feat:(integration): add MixedbreadAICrossEncoder for reranking documents

[adityaiiitmk]

This PR introduces the MixedbreadAICrossEncoder, a new cross-encoder integration that leverages the mixedbread-ai/mxbai-rerank-base-v2 model for reranking documents.
It provides a standardized interface for computing similarity scores between queries and documents, with optional score normalization.

Key Changes

• Added new class: MixedbreadAICrossEncoder

  • Inherits from BaseCrossEncoder.
  • Supports configurable model_name, model_kwargs, and normalize_scores.
  • Provides score() method to compute relevance scores for query-document pairs.
  • Includes normalization utility for scaling scores to [0, 1].

• Handles ImportError with a clear message and installation instruction for mxbai-rerank.

• Default model: mixedbread-ai/mxbai-rerank-base-v2.

Example Usage

from langchain_community.cross_encoders.mixedbreadai import MixedbreadAICrossEncoder

model_name = "mixedbread-ai/mxbai-rerank-base-v2"
model_kwargs = {'top_k': 10}
mb = MixedbreadAICrossEncoder(
    model_name=model_name,
    model_kwargs=model_kwargs
)

text_pairs = [
    ("What is climate change?", "Climate change leads to droughts and floods."),
    ("What is climate change?", "It refers to shifts in weather patterns.")
]
scores = mb.score(text_pairs)
print(scores)

Why This Change?

• Provides a seamless integration of Mixedbread’s reranker into the LangChain ecosystem.
• Enables more accurate document reranking in retrieval and search pipelines.
• Enhances flexibility with support for top_k, custom kwargs, and optional score normalization.

Dependencies

• Requires [mxbai-rerank](https://pypi.org/project/mxbai-rerank/) package.

  pip install mxbai-rerank