mmca 0.0.2

MMCA - Pytorch

Multi-Modal Causal Attention

The open source community's implementation of the all-new Multi-Modal Causal Attention from "DeepSpeed-VisualChat: Multi-Round Multi-Image Interleave Chat via Multi-Modal Causal Attention"

Paper Link

Appreciation

• Lucidrains
• Agorians

Install

pip install mmca

Usage

import torch 
from mmca.main import MultiModalCausalAttention


attn = MultiModalCausalAttention(dim=512, heads=8)

x = torch.randn(1, 10, 512)
y = torch.randn(1, 20, 512)

#create a mask for the text
# mask = torch.ones(1, 20).bool()

x, y = attn(x, y)

print(x)
# print(y)

Architecture

Todo

License

MIT

Citations