@baudm I have a similar issue with loading the converted ONNX model. I am able to successfully convert the model to ONNX, but when I try to load and check if the model is well-formed I get the error.

import torch
import onnx

# Load PyTorch model
parseq = torch.hub.load('baudm/parseq', 'parseq', pretrained=True).eval()
dummy_input = torch.rand(1, 3, *parseq.hparams.img_size)

# Convert to ONNX
parseq.to_onnx('pairseq.onnx', dummy_input, opset_version=14)

# Load the ONNX model
onnx_model = onnx.load('pairseq.onnx')

# Check ONNX model
onnx.checker.check_model(onnx_model, full_check=True)


---------------------------------------------------------------------------
InferenceError                            Traceback (most recent call last)
Input In [1], in <cell line: 15>()
     12 onnx_model = onnx.load('pairseq.onnx')
     14 # Check ONNX model
---> 15 onnx.checker.check_model(onnx_model, full_check=True)

File /opt/venv/lib/python3.8/site-packages/onnx/checker.py:108, in check_model(model, full_check)
    106 C.check_model(protobuf_string)
    107 if full_check:
--> 108     onnx.shape_inference.infer_shapes(model, check_type=True, strict_mode=True)

File /opt/venv/lib/python3.8/site-packages/onnx/shape_inference.py:34, in infer_shapes(model, check_type, strict_mode, data_prop)
     32 if isinstance(model, (ModelProto, bytes)):
     33     model_str = model if isinstance(model, bytes) else model.SerializeToString()
---> 34     inferred_model_str = C.infer_shapes(model_str, check_type, strict_mode, data_prop)
     35     return onnx.load_from_string(inferred_model_str)
     36 elif isinstance(model, str):

InferenceError: [ShapeInferenceError] (op_type:CumSum, node name: CumSum_2527): x typestr: T, has unsupported type: tensor(bool)

Support onnx #12

Description

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions