Issue Checkpoints

[DavideNapolitano]

Hi,
I'm trying to run the inference.py file, but I think some files are missing.
In detail, eva_vit_g.pth and instruct_blip_vicuna13b_trimmed.pth files are referenced in your code (loaded from local folders) but missing in the repo.
Can you update the code and share the remaining files?
Thank you for your help!

[BriansIDP]

VIT: https://storage.googleapis.com/sfr-vision-language-research/LAVIS/models/BLIP2/eva_vit_g.pth
Instructblip: https://storage.googleapis.com/sfr-vision-language-research/LAVIS/models/InstructBLIP/instruct_blip_vicuna13b_trimmed.pth

[patrickamadeus]

Hi there @BriansIDP, May I ask which Vicuna-13b checkpoint you guys used? I have tried vicuna-1.5-13b but it gave pretty bad result.

[BriansIDP]

Hi. Please refer to this repo for the latest model and instructions: https://github.com/bytedance/SALMONN/tree/main/video_salmonn. Thank you!

[Aafiya-H]

@patrickamadeus Were you able to find the vicuna checkpoint? Also which one is the original model weight (lmsys/vicuna-13b-v1.5 or lmsys/vicuna-13b-v1.5-16k)