Using flauBERT for similarities between sentences about flaubert HOT 3 CLOSED

Hi @QuentinSpalla,

The current fine-tuning code of XLM does not save the best models (I'm sorry that you spent your time fine-tuning without having a trained model at the end). I did not realise this issue as I only needed the validation scores and prediction files.

I can add some code that allows saving the best model and loading it for computing embeddings. However, since I plan to migrate all the fine-tuning tasks to Hugging Face's transformers (for ease of comparison with other methods), it's difficult for me to do it now. Could you please wait to use transformers pipeline? I will push an update in the next few days for the XNLI task (with the weights of the fine-tuned model so that you won't have to train it again).

from flaubert.

HI @formiel

Thank you for your answer. Yes, it could be great using huggingface transformers for flauBERT.

Thanks

from flaubert.

@QuentinSpalla is it ok for you ? we can close this thread now I guess.

from flaubert.