DeepSeek-VL icon indicating copy to clipboard operation
DeepSeek-VL copied to clipboard

多图的数量上限

Open JUNJIE99 opened this issue 1 year ago • 2 comments
trafficstars

请问DeepSeek-VL最多支持多少张图片输入?

JUNJIE99 avatar May 30 '24 14:05 JUNJIE99

请问DeepSeek-VL最多支持多少张图片输入?

测试过6-7张没出现问题。

Benjamin-eecs avatar May 30 '24 14:05 Benjamin-eecs

@Benjamin-eecs 你好,在使用‘deepseek-ai/deepseek-vl-7b-chat’时,输入8张图像发现输出是混乱的,请问可能是什么原因呢?代码参照simple-inference-example

输入如下:

Image 1 is <image_placeholder>. Image 2 is <image_placeholder>. Image 3 is <image_placeholder>. Image 4 is <image_placeholder>. Image 5 is <image_placeholder>. Image 6 is <image_placeholder>. Image 7 is <image_placeholder>. Image 8 is <image_placeholder>. Question: What action do these images show? A. spinning remote that quickly stops spinning B. holding toy next to remote C. moving remote and small remote away from each other D. moving remote and remote away from each other

这时的token数量是4749

输出如下:

'\nand your task.'

Terminal-K avatar Dec 27 '24 07:12 Terminal-K