airaria / TextBrewer

A PyTorch-based knowledge distillation toolkit for natural language processing
http://textbrewer.hfl-rc.com
Apache License 2.0
1.59k stars 239 forks source link

请问作者有做中文数据增强吗? #40

Closed hahlw closed 3 years ago

hahlw commented 3 years ago

Tinybert(4L,"hidden_size": 312,"intermediate_size": 1200,)模型随机初始化,结果相差bert-base 9个点(acc83%降到74%)。这种结果正常吗?

airaria commented 3 years ago

什么任务,什么配置?

hahlw commented 3 years ago

什么任务,什么配置?

文本分类任务,单卡TITAN XP(12g) 训练了30个epoch,选用比较好的epoch测试是75.94%,bertbase是80.02%(之前的指标观测错了)这个指标算是正常吗?我加了attentionloss 但是loss是先上升后下降。这个正常吗?

airaria commented 3 years ago

看起来正常,如果没调过超参,再调一调应该还有上升空间

stale[bot] commented 3 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

stale[bot] commented 3 years ago

Closing the issue, since no updates observed. Feel free to re-open if you need any further assistance.