模型转onnx后，verify计算误差报错

[a773783082]

Question

onnx模型转成功了，但是后续的verify过程导入onnx模型报错。想问下该怎么解决？

## version
onnx                         1.17.0
onnxruntime-gpu              1.14.0
onnxsim                      0.4.36
onnx-graphsurgeon            0.5.2
torch                        1.11.0+cu113

WARNING: The shape inference of prim::Constant type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of prim::Constant type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
Successfully exported ONNX model: /workspace/log_7cls_det/salience_detr_swin_l_480_800/2024-10-15-00_53_43/best_ap_save.onnx
Uncaught exception
Traceback (most recent call last):
  File "/workspace/train_code/Salience-DETR-main/tools/pytorch2onnx.py", line 145, in <module>
    pytorch2onnx()
  File "/workspace/train_code/Salience-DETR-main/tools/pytorch2onnx.py", line 132, in pytorch2onnx
    onnx_model = ONNXDetector(args.save_file)
  File "/workspace/train_code/Salience-DETR-main/tools/pytorch2onnx.py", line 21, in __init__
    self.session = onnxruntime.InferenceSession(
  File "/root/miniconda3/lib/python3.9/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 360, in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
  File "/root/miniconda3/lib/python3.9/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 397, in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_path, True, self._read_config_from_model)
onnxruntime.capi.onnxruntime_pybind11_state.InvalidGraph: [ONNXRuntimeError] : 10 : INVALID_GRAPH : Load model from /workspace/log_7cls_det/salience_detr_swin_l_480_800/2024-10-15-00_53_43/best_ap_save.onnx failed:This is an invalid model. Type Error: Type 'tensor(int32)' of input parameter (onnx::TopK_19989) of operator (TopK) in node (TopK_16698) is invalid.

补充信息

No response

[xiuqhou]

看报错信息是topk运算的输入参数类型不对，应该输入的是int64类型的参数。
我这边没有复现出来这个问题，但代码中用到topk的地方不多，你可以直接全文搜索，将topk的k参数换成int64类型再导出试一试，类似这样：

# 修改前
valid_score.topk(level_token_nums[level_idx], dim=1)
# 修改后
valid_score.topk(level_token_nums[level_idx].to(torch.int64), dim=1)

[a773783082]

@xiuqhou ，感谢，修改后成功了，👍，只是误差有点大

Successfully exported ONNX model: /workspace/log_7cls_det/salience_detr_swin_l_480_800/2024-10-15-00_53_43/best_ap_save.onnx
Uncaught exception
Traceback (most recent call last):
  File "/workspace/train_code/Salience-DETR-main/tools/pytorch2onnx.py", line 145, in <module>
    pytorch2onnx()
  File "/workspace/train_code/Salience-DETR-main/tools/pytorch2onnx.py", line 138, in pytorch2onnx
    np.testing.assert_allclose(
  File "/root/miniconda3/lib/python3.9/site-packages/numpy/testing/_private/utils.py", line 1592, in assert_allclose
    assert_array_compare(compare, actual, desired, err_msg=str(err_msg),
  File "/root/miniconda3/lib/python3.9/contextlib.py", line 79, in inner
    return func(*args, **kwds)
  File "/root/miniconda3/lib/python3.9/site-packages/numpy/testing/_private/utils.py", line 862, in assert_array_compare
    raise AssertionError(msg)
AssertionError: 
Not equal to tolerance rtol=0.001, atol=1e-05
The numerical values are different between Pytorch and ONNXBut it does not necessarily mean the exported ONNX is problematic.
Mismatched elements: 255 / 300 (85%)
Max absolute difference: 0.00075046
Max relative difference: 0.04880792
 x: array([0.043474, 0.041977, 0.041789, 0.04153 , 0.041011, 0.040653,
       0.040344, 0.039332, 0.038988, 0.038698, 0.03841 , 0.038371,
       0.038008, 0.037157, 0.036468, 0.035893, 0.035884, 0.035836,...
 y: array([0.043971, 0.042558, 0.042212, 0.041955, 0.040835, 0.040663,
       0.040282, 0.03931 , 0.039249, 0.039102, 0.038791, 0.038654,
       0.038188, 0.037154, 0.036845, 0.036507, 0.036416, 0.036176,...