fairseq How to convert a fairseq transformer model into ONNX format?

How to convert a fairseq transformer model into ONNX format?

Open wolfshow opened this issue 5 years ago • 10 comments

Any documentation or tools? Thanks!

Jul 23 '19 08:07 wolfshow

I would like to contribute if someone can guide me

Aug 08 '19 03:08 adikshit

Also interested in being able to do this..

Nov 07 '19 07:11 echan00

i need it too!tks! i write code to convert onnx but it is not work! `import torch import onnx

en2de = torch.hub.load('pytorch/fairseq', 'transformer.wmt19.en-de.single_model') output = en2de.translate('Hello world', beam=5) #print("This is output:", output) model = en2de.eval() # disable dropout #print(model)

dummy_input1 = torch.ones(1, 4, device='cpu', dtype=torch.long) dummy_input2 = torch.ones(1, device='cpu', dtype=torch.long) dummy_input3 = torch.ones(5, 5, device='cpu', dtype=torch.long) inputs = (dummy_input1, dummy_input2, dummy_input3)

input_names = ["src_tokens", "src_lengths", "prev_output_tokens"] #output_names = ["start_logits","end_logits"] torch.onnx.export(model, inputs, "fairseq_transformer.onnx", verbose=True, input_names=input_names, output_names=["output"], export_params=True, opset_version=13, do_constant_folding = True, )#operator_export_type=torch.onnx.OperatorExportTypes.ONNX_ATEN_FALLBACK #checker onnx_file = onnx.load("fairseq_transformer.onnx") onnx.checker.check_model(onnx_file)`

The error sheetshot is :

Jan 27 '22 06:01 rGitcy

This issue has been automatically marked as stale. If this issue is still affecting you, please leave any comment (for example, "bump"), and we'll keep it open. We are sorry that we haven't been able to prioritize it yet. If you have any new additional information, please include it with your comment!