Source code for autorag.nodes.passagereranker.sentence_transformer

from typing import List, Tuple

import pandas as pd
import torch
from sentence_transformers import CrossEncoder
from tqdm import tqdm

from autorag.nodes.passagereranker.base import passage_reranker_node
from autorag.utils.util import flatten_apply, make_batch, select_top_k, sort_by_scores




@passage_reranker_node
def sentence_transformer_reranker(
	queries: List[str],
	contents_list: List[List[str]],
	scores_list: List[List[float]],
	ids_list: List[List[str]],
	top_k: int,
	batch: int = 64,
	sentence_transformer_max_length: int = 512,
	model_name: str = "cross-encoder/ms-marco-MiniLM-L-2-v2",
) -> Tuple[List[List[str]], List[List[str]], List[List[float]]]:
	"""
	Rerank a list of contents based on their relevance to a query using Sentence Transformer model.

	:param queries: The list of queries to use for reranking
	:param contents_list: The list of lists of contents to rerank
	:param scores_list: The list of lists of scores retrieved from the initial ranking
	:param ids_list: The list of lists of ids retrieved from the initial ranking
	:param top_k: The number of passages to be retrieved
	:param batch: The number of queries to be processed in a batch
	:param sentence_transformer_max_length: The maximum length of the input text for the Sentence Transformer model
	:param model_name: The name of the Sentence Transformer model to use for reranking
	    Default is "cross-encoder/ms-marco-MiniLM-L-2-v2"
	:return: tuple of lists containing the reranked contents, ids, and scores
	"""
	device = "cuda" if torch.cuda.is_available() else "cpu"
	model = CrossEncoder(
		model_name, max_length=sentence_transformer_max_length, device=device
	)

	nested_list = [
		list(map(lambda x: [query, x], content_list))
		for query, content_list in zip(queries, contents_list)
	]
	rerank_scores = flatten_apply(
		sentence_transformer_run_model, nested_list, model=model, batch_size=batch
	)

	df = pd.DataFrame(
		{
			"contents": contents_list,
			"ids": ids_list,
			"scores": rerank_scores,
		}
	)
	df[["contents", "ids", "scores"]] = df.apply(
		sort_by_scores, axis=1, result_type="expand"
	)
	results = select_top_k(df, ["contents", "ids", "scores"], top_k)

	del model
	if torch.cuda.is_available():
		torch.cuda.empty_cache()

	return (
		results["contents"].tolist(),
		results["ids"].tolist(),
		results["scores"].tolist(),
	)





def sentence_transformer_run_model(input_texts, model, batch_size: int):
	batch_input_texts = make_batch(input_texts, batch_size)
	results = []
	for batch_texts in tqdm(batch_input_texts):
		with torch.no_grad():
			pred_scores = model.predict(sentences=batch_texts, apply_softmax=True)
		results.extend(pred_scores.tolist())
	return results