[Python] Advise needed on training a domain-specific QA model for radiation therapy

I have assembled a corpus of radiation therapy materials, including scanned textbooks (PDFs), structured Q&A pairs, and other reference documents. I want to build an AI system that can accurately answer domain-specific questions by leveraging this corpus.

I’m trying to decide between two approaches:

Fine-tuning a language model on my domain-specific corpus Using Retrieval-Augmented Generation (RAG) to pull relevant context at inference time

I have the following question

1. What are the pros and cons of each approach for this use case?
2. Which of the 2 approach is more Accurate and reliable?
3. Also, IF following Fine-tuning which model would be better for this approach?
4. Roughly how much domain-specific data is needed to achieve good QA performance with fine-tuning?
5. Are there any better alternatives than the above two?

Continue reading...