How to use max_length (number max of tokens) as an argument?

[piegu]

Hi,

max_length is defined here with the value 32.

How to modify it?

Thank you.

[blakechi]

Hello @piegu , thank you for raising up this issue.

So the max_length will be set to the maximum possible length according to the model_body (sentence-transformers) you use. Please see here for details.

Hope this resolve your issue. If not, I think we can add one more argument when initializing SetFitModel with an assert checking whether the given max_length exceeds the acceptable one the model_body can handle.

[piegu]

Hi @blakechi,

Thank you for your reply.
However, I don't want to set max_length to the maximum possible length. I want to be able to set it to any value via an argument.

Example: texts in my dataset have a maximum of 128 tokens. Then, I want to be able to set this value as max_length.

Note: when using transformers from Hugging Face, I configure the max_length value in the tokenizer through the argument max_length.

[blakechi]

Sure, I think we can achieve it as what I mentioned here:

Hope this resolve your issue. If not, I think we can add one more argument when initializing SetFitModel with an assert checking whether the given max_length exceeds the acceptable one the model_body can handle.

@lewtun How do you think? I can open a PR for this 😀