ymcui / Chinese-LLaMA-Alpaca-2

中文LLaMA-2 & Alpaca-2大模型二期项目 + 64K超长上下文模型 (Chinese LLaMA-2 & Alpaca-2 LLMs with 64K long context models)
Apache License 2.0
7.06k stars 578 forks source link

Chinese-Alpaca-2的7B指令精调,用作分类,微调的结果accuracy为什么16bit比8bit只高0.001,另外13B模型不升反降? #498

Closed feifei05 closed 8 months ago

feifei05 commented 9 months ago

提交前必须检查以下项目

问题类型

模型训练与精调

基础模型

Chinese-Alpaca-2 (7B/13B)

操作系统

Linux

详细描述问题

# Chinese-Alpaca-2的7B指令精调,用作分类,输出答案在10个中文字以内,微调100epochs的结果accuracy为什么16bit比8bit只高0.001,另外13B模型不升反降,这问题主要从哪些角度考虑?

依赖情况(代码类问题务必提供)

# transformers=4.34.0
# sentencepiece= 0.1.99

运行日志或截图

# 7B的8bit微调100epochs分类任务,测试集结果Class,precision,recall,f1-score,support
weighted avg,0.6490202416435714,0.6545275590551181,0.6430212355260709,2032.0

7B的16bit微调100epochs分类任务,测试集结果Class,precision,recall,f1-score,support
weighted avg,0.6485059628314677,0.655511811023622,0.6441143713844812,2032.0
ymcui commented 8 months ago

特定任务问题还请自行debug

github-actions[bot] commented 8 months 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 consideration.

github-actions[bot] commented 8 months ago

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