[ONNXRuntimeError] : 9 : NOT_IMPLEMENTED : Could not find an implementation for the node ArgMax_1094:ArgMax(11)

[dengfenglai321]

hi i run
onnx_session1 = onnxruntime.InferenceSession("./pretrained/textmodel.onnx")
and generate error as below:

File "D:\anaconda3\lib\site-packages\onnxruntime\capi\onnxruntime_inference_collection.py", line 312, in _create_inference_session
sess.initialize_session(providers, provider_options)
onnxruntime.capi.onnxruntime_pybind11_state.NotImplemented: [ONNXRuntimeError] : 9 : NOT_IMPLEMENTED : Could not find an implementation for the node ArgMax_1094:ArgMax(11)

ONNX - 1.7.0
onnxruntime - 1.7.0

could you help me to solve this problem?

and the onnx model i transfered here

[ytaous]

Hi, can u pls check what's the type you are using for ArgMax. Likely it's not supported for opset 11.
ref - #9760

search for ArgMax for either CPU/CUDA provider:
ref - https://github.com/microsoft/onnxruntime/blob/master/docs/OperatorKernels.md

If CUDA, there's an PR in progress - #9700

[dengfenglai321]

Hi, can u pls check what's the type you are using for ArgMax. Likely it's not supported for opset 11. ref - #9760

search for ArgMax for either CPU/CUDA provider: ref - https://github.com/microsoft/onnxruntime/blob/master/docs/OperatorKernels.md

If CUDA, there's an PR in progress - #9700

the type using for ArgMax is int64. and only can be int64, could yuo give me some advices?
which opset can support int64 for ArgMax?

[dengfenglai321]

I solve it , only change ArgMax's input to int32, everything else stays the same, and it ok!!

[greyovo]

To make it clearer: it's a matter of converting the input vector to int32 when converting the model, and then executing torch.onnx.export, i.e.:

def export_text_encoder():
    text_encoder.eval()
    text = "A Diagram"
    input_tensor: Tensor = tokenizer(text)
    input_tensor = input_tensor.to(torch.int32)   # <---- here!
    model_text ='mobileclip-text-encoder.onnx'
    torch.onnx.export(text_encoder, input_tensor, model_text)