search

salesforce / LAVIS

LAVIS - A One-stop Library for Language-Vision Intelligence
BSD 3-Clause "New" or "Revised" License
10k stars 975 forks source link

No output of InstructBLIP-vicuna #316

Open IQ250 opened 1 year ago

IQ250 commented 1 year ago

I run InstructBLIP successfully when LLM is flant5xl or flant5xxl, but when I switch LLM as vicuna-7b-v1.1, the output is a string of nothing(['']). Actually, when I use vicuna-7b-v0, there are some reasonable outputs (like 'the image features ...', 'in this image, there is ...' ), but sometimes repeated answer or '### Assistant: ....' happens.

LiJunnan1992 commented 1 year ago

Please use Vicuna v1.1 instead of v0.

Related issue #297