onnxruntime InferenceSession error

[ruajgmeiogaui]

Describe the issue

I have loaded a mdoel.pt and transfer it to onnx.But when I try to run ort.InferenceSession,such error occurs:onnxruntime.capi.onnxruntime_pybind11_state.NotImplemented: [ONNXRuntimeError] : 9 : NOT_IMPLEMENTED : Could not find an implementation for EyeLike(9) node with name '/blocks.0/intra_mossformer/EyeLike'.

To reproduce

My code is as follows:
def inference(args):
device = torch.device('cuda') if args.use_cuda==1 else torch.device('cpu')
print(device)
print('creating model...')
model = network_wrapper(args).se_network
model.to(device)

print('loading model ...')
reload_for_eval(model, args.checkpoint_dir, args.use_cuda,args.debug)
model.eval()
example_input = torch.randn((1,2,161,201)).to(device)
onnx_path = os.path.join(args.checkpoint_dir, 'model.onnx')
torch.onnx.export(
    model,
    example_input,
    onnx_path,
    export_params=True,  # 导出模型参数
    opset_version=13,  # ONNX opset 版本
    do_constant_folding=True,  # 是否执行常量折叠优化
    input_names=["input"],  # 输入节点名称
    output_names=["output"],  # 输出节点名称
    dynamic_axes={  # 动态轴支持
        "input": {0: "batch_size"},  # 允许输入的 batch_size 动态变化
        "output": {0: "batch_size"}
    }
)

The original code can be found in : https://github.com/modelscope/ClearerVoice-Studio/tree/main/train/speech_enhancement
the model.pt can be found in: https://huggingface.co/alibabasglab/MossFormerGAN_SE_16K
Thank for your reply

Urgency

No response

Platform

Linux

OS Version

VERSION="20.04.4 LTS (Focal Fossa)"

ONNX Runtime Installation

Built from Source

ONNX Runtime Version or Commit ID

onnx 1.13.0 onnxruntime 1.13.1

ONNX Runtime API

Python

Architecture

X64

Execution Provider

CUDA

Execution Provider Library Version

CUDA Version: 12.2

[justinchuby]

Please consider doing two things:

1. Update onnx, onnxruntime and PyTorch to the latest versions
2. For model export, consider testing with torch.onnx.export(..., dynamo=True, report=True) using the latest torch-nightly. Attach the generated report if there is an error. Thanks!