Skip to content

cyan2k/molmo-7b-bnb-4bit

Repository files navigation

Molmo-7B-O and Molmo-7B-D BnB 4bit quants

Huggingface:

https://huggingface.co/cyan2k/molmo-7B-O-bnb-4bit

Todo

  • Small Gradio/Streamlit/Shiny UI app

Create venv

python -m venv venv
(mac/linux)
source venv/bin/activate

(win)
 .\venv\Scripts\activate

Install dependencies

Fitting PyTorch for your cuda version

(cuda 12.4)
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu124

(cuda 12.1)
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu121

(cuda 11.8)
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118

Additional stuff

pip install einops tensorflow-cpu transformers accelerate bitsandbytes

NOTE: If you use the "no_tensorflow" version of the script above you do not need to install tensorflow. However, you MUST also replace the current version of image_preprocessing_molmo.py downloaded from the huggingface repo with the one in this repository above.

Run

python molmo-7B-O-bnb-4bit.py

# A dark gray cat with black stripes and a long tail is perched on a white carpet, its body oriented to the left while its head is turned to face the camera. The cat's striking yellow eyes with black pupils are prominently visible. Its ears are perked up, and its front legs are stretched out in front of it. The cat's tail extends off the right side of the image. Behind the cat, there is a white piano with a stack of books on top. An open book of sheet music is visible on the piano, with the pages spread out. The wall behind the piano is white, and the piano casts a shadow on it. The cat's whiskers are also noticeable, adding to its alert and attentive appearance.

python molmo-7B-D-bnb-4bit.py

NOTE: If you use the "no_tensorflow" version of the script above you must enter into the script the hardcoded path to the model on your computer as well as a hardcoded path to the image you want to process. As such, you must download the huggingface repo files first in order have a hardcoded path to specify.

About

4bit bitsandbytes quants of the best 7B vlms

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  

Languages