已经使用LoRa技术及进行了模型微调,产生了很多文件,如何使用微调后的模型呢? #97
Replies: 7 comments 1 reply
-
同问,使用SFT或者LORA的脚本训练后,生产出的模型需要怎么进行Chat或者推理?似乎没办法再用model.chat的方式了,是不是需要对输入做一些前置处理? |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
15000step应该数据集一次还没迭代完吧?batch_size是1的话,并且使用一张卡。你的epoch=8是怎么算出来的?我现在也在微调,还没跑完,不知道结果怎么样呢。出来结果了我们可以对照一下 |
Beta Was this translation helpful? Give feedback.
-
这是我训练的模型,输出也很烂,不知道是为什么? |
Beta Was this translation helpful? Give feedback.
-
我是4张卡,--per_device_train_batch_size 2,lora的训练效果很差。 |
Beta Was this translation helpful? Give feedback.
-
有没有试过全量微调啊,用float32的话大概需要多少显存? |
Beta Was this translation helpful? Give feedback.
-
Feature request / 功能建议
使用LoRa技术微调,产生了下面的这些文件,有没有使用微调后的模型和原始模型一起进行预测的代码呢?
Beta Was this translation helpful? Give feedback.
All reactions