kosmosg 0.0.4

kosmosg - Pytorch

KosmosG

My implementation of the model KosmosG from "KOSMOS-G: Generating Images in Context with Multimodal Large Language Models"

Installation

pip install kosmosg

Usage

import torch
from kosmosg.main import KosmosG

# usage
img = torch.randn(1, 3, 256, 256)
text = torch.randint(0, 20000, (1, 1024))

model = KosmosG()
output = model(img, text)
print(output)

Architecture

text, image => KosmosG => text tokens with multi modality understanding

License

MIT

Todo

• Create Aligner in pytorch
• Create Diffusion module
• Integrate these pieces
• Create a training script