[Bug]: Qwen2.5-7B-Instruct 支持 128K tokens,为啥 config 里面写 32k呢,如果我基于qwen2.5 要训练一个大于 32k 的模型,需要怎么做呢 #1134
Unanswered
tao-githup
asked this question in
Q&A
Replies: 3 comments 2 replies
|
Please first read the README/Modelcard: https://huggingface.co/Qwen/Qwen2.5-7B-Instruct#processing-long-texts |
0 replies
|
@jklj077 这是推理的时候吧,如果是基于他做 sft,直接把max_position_embeddings 改成 128K 行吗,我的训练文本比较长 |
0 replies
|
2 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Model Series
Qwen2.5
What are the models used?
Qwen2.5-7B-Instruct
What is the scenario where the problem happened?
sft
Is this a known issue?
Information about environment
no
Log output
Description
Steps to reproduce
This happens to Qwen2.5-xB-Instruct-xxx and xxx.
The problem can be reproduced with the following steps:
All reactions