Issue converting owlv2 model to ONNX format about optimum HOT 4 CLOSED

Hi @n9s8a the message ValueError: Trying to export a owlv2 model, that is a custom or unsupported architecture, but no custom onnx configuration was passed as custom_onnx_configs. Please refer to https://huggingface.co/docs/optimum/main/en/exporters/onnx/usage_guides/export_a_model#custom-export-of-transformers-models for an example on how to export custom models. Please open an issue at https://github.com/huggingface/optimum/issues if you would like the model type owlv2 to be supported natively in the ONNX export.

means that owlv2 ONNX export is not yet supported in Optimum. I'll add it to my todo

from optimum.

Hi @fxmarty, Thankyou for your response and for clarifying the situation regarding the owlv2 ONNX export. I'm using owlv2 model in my project. Could you please provide an estimated timeline for when this support might be available?

Additionally, in the meantime, would it be feasible to use any external methods or workarounds to achieve ONNX conversion for the owlv2 model?

Your insights on these matters would be greatly appreciated.

from optimum.

@n9s8a turns out there is an open PR for the ONNX export of owlv2: #1689. It should be a matter of days for it to be merged!

from optimum.

Hi @n9s8a, #1689 is merged. I also recommend you to use Transformers from source for now as huggingface/transformers#29326 fixed a bug in the ONNX export of owlvit/owlv2 and is not yet in a release.

optimum-cli export onnx -m google/owlv2-base-patch16 --task 'zero-shot-object-detection' --framework 'pt' owlv2_onnx

now works.

from optimum.