ONNX inference issue

[YenYunn]

when i run this code, i got the warning
import torch
import onnx

parseq = load_from_checkpoint('pretrained=parseq').eval()
parseq.refine_iters = 0
parseq.decode_ar = False

image = torch.rand(1, 3, *parseq.hparams.img_size)
parseq.to_onnx('parseq.onnx', image, do_constant_folding=True, opset_version=14)

onnx_model = onnx.load('parseq.onnx')
onnx.checker.check_model(onnx_model, full_check=True)

then, when I use this ONNX model for inference, I encounter the following error

and this is my code

[YenYunn]

my version
onnx==1.15.0
onnxruntime-gpu==1.17.1
torch==2.1.1+cu118
pytorch-lightning==2.1.0

[IceboxDev]

Exactly same issue with:

• onnx==1.15.0
• torch==2.2.1
• pytorch-lightning==2.2.1

Stacktrace:

(.venv) [mantas@WS21 parseq]$ python onnx_runtime.py
Traceback (most recent call last):
  File "/home/mantas/Documents/Projects/parseq/onnx_runtime.py", line 4, in <module>
    session = ort.InferenceSession(model_path)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/mantas/Documents/Projects/parseq/.venv/lib/python3.11/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 419, in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
  File "/home/mantas/Documents/Projects/parseq/.venv/lib/python3.11/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 472, in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_path, True, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : Load model from modified.onnx failed:Type Error: Type parameter (T) of Optype (Where) bound to different types (tensor(bool) and tensor(float) in node (/Where_23).

[YenYunn]

@baudm
Hello, author,

I am currently encountering some technical issues and would appreciate your assistance. Firstly, I would like to inquire about how to resolve the aforementioned problem. Secondly, after using an older version of the project and converting it to ONNX format, I noticed a significant discrepancy between the output results and the pre-trained model. Regarding this issue, could you provide some suggestions to help address this problem?

Thank you very much for taking the time to respond amidst your busy schedule.

[GIN-zhao]

Exactly same issue with:

• onnx==1.15.0
• torch==2.2.1
• pytorch-lightning==2.2.1

Stacktrace:

(.venv) [mantas@WS21 parseq]$ python onnx_runtime.py
Traceback (most recent call last):
  File "/home/mantas/Documents/Projects/parseq/onnx_runtime.py", line 4, in <module>
    session = ort.InferenceSession(model_path)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/mantas/Documents/Projects/parseq/.venv/lib/python3.11/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 419, in __init__
    self._create_inference_session(providers, provider_options, disabled_optimizers)
  File "/home/mantas/Documents/Projects/parseq/.venv/lib/python3.11/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 472, in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_path, True, self._read_config_from_model)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : Load model from modified.onnx failed:Type Error: Type parameter (T) of Optype (Where) bound to different types (tensor(bool) and tensor(float) in node (/Where_23).

in parseq/models.py line-117, update tensor.bool to tensor.float,may be the version of torch and onnxruntime-gpu