search

princeton-nlp / SimCSE

[EMNLP 2021] SimCSE: Simple Contrastive Learning of Sentence Embeddings https://arxiv.org/abs/2104.08821
MIT License
3.31k stars 502 forks source link

About model format conversion #262

Closed fzxxg closed 6 months ago

fzxxg commented 7 months ago

I didn’t reproduce using your code, can I use your evaluation file? But I noticed that your model format needs to be converted. If I don’t convert it, it can still run successfully, but what impact will this have? The following is the code I executed for the evaluation file: python evaluation.py \ --model_name_or_path bert-base-uncased-sts \ --pooler cls_before_pooler \ --task_set sts/transfer \ --mode test

fzxxg commented 7 months ago

I only saved the BERT parameters in the model.

seanswyi commented 6 months ago

I believe the output format is safetensors. There should be any actual impact.

gaotianyu1350 commented 6 months ago

Hi @fzxxg , if you use our training code and don't convert it, the checkpoint won't be able to perform as well as the paper reported; however, if you are evaluating other models, it should be fine (as long as the inference for those models are strictly the same as how they are supposed to be used.

fzxxg commented 6 months ago

Thank you for your reply. I will close this issue.