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
7.01k stars 571 forks source link

如何开启13B全量预训练? #484

Closed huwei02 closed 6 months ago

huwei02 commented 6 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-LLaMA-2 (7B/13B)

操作系统

Linux

详细描述问题

# 请在此处粘贴运行代码(请粘贴在本代码块里)

运行的是Chinese-LLaMA-2-13B

run_pt.sh中干掉了lora相关配置,但是跑起来,参数还是没有130亿,请问这个是什么原因?怎么处理呢? Number of trainable parameters = 572,784,640

依赖情况(代码类问题务必提供)

# 请在此处粘贴依赖情况(请粘贴在本代码块里)

sentencepiece 0.1.91

运行日志或截图

# 请在此处粘贴运行日志(请粘贴在本代码块里)

[INFO|trainer.py:1723] 2023-12-27 11:49:52,215 >> Running training [INFO|trainer.py:1724] 2023-12-27 11:49:52,215 >> Num examples = 445,413 [INFO|trainer.py:1725] 2023-12-27 11:49:52,215 >> Num Epochs = 1 [INFO|trainer.py:1726] 2023-12-27 11:49:52,215 >> Instantaneous batch size per device = 4 [INFO|trainer.py:1729] 2023-12-27 11:49:52,215 >> Total train batch size (w. parallel, distributed & accumulation) = 16 [INFO|trainer.py:1730] 2023-12-27 11:49:52,215 >> Gradient Accumulation steps = 4 [INFO|trainer.py:1731] 2023-12-27 11:49:52,215 >> Total optimization steps = 27,838 [INFO|trainer.py:1732] 2023-12-27 11:49:52,217 >> Number of trainable parameters = 572,784,640

huwei02 commented 6 months ago

done,已解决