Giters

KLUE-benchmark / KLUE

📖 Korean NLU Benchmark

Home Page:https://klue-benchmark.com

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

NLI performance

LittlePea13 opened this issue · comments

commented

Hi there,
Could you share the hyperparameters for the NLI task using Roberta-large? I am unable to replicate the 0.89 performance, and after trying the different setups I get a performance around 0.86 tops. Also, is there any plan to release checkpoints for each of the tasks?

Thanks!

commented

Hi, @LittlePea13 Thanks for your interest on KLUE!

KLUE team is planning to release finetuning pipeline and training script at the end of July. You would get detailed training hyperparameters after the release.
About releasing checkpoints, frankly speaking, we haven't think of it, since our results are "baselines" - we didn't put a lot of effort to get maximum finetuning results. But I will think about it! Thx