search

CLUEbenchmark / CLUE

中文语言理解测评基准 Chinese Language Understanding Evaluation Benchmark: datasets, baselines, pre-trained models, corpus and leaderboard
http://www.CLUEbenchmarks.com
3.97k stars 540 forks source link

WSC模型使用Chinese-wwm-bert预训练模型accuracy只有50% #96

Closed viviqi closed 4 years ago

viviqi commented 4 years ago

WSC模型使用Chinese-wwm-bert预训练模型accuracy只有50%，看发布的基线模型有62%左右，大家有遇到这类精度相差很大的情况吗？求教

lonePatient commented 4 years ago

@viviqi 可参考下baseline里面的一些参数数值，正常情况下bert-base模型accuracy在72~76之间左右