RAGchain.reranker.llm package

Submodules

RAGchain.reranker.llm.llm module

The original code is from [RankGPT](https://github.com/sunnweiwei/RankGPT). I modified the code to fit the RAGchain framework.

class RAGchain.reranker.llm.llm.LLMReranker(model_name: str = 'gpt-3.5-turbo', api_base: str | None = None, *args, **kwargs)

Bases: BaseReranker

LLMReranker is a reranker based on RankGPT (https://github.com/sunnweiwei/RankGPT). The LLM rerank the passages by question. This reranker only supports the OpenAI models only.

find_passages(target_passages: List[Passage], content: str)
invoke(input: Input, config: RunnableConfig | None = None) → Output

Transform a single input into an output. Override to implement.

Args:

input: The input to the runnable. config: A config to use when invoking the runnable.

The config supports standard keys like ‘tags’, ‘metadata’ for tracing purposes, ‘max_concurrency’ for controlling how much work to do in parallel, and other keys. Please refer to the RunnableConfig for more details.

Returns:

The output of the runnable.

make_item(query: str, passages: List[Passage]) → dict
make_passages(items: dict, original_passages: List[Passage]) → List[Passage]
rerank(query: str, passages: List[Passage]) → List[Passage]

Reranks a list of passages based on a specific ranking algorithm.

Parameters:

  • passages (List[Passage]) – A list of Passage objects representing the passages to be reranked.

  • query – str: The query that was used for retrieving the passages.

Returns:

The reranked list of passages.

Return type:

List[Passage]

rerank_sliding_window(query: str, passages: List[Passage], window_size: int) → List[Passage]

Reranks a list of passages based on a specific ranking algorithm with sliding window. This function is useful when the model input token size is limited like LLMs.

Parameters:

  • passages – (List[Passage]): The list of passages to be reranked.

  • query – str: The query that was used for retrieving the passages.

  • window_size – (int): The size of the sliding window used for reranking.

Returns:

List[Passage]: The reranked list of passages.

RAGchain.reranker.llm.rank_gpt module

This code is from RankGPT repo and modified a little bit for integration. Please go to https://github.com/sunnweiwei/RankGPT if you need more information.

class RAGchain.reranker.llm.rank_gpt.SafeOpenai(keys=None, start_id=None, proxy=None, api_base: str | None = None)

Bases: object

chat(*args, return_text=False, reduce_length=False, **kwargs)
text(*args, return_text=False, reduce_length=False, **kwargs)
RAGchain.reranker.llm.rank_gpt.clean_response(response: str)
RAGchain.reranker.llm.rank_gpt.create_permutation_instruction(item=None, rank_start=0, rank_end=100, model_name='gpt-3.5-turbo')
RAGchain.reranker.llm.rank_gpt.get_post_prompt(query, num)
RAGchain.reranker.llm.rank_gpt.get_prefix_prompt(query, num)
RAGchain.reranker.llm.rank_gpt.max_tokens(model)
RAGchain.reranker.llm.rank_gpt.num_tokens_from_messages(messages, model='gpt-3.5-turbo-0301')

Returns the number of tokens used by a list of messages.

RAGchain.reranker.llm.rank_gpt.permutation_pipeline(item=None, rank_start=0, rank_end=100, model_name='gpt-3.5-turbo', api_key=None, api_base=None)
RAGchain.reranker.llm.rank_gpt.receive_permutation(item, permutation, rank_start=0, rank_end=100)
RAGchain.reranker.llm.rank_gpt.remove_duplicate(response)
RAGchain.reranker.llm.rank_gpt.run_llm(messages, api_key=None, api_base: str | None = None, model_name='gpt-3.5-turbo')
RAGchain.reranker.llm.rank_gpt.sliding_windows(item=None, rank_start=0, rank_end=100, window_size=20, step=10, model_name='gpt-3.5-turbo', api_key=None, api_base=None)

Module contents