bge-reranker-v2-m3 - Pinecone Docs

When to use it

This is an open source, high performance, multilingual model. It works well on messy data and short queries expected to return medium-length passages of text (1-2 paragraphs).Reranking models are designed to provide superior accuracy over retriever models but are much slower, so this model shouldn’t be used with more than a few hundred documents. Due to the slowness of rerankers, we recommend using them in a two-stage retrieval system: use a retrieval to pull in a smaller number of documents from a larger database and then rerank the smaller number of documents using a reranker.

Inputs, limits, and pricing

Input	Limits	Price
Text	1024 tokens per query and document pair, 100 documents per request	$2.00 per 1k requests

Monthly and per-minute limits depend on your plan. See Rate limits.

Quickstart

Installation

pip install -U pinecone

Rerank

To use bge-reranker-v2-m3 for standalone reranking with the rerank endpoint, see Rerank results.

from pinecone import Pinecone

pc = Pinecone("API-KEY")

query = "Tell me about Apple's products"
results = pc.inference.rerank(
    model="bge-reranker-v2-m3",
    query=query,
    documents=[
"Apple is a popular fruit known for its sweetness and crisp texture.",	
"Apple is known for its innovative products like the iPhone.",
"Many people enjoy eating apples as a healthy snack.",
"Apple Inc. has revolutionized the tech industry with its sleek designs and user-friendly interfaces.",
"An apple a day keeps the doctor away, as the saying goes.",
    ],
    top_n=3,
    return_documents=True,
    parameters= {
        "truncate": "END"
    }
)

print(query)
for r in results.data:
  print(r.score, r.document["text"])

Parameters

Set rank_fields on the rerank request itself, and set the other parameters in its parameters object. This applies to both standalone reranking and integrated reranking.

Parameter	Type	Required/Optional	Description	Default
`truncate`	string	Optional	How to handle inputs longer than those supported by the model. Accepted values: `END` or `NONE`. `END` truncates the input sequence at the input token limit. `NONE` returns an error when the input exceeds the input token limit.	`NONE`
`rank_fields`	array of strings	Optional	The field to use for reranking. The model supports only a single rerank field.	`["text"]`

Good to know

This model returns a relevance score for a query and passage. A sigmoid function can map the relevance score to a float value in the range [0,1].

​When to use it

​Inputs, limits, and pricing

​Quickstart

​Installation

​Rerank

​Parameters

​Good to know