search

SakanaAI / evolutionary-model-merge

Official repository of Evolutionary Optimization of Model Merging Recipes
Apache License 2.0
1.21k stars 88 forks source link

Can't response three traffic lights' colors in one image using your EvoVLM #10

Open yiyiwwang opened 4 months ago

yiyiwwang commented 4 months ago

Hello, I tested the image in your paper Table 6 Example 3 with your model EvoVLM-7b, I got the following answer, not the same to your good results with (A)(B)(C) in the paper. image

image

I have two questions:

  1. The answer seems not recognize the three subimage (A)(B)(C), and do not reply three colors. Why?
  2. Why does the answer repeat the question and response many times?

Thank you very much.