FurkanGozukara
commented
6 months ago Just made new update
16 February 2024 Update: 4-bit, 8-bit, 16-bit and 32-bit loading options added to the Kosmos-2 I think Kosmos-2 is the very best caption model if you have a lower VRAM GPU 4-bit of Kosmos 2 only uses 2 GB VRAM and 32-bit uses only 7.5 GB VRAM Please reinstall with newest Kosmos-2_v2.zip
After working whole day I finally coded it and published it.
Also I found a better prompt that captions better.
You can download auto installer at here : https://www.patreon.com/posts/90744385
The batch image captioning models we have right now as follows:
CogVML with quantization 4-bit, 8-bit, 16-bit LLaVA including 34b with quantization such as 4-bit, 8-bit, 16-bit Blip2 Models Clip Vision Models Kosmos-2 Model Kosmos-2 supports both single image captioning and also batch image captioning. I also did some research to find a good prompt.
1 click to install both on Windows, RunPod & Linux.
Generates its own venv so will never conflict with no any other app you have.
Here news about them : https://www.patreon.com/posts/sota-image-model-98499462