0.77 dev accuracy on quora question pairs, but 0.91 is the state of the art

openai / finetune-transformer-lm

Code and model for the paper "Improving Language Understanding by Generative Pre-Training"

MIT License

2.15k stars 503 forks source link

Closed weizier closed 6 years ago

weizier commented 6 years ago

I ran quora_similarity.py, and didn't modify any codes, then I got 0.77 accuracy for 0.4 class balance, any ideas for this? Thanks.

madisonmay commented 6 years ago

I think this issue can be closed @Newmu since it was intended to be opened on https://github.com/IndicoDataSolutions/finetune.