An open-source implementation of LongVA for facilitating the large multi-modal model community.

Construct Datasets for Finetuning

{ "id": id of the video/image data, "video": dir to the video/image data, "conversations": [ { "from": "human", "value": "\n Instruction" }, { "from": "gpt", "value": "Output" }, { "from": "human", "value": "\n Instruction" }, { "from": "gpt", "value": "Output" } ] },