Retrievers API¶

TempoEval includes a comprehensive retrieval module to help you build and benchmark temporal RAG systems.

Sparse Retrievers¶

The standard baseline. Uses pyserini or rank_bm25.

from tempoeval.retrieval import BM25Retriever
retriever = BM25Retriever()

State-of-the-art embedding models.

High performance dense retrieval.

from tempoeval.retrieval import BGERetriever
# Defaults to 'BAAI/bge-large-en-v1.5'
retriever = BGERetriever()

Contrastive training for dense retrieval.

from tempoeval.retrieval import ContrieverRetriever
retriever = ContrieverRetriever()

Load any model from Hugging Face.

from tempoeval.retrieval import SentenceTransformerRetriever
retriever = SentenceTransformerRetriever("sentence-transformers/all-mpnet-base-v2")

Nomic Embed Text v1 (Long context).

from tempoeval.retrieval import NomicRetriever

Instruction-finetuned embeddings.

from tempoeval.retrieval import InstructorRetriever
retriever = InstructorRetriever(instruction="Represent the query for retrieval:")

Embedding models from OpenAI.

from tempoeval.retrieval import AzureOpenAIRetriever
retriever = AzureOpenAIRetriever(
    deployment="text-embedding-3-small"
)

Cohere Command R / Embed v3.

from tempoeval.retrieval import CohereRetriever

Voyage AI embeddings (voyage-large-2).

from tempoeval.retrieval import VoyageRetriever

Using LLMs as retrievers (Score function or encoder).

from tempoeval.retrieval import QwenE5SFRetriever
retriever = QwenE5SFRetriever(model_id="qwen2") # or "e5", "sf"

Generative Representational Instruction Tuning.

from tempoeval.retrieval import GritLMRetriever

Diverse LLM-based retriever.

from tempoeval.retrieval import DiverRetriever