Using llava-next-110b for inference, the return is ['']

LLaVA-VL / LLaVA-NeXT

Apache License 2.0

2.52k stars 186 forks source link

Using llava-next-110b for inference, the return is [''] #63

Open J0eky opened 3 months ago

J0eky commented 3 months ago

I am seeking to utilize the llava-next-110b model to generate complex descriptions for an input image. Having tested numerous images, I've encountered instances where the inference result is empty, indicated as ['']. Could you please enlighten me on the possible reasons behind receiving an output of [''], thank you.

greenhestu commented 2 months ago

I think I observed the same behavior when the model had special token as output (e.g. <|im_start|> ). What was your prompt and implementation?