Rerank QA Content

Description

The Rerank operator is used to reorder the list of relevant documents for a query. It uses the MS MARCO Cross-Encoders model to get the relevant scores and then reorders the documents.

Code Example

• Run with ops

from towhee import ops

op = ops.rerank(threshold=0)
res = op('What is Towhee?',
         ['Towhee is Towhee is a cutting-edge framework to deal with unstructure data.', 'I do not know about towhee', 'Towhee has many powerful operators.', 'The weather is good' ])

• Run a pipeline

from towhee import ops, pipe, DataCollection

p = (pipe.input('query', 'doc')
         .map(('query', 'doc'), ('doc', 'score'), ops.rerank(threshold=0))
         .flat_map(('doc', 'score'), ('doc', 'score'), lambda x, y: [(i, j) for i, j in zip(x, y)])
         .output('query', 'doc', 'score')
     )

DataCollection(p('What is Towhee?',
                 ['Towhee is Towhee is a cutting-edge framework to deal with unstructure data.', 'I do not know about towhee', 'Towhee has many powerful operators.', 'The weather is good' ])
              ).show()

Factory Constructor

Create the operator via the following factory method

towhee.rerank(model_name: str = 'cross-encoder/ms-marco-MiniLM-L-12-v2')

Parameters:

model_name: str

​ The model name of CrossEncoder, you can set it according to the Model List.

​ threshold: float

​ The threshold for filtering with score device: str

Interface

This operator is used to sort the documents of the query content and return the score, and can also set a threshold to filter the results.

Parameters:

query: str

The query content.

​ docs: list

A list of sentences to check the correlation with the query content.

Return: List[str], List[float]

The list of documents after rerank and the list of corresponding scores.

More Resources

• The guide to rerank-english-v3.0 | Cohere: rerank-english-v3.0: a reranking model for English documents and semi-structured data (JSON); context length: 4096 tokens.
• Optimizing RAG with Rerankers: The Role and Trade-offs - Zilliz blog: Rerankers can enhance the accuracy and relevance of answers in RAG systems, but these benefits come with increased latency and computational costs.
• What Are Rerankers and How They Enhance Information Retrieval - Zilliz blog: Rerankers are specialized components in information retrieval systems that perform a crucial second-stage evaluation of search results.
• Building an Intelligent QA System with NLP and Milvus - Zilliz blog: The Next-Gen QA Bot is here
• The guide to rerank-english-v2.0 | Cohere: rerank-english-v2.0: a reranking model for English language documents with a context length of 512 tokens.