QwenLM / Qwen-VL

The official repo of Qwen-VL (通义千问-VL) chat & pretrained large vision language model proposed by Alibaba Cloud.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

[BUG] qwen-vl 第一阶段训练loss下降后升高了,模型训崩了

liuheng0111 opened this issue · comments

是否已有关于该错误的issue或讨论? | Is there an existing issue / discussion for this?

  • 我已经搜索过已有的issues和讨论 | I have searched the existing issues / discussions

该问题是否在FAQ中有解答? | Is there an existing answer for this in FAQ?

  • 我已经搜索过FAQ | I have searched FAQ

当前行为 | Current Behavior

模拟qwen-vl的第一阶段训练,采用了相同的训练参数,也对vit采用了learning rate decay,训练过程中loss先慢慢下降,但后面升高了,升高之后发现模型训崩了,排除了训练数据问题,不知道是哪里的问题,放开vit训练久了之后就崩了?
image

期望行为 | Expected Behavior

No response

复现方法 | Steps To Reproduce

No response

运行环境 | Environment

- OS:
- Python:
- Transformers:
- PyTorch:
- CUDA (`python -c 'import torch; print(torch.version.cuda)'`):

备注 | Anything else?

No response

大佬有查到原因吗?