pharmapsychotic / clip-interrogator

Image to prompt with BLIP and CLIP
MIT License
2.58k stars 429 forks source link

How to speed up? Which torch version is suggested #106

Open FurkanGozukara opened 7 months ago

FurkanGozukara commented 7 months ago

I am testing Torch 2.0.1, 2.1 and 1.13.1

The speed changes but the results are also changing

For example 2.0.1 is slower but it ends quicker number of steps and produces better output

image

Vordlex commented 4 months ago

Any news?

FurkanGozukara commented 4 months ago

i made my own version with batch captioning working great

https://www.patreon.com/posts/sota-image-for-2-90744385

captioners_clip_interrogator_v2.zip LLaVA_auto_install_v3.zip Qwen-VL_v3.zip blip2_captioning_v1.zip CogVLM_v7.zip Kosmos-2_v5.zip

Vordlex commented 4 months ago

I'm happy that you got results, but for me it doesn't make sense to pay to test/use