mixedbread Rerank Cookbook

mixedbread.ai has released three fully open-source reranker models under the Apache 2.0 license. For more in-depth information, you can check out their detailed blog post. The following are the three models:

1. mxbai-rerank-xsmall-v1
2. mxbai-rerank-base-v1
3. mxbai-rerank-large-v1

In this notebook, we’ll demonstrate how to use the mxbai-rerank-base-v1 model with the SentenceTransformerRerank module in LlamaIndex. This setup allows you to seamlessly swap in any reranker model of your choice using the SentenceTransformerRerank module to enhance your RAG pipeline.

Installation

!pip install llama-index
!pip install sentence-transformers

Set API Keys

import os

os.environ["OPENAI_API_KEY"] = "YOUR OPENAI API KEY"

from llama_index.core import (
    VectorStoreIndex,
    SimpleDirectoryReader,
)

from llama_index.core.postprocessor import SentenceTransformerRerank

Download Data

!mkdir -p 'data/paul_graham/'
!wget 'https://raw.githubusercontent.com/run-llama/llama_index/main/docs/docs/examples/data/paul_graham/paul_graham_essay.txt' -O 'data/paul_graham/paul_graham_essay.txt'

--2024-03-01 09:52:09--  https://raw.githubusercontent.com/run-llama/llama_index/main/docs/docs/examples/data/paul_graham/paul_graham_essay.txt
Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.110.133, 185.199.108.133, 185.199.109.133, ...
Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.110.133|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 75042 (73K) [text/plain]
Saving to: ‘data/paul_graham/paul_graham_essay.txt’

data/paul_graham/pa 100%[===================>]  73.28K  --.-KB/s    in 0.007s

2024-03-01 09:52:09 (9.86 MB/s) - ‘data/paul_graham/paul_graham_essay.txt’ saved [75042/75042]

Load Documents

documents = SimpleDirectoryReader("./data/paul_graham/").load_data()

Build Index

index = VectorStoreIndex.from_documents(documents=documents)

Define postprocessor for mxbai-rerank-base-v1 reranker

from llama_index.core.postprocessor import SentenceTransformerRerank

postprocessor = SentenceTransformerRerank(
    model="mixedbread-ai/mxbai-rerank-base-v1", top_n=2
)

Create Query Engine

We will first retrieve 10 relevant nodes and pick top-2 nodes using the defined postprocessor.

query_engine = index.as_query_engine(
    similarity_top_k=10,
    node_postprocessors=[postprocessor],
)

Test Queries

response = query_engine.query(
    "Why did Sam Altman decline the offer of becoming president of Y Combinator?",
)

print(response)

Sam Altman initially declined the offer of becoming president of Y Combinator because he wanted to start a startup focused on making nuclear reactors.

response = query_engine.query(
    "Why did Paul Graham start YC?",
)

print(response)

Paul Graham started YC because he and his partners wanted to create an investment firm where they could implement their own ideas and provide the kind of support to startups that they felt was lacking when they were founders themselves. They aimed to not only make seed investments but also assist startups with various aspects of setting up a company, similar to the help they had received from others in the past.