GATED XATTN-DENSE

GATED XATTN-DENSE是Flamingo模型中的一个cross-attention层，用于桥接不同模态信息之间的交互。这个层的训练是Flamingo模型中唯一需要进行的训练步骤，而vision encoder和LM则不需要额外的训练。通过利用CV领域和NLP领域的已有成果，Flamingo的缝合方案充分利用了这些成果，从而在模型训练成本上具有明显的优势。\[1\]
#### 引用[.reference_title]
- *1* *2* [SEEChat](https://blog.csdn.net/qq_29788741/article/details/131388003)[target="_blank" data-report-click={"spm":"1018.2226.3001.9630","extra":{"utm_source":"vip_chatgpt_common_search_pc_result","utm_medium":"distribute.pc_search_result.none-task-cask-2~all~insert_cask~default-1-null.142^v91^control_2,239^v3^insert_chatgpt"}} ] [.reference_item]
- *3* [360 人工智能研究院正式开源中文多模态对话模型 SEEChat](https://blog.csdn.net/csdnnews/article/details/131485747)[target="_blank" data-report-click={"spm":"1018.2226.3001.9630","extra":{"utm_source":"vip_chatgpt_common_search_pc_result","utm_medium":"distribute.pc_search_result.none-task-cask-2~all~insert_cask~default-1-null.142^v91^control_2,239^v3^insert_chatgpt"}} ] [.reference_item]
[ .reference_list ]