================')
try:
testing.assert_allclose(output_0["out"], output_1["out"])
print("tvm_opt_0 does not trigger assertion")
except AssertionError as e:
print("tvm_opt_0 triggers assertion")
print(e)
print('=========================')
```
--
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]
