jikechao opened a new issue, #15160:
URL: https://github.com/apache/tvm/issues/15160
The input_dtype == 'float64' will lead to a check failure and throw "Check
failed: from_size == to_size (3360 vs. 1680) : TVMArrayCopyFromTo: The size
must exactly match".
This bug looks similar to this PR #7665.
### Actual behavior
```
Traceback (most recent call last):
File "test.py", line 28, in <module>
res_tvm = model(tvm.nd.array(test_x_tvm.astype('float64'))).numpy()
File "/workplace/software/tvm/tvm_/python/tvm/relay/build_module.py", line
537, in _graph_wrapper
gmodule.set_input(i, arg)
File "/workplace/software/tvm/tvm_/python/tvm/contrib/graph_executor.py",
line 199, in set_input
v.copyfrom(value)
File "/workplace/software/tvm/tvm_/python/tvm/runtime/ndarray.py", line
143, in copyfrom
source_array.copyto(self)
File "/workplace/software/tvm/tvm_/python/tvm/runtime/ndarray.py", line
251, in copyto
return self._copyto(target)
File "/workplace/software/tvm/tvm_/python/tvm/_ffi/_ctypes/ndarray.py",
line 90, in _copyto
check_call(_LIB.TVMArrayCopyFromTo(self.handle, target_nd.handle, None))
File "/workplace/software/tvm/tvm_/python/tvm/_ffi/base.py", line 348, in
check_call
raise get_last_ffi_error()
tvm._ffi.base.TVMError: Traceback (most recent call last):
49: 0x000055c08bd49b32
at ../sysdeps/x86_64/elf/start.S:103
48: __libc_start_main
47: Py_BytesMain
at /tmp/build/80754af9/python_1589913560097/work/Modules/main.c:1089
46: Py_RunMain
at /tmp/build/80754af9/python_1589913560097/work/Modules/main.c:650
45: pymain_run_python
at /tmp/build/80754af9/python_1589913560097/work/Modules/main.c:571
44: pymain_run_file
at /tmp/build/80754af9/python_1589913560097/work/Modules/main.c:387
43: PyRun_SimpleFileExFlags
at
/tmp/build/80754af9/python_1589913560097/work/Python/pythonrun.c:428
42: PyRun_FileExFlags
at
/tmp/build/80754af9/python_1589913560097/work/Python/pythonrun.c:1063
41: run_mod
at
/tmp/build/80754af9/python_1589913560097/work/Python/pythonrun.c:1147
40: run_eval_code_obj
at
/tmp/build/80754af9/python_1589913560097/work/Python/pythonrun.c:1125
39: PyEval_EvalCode
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:718
38: PyEval_EvalCodeEx
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4327
37: _PyEval_EvalCodeWithName
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4298
36: _PyEval_EvalFrameDefault
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:3500
35: call_function
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4963
34: _PyObject_Vectorcall
at
/tmp/build/80754af9/python_1589913560097/work/Include/cpython/abstract.h:127
33: _PyFunction_Vectorcall
at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:435
32: _PyEval_EvalCodeWithName
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4298
31: _PyEval_EvalFrameDefault
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:3486
30: call_function
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4963
29: _PyObject_Vectorcall
at
/tmp/build/80754af9/python_1589913560097/work/Include/cpython/abstract.h:127
28: _PyFunction_Vectorcall
at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:435
27: _PyEval_EvalCodeWithName
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4298
26: _PyEval_EvalFrameDefault
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:3486
25: call_function
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4963
24: _PyObject_Vectorcall
at
/tmp/build/80754af9/python_1589913560097/work/Include/cpython/abstract.h:127
23: _PyFunction_Vectorcall
at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:410
22: function_code_fastcall
at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:283
21: _PyEval_EvalFrameDefault
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:3486
20: call_function
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4963
19: _PyObject_Vectorcall
at
/tmp/build/80754af9/python_1589913560097/work/Include/cpython/abstract.h:127
18: _PyFunction_Vectorcall
at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:435
17: _PyEval_EvalCodeWithName
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4298
16: _PyEval_EvalFrameDefault
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:3486
15: call_function
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4963
14: _PyObject_Vectorcall
at
/tmp/build/80754af9/python_1589913560097/work/Include/cpython/abstract.h:127
13: _PyFunction_Vectorcall
at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:410
12: function_code_fastcall
at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:283
11: _PyEval_EvalFrameDefault
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:3469
10: call_function
at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4963
9: _PyObject_Vectorcall
at
/tmp/build/80754af9/python_1589913560097/work/Include/cpython/abstract.h:125
8: _PyObject_MakeTpCall
at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:159
7: PyCFuncPtr_call
at /usr/local/src/conda/python-3.8.3/Modules/_ctypes/_ctypes.c:4201
6: _ctypes_callproc
at /usr/local/src/conda/python-3.8.3/Modules/_ctypes/callproc.c:1199
5: _call_function_pointer
at /usr/local/src/conda/python-3.8.3/Modules/_ctypes/callproc.c:871
4: ffi_call_int
3: ffi_call_unix64
2: TVMArrayCopyFromTo
1: tvm::runtime::NDArray::CopyFromTo(DLTensor const*, DLTensor*, void*)
0: _ZN3tvm7runtime6detail
File "/workplace/software/tvm/tvm_/src/runtime/ndarray.cc", line 270
TVMError:
```
### Environment
Any environment details, such as: Operating System, TVM version, etc
### Steps to reproduce
```
import tvm
import tvm.relay as relay
import numpy as np
from tensorflow import keras
from tensorflow.keras import layers, models
input_shape = (1, 5, 7, 6, 2)
input_data = np.random.random(size=input_shape)
x = layers.Input(shape=input_shape[1:], dtype='float64')
layer = keras.layers.Conv3D(filters=1, kernel_size=3)
layer.set_weights(layer.get_weights())
y = layer(x)
model = models.Model(x, y)
model.summary()
res_keras = model(input_data)
shape_dict = {'input_1': input_shape}
mod, params = relay.frontend.from_keras(model, shape_dict,layout='NDHWC')
with tvm.transform.PassContext(opt_level=3):
model = relay.build_module.create_executor("graph", mod, tvm.cpu(0),
'llvm', params).evaluate()
test_x_tvm = input_data
res_tvm = model(tvm.nd.array(test_x_tvm.astype('float64'))).numpy()
np.testing.assert_allclose(res_keras, res_tvm, atol=1e-3, rtol=1e-3)
```
cc @comaniac @tkonolige
--
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]
