Skip to content

Llm

LLMReranking

Bases: BaseReranking

Source code in libs\kotaemon\kotaemon\indices\rankings\llm.py 
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
class LLMReranking(BaseReranking):
    llm: BaseLLM
    prompt_template: PromptTemplate = PromptTemplate(template=RERANK_PROMPT_TEMPLATE)
    top_k: int = 3
    concurrent: bool = True

    def run(
        self,
        documents: list[Document],
        query: str,
    ) -> list[Document]:
        """Filter down documents based on their relevance to the query."""
        filtered_docs = []
        output_parser = BooleanOutputParser()

        if self.concurrent:
            with ThreadPoolExecutor() as executor:
                futures = []
                for doc in documents:
                    _prompt = self.prompt_template.populate(
                        question=query, context=doc.get_content()
                    )
                    futures.append(executor.submit(lambda: self.llm(_prompt).text))

                results = [future.result() for future in futures]
        else:
            results = []
            for doc in documents:
                _prompt = self.prompt_template.populate(
                    question=query, context=doc.get_content()
                )
                results.append(self.llm(_prompt).text)

        # use Boolean parser to extract relevancy output from LLM
        results = [output_parser.parse(result) for result in results]
        for include_doc, doc in zip(results, documents):
            if include_doc:
                filtered_docs.append(doc)

        # prevent returning empty result
        if len(filtered_docs) == 0:
            filtered_docs = documents[: self.top_k]

        return filtered_docs

run 

1
run(documents, query)

Filter down documents based on their relevance to the query.

Source code in libs\kotaemon\kotaemon\indices\rankings\llm.py 
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
def run(
    self,
    documents: list[Document],
    query: str,
) -> list[Document]:
    """Filter down documents based on their relevance to the query."""
    filtered_docs = []
    output_parser = BooleanOutputParser()

    if self.concurrent:
        with ThreadPoolExecutor() as executor:
            futures = []
            for doc in documents:
                _prompt = self.prompt_template.populate(
                    question=query, context=doc.get_content()
                )
                futures.append(executor.submit(lambda: self.llm(_prompt).text))

            results = [future.result() for future in futures]
    else:
        results = []
        for doc in documents:
            _prompt = self.prompt_template.populate(
                question=query, context=doc.get_content()
            )
            results.append(self.llm(_prompt).text)

    # use Boolean parser to extract relevancy output from LLM
    results = [output_parser.parse(result) for result in results]
    for include_doc, doc in zip(results, documents):
        if include_doc:
            filtered_docs.append(doc)

    # prevent returning empty result
    if len(filtered_docs) == 0:
        filtered_docs = documents[: self.top_k]

    return filtered_docs