bosquet.eval.qna-generator — io.github.zmedelis/bosquet 2025.03.28

answering-prompt clj

(answering-prompt question-count)

context-prompt-block clj

document->dataset clj

(document->dataset opts document-file dataset-file)

Given a document-file create a QnA dataset. Save it to dataset-file

format-constraints clj

generate-qna-dataset clj

(generate-qna-dataset {q-count query-count chunk-count max-chunks} document)

Generate a QnA dataset. First a document will be loaded and split into sentence chunks. The size of the chunk is a function of how many questions we are generating per chunk.

Passed in options control the scope of the QnA generation process:

• query-count: how many queries per chunk to generate
• max-chunks: how many chunks to take, nil will take them all

load-qna-dataset clj

(load-qna-dataset file-name)

Load eval tuples from file (created via save-dataset)

max-chunks clj

qna->eval-dataset clj

(qna->eval-dataset ds-id ds-data)

Convert QnA data to a dataset format - a list of question to answer tuples

query-count clj

question-building-prompts clj

(question-building-prompts question-count)

question-set clj

response-set clj

save-dataset clj

(save-dataset ds-file ds)