Unlocking Causal Attention into Modality-Mutual Attention for Multimodal LLMs (opens in new tab)

(github.com)

1 pointscountWSS1y ago1 comments

1 comments

1 comments · 1 top-level

AKI, a novel MLLM that unlocks causal attention into modality-mutual attention (MMA) to enable image tokens to attend to text tokens.

j / k navigate · click thread line to collapse