finetuning pytorch model on WMT

google-research / text-to-text-transfer-transformer

Code for the paper "Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer"

https://arxiv.org/abs/1910.10683

Apache License 2.0

6.19k stars 757 forks source link

finetuning pytorch model on WMT #489

Open rabeehkarimimahabadi opened 4 years ago

rabeehkarimimahabadi commented 4 years ago

Hi, I finetuned the pytorch model on WMT, from scratch, after 50000 iterations, I was getting 1 blue score.
I think the way data processing is done, does not match the way decoding works in huggingface model, resulting in low performance for the HF model. It would be great if this can be fixed. thanks.

rabeehk commented 4 years ago

Hi I am not sure if the issue is with setting, could you please run your codes on WMT and verify it is working? I tried multiple times, and the bleu score is hardly get to 1. thanks.