I want to set up a model to take two news articles and and summarize them into one. I want it follow a format of, x and y agree on these points, then x said this also while y felt this was important. I also want to use the GPT-J because it is flash and new, along with really good test scores. Or should i use an older more trained model like BART.
Second, is there a easy way im missing, or do i have to pass it to the model about 4 times. My thought is to extract into bullet points each article. Then organize and group them with a classifier. This then hits a either abstract summarize or text generator actually write the new summery. I think it will dodge some max token issues by how its naturally is chunked out.
Does this makes sense and if so please let me know what libraries and models to specifically look and and more importantly avoid.