Unlocking Causal Attention into Modality-Mutual Attention for Multimodal LLMs github.com 1 points by countWSS 7 hours ago
countWSS 7 hours ago AKI, a novel MLLM that unlocks causal attention into modality-mutual attention (MMA) to enable image tokens to attend to text tokens.
AKI, a novel MLLM that unlocks causal attention into modality-mutual attention (MMA) to enable image tokens to attend to text tokens.