Three fine-tuning directions: simple medical question-and-answer clinical question-and-answer imaging
general process: 1 data set preparation 2 model base selection 3 fine-tuning 4 case disassembly
1 Dataset preparation: two types, one text and one image
extended, multimodal
2 Model base selection
Multimodal processing of all videos, texts, and digital humans will be threatened, and the digital human race track.
Multimodal processing will greatly promote the medical field.
The multimodal architecture of
the entire large image model is not mature at this stage. There is no standard plan for the plan of each doctor, and there is no way to be universal, and it will not be replaced by a general large model. Although the diagnosis may be the same, the treatment plan is different, and only the stage of diagnosis and treatment can be presented. Increase the knowledge base to do some knowledge alignment.
3 fine-tuning
