Spaces:
Running on Zero
Running on Zero
File size: 1,258 Bytes
e8c5cc2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 | """Structural types for the duck-typed interfaces in this project.
Several functions accept "anything that can score pairs" or "anything that can
search" - the torch reranker and the ONNX reranker are interchangeable, and that
interchangeability is the whole point of the row-6 benchmark. Protocols express
that without forcing a shared base class.
"""
from __future__ import annotations
from typing import Protocol
import numpy as np
from src.data.loader import Corpus
Run = dict[str, dict[str, float]]
class Scorer(Protocol):
"""Anything that turns (query, passage) pairs into relevance scores."""
def score(self, pairs: list[tuple[str, str]]) -> np.ndarray: ...
class RerankerLike(Scorer, Protocol):
"""A Scorer that can also reorder a whole retrieval run.
Implemented by both src.rerank.model.Reranker (torch) and
src.rerank.onnx_runtime.OnnxReranker (onnxruntime).
"""
def rerank(
self,
queries: dict[str, str],
run: Run,
corpus: Corpus,
top_k: int = 50,
) -> tuple[Run, list[float]]: ...
class SearchRetriever(Protocol):
"""A first-stage retriever: BM25Retriever or DenseRetriever."""
def search(self, queries: dict[str, str], top_k: int = 50) -> Run: ...
|