[Python] Getting confidence scores for fields extracted from image-to-text LLM inference

I am extracting details from cheques using fine-tuned Qwen 2.5 VL 7B model. I want to have confidence scores for the fields as in azure document intelligence.

A few ways I think I can do it is using rule-based approach for each field. Another way is to train a classifier model for it. I am confused which method to choose. Is there any better way?

Continue reading...