THUDM / CogVLM2

GPT4V-level open-source multi-modal model based on Llama3-8B
Apache License 2.0
2.02k stars 134 forks source link

Segmentation fault #75

Closed wuxiaolianggit closed 3 months ago

wuxiaolianggit commented 4 months ago

System Info / 系統信息

cuda11.8

Who can help? / 谁可以帮助到您?

No response

Information / 问题信息

Reproduction / 复现过程

2024-06-02 09:26:33 - Your app is available at http://localhost:8000 2024-06-02 09:26:52 - Translation file for zh-CN not found. Using default translation en-US. 2024-06-02 09:26:53 - Translated markdown file for zh-CN not found. Defaulting to chainlit.md. 2024-06-02 09:26:56 - Translation file for zh-CN not found. Using default translation en-US. 2024-06-02 09:27:17 - Translation file for zh-CN not found. Using default translation en-US. 2024-06-02 09:27:25 - Translation file for zh-CN not found. Using default translation en-US. 2024-06-02 09:27:39 - Translation file for zh-CN not found. Using default translation en-US. Segmentation fault

Expected behavior / 期待表现

如何解决这个问题呢

zRzRzRzRzRzRzR commented 4 months ago

啥配置,然后triton最好是用cuda 12的,11.8可能出现驱动 triton cuda不适配问题

cuppersd commented 3 months ago

啥配置,然后triton最好是用cuda 12的,11.8可能出现驱动 triton cuda不适配问题

后期能支持一下cuda11.8或者11.4吗?我的显卡比较老了,跪求支持低版本的cuda

zRzRzRzRzRzRzR commented 3 months ago

11.8是可以的,但是你需要自己寻找一下triton,不够老卡还会面临无法使用bf16推理导致极小概率出现模型溢出的问题。总体使用效果差距很小,可以在11.8上尝试。