Why is the original model required?

[mgoldenbe]

I tried removing the original files from .cache and leaving only the .onnx file. However, when I ran the script it downloaded the original model again. How can I get rid of the dependency on the original model? In fact, given onnx, I should only need onnxruntime to make inferences. So, how do I get rid of the dependencies such as torch?

[Sayed-Ameer]

did you find a solution, for the whole model dependency?

[antonmosin]

This package not only converts the model to onnx format, but also instantiates it to run inference. To run inference, you need the tokenizer, which is expected in the cache_folder. The tokenizer is instantiated by the transformer library, which depends on torch, among other things.

[davidberenstein1957]

Sorry for the late response but I've got good news! Sentence Transformers has an awesome maintainer again @tomaarsen, along with awesome ONNX integrations through optimum, therefore I decided to archive this project. https://github.com/UKPLab/sentence-transformers. Take a look here https://www.philschmid.de/optimize-sentence-transformers.