For example, when I am attaching an image file to my request “Translate File” it wants to call function extract_text_from_image or translateText. We started to experience this behavior very recently, it used to work fine before.
Expected behavior: it should not try to call extract_text_from_image/translateText functions, but should recognize text on images using model internal multi modal capabilities.
It is not reproducible all the time, but happens now in 50-60% percent of cases.