Sliding Window/Convolution over Large text input

For example, given a large python file or textbook chapter, use a sliding window over the file where each block size gets inputted into the model API until EOF and then you concatenate all outputs. Has anyone tried this?


Yes. It’s not bad. Another approach is to use an extractive summarizer for the transection, then use GPT-3 to write abstracts for each. I also think it should be possible to keep a running summary of the document as a whole (call it S) and update S every time you read a chunk. This is basically what the human brain does.