opensearch-neural-sparse-encoding-doc

Select the model

The model should be selected considering search relevance, model inference and retrieval efficiency(FLOPS). We benchmark models' zero-shot performance on a subset of BEIR benchmark: TrecCovid,NFCorpus,NQ,HotpotQA,FiQA,ArguAna,Touche,DBPedia,SCIDOCS,FEVER,Climate FEVER,SciFact,Quora.

Overall, the v2 series of models have better search relevance, efficiency and inference speed than the v1 series. The specific advantages and disadvantages may vary across different datasets.

Overview

• Paper: Towards Competitive Search Relevance For Inference-Free Learned Sparse Retrievers
• Fine-tuning sample: opensearch-sparse-model-tuning-sample

This is a learned sparse retrieval model. It encodes the documents to 30522 dimensional sparse vectors. For queries, it just use a tokenizer and a weight look-up table to generate sparse vectors. The non-zero dimension index means the corresponding token in the vocabulary, and the weight means the importance of the token. And the similarity score is the inner product of query/document sparse vectors.

The training datasets includes MS MARCO, eli5_question_answer, squad_pairs, WikiAnswers, yahoo_answers_title_question, gooaq_pairs, stackexchange_duplicate_questions_body_body, wikihow, S2ORC_title_abstract, stackexchange_duplicate_questions_title-body_title-body, yahoo_answers_question_answer, searchQA_top5_snippets, stackexchange_duplicate_questions_title_title, yahoo_answers_title_answer.

OpenSearch neural sparse feature supports learned sparse retrieval with lucene inverted index. Link: https://opensearch.org/docs/latest/query-dsl/specialized/neural-sparse/. The indexing and search can be performed with OpenSearch high-level API.

Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers.sparse_encoder import SparseEncoder

# Download from the 🤗 Hub
model = SparseEncoder("opensearch-project/opensearch-neural-sparse-encoding-doc-v2-distill")

query = "What's the weather in ny now?"
document = "Currently New York is rainy."

query_embed = model.encode_query(query)
document_embed = model.encode_document(document)

sim = model.similarity(query_embed, document_embed)
print(f"Similarity: {sim}")
# Similarity: tensor([[17.5307]])

decoded_query = model.decode(query_embed)
decoded_document = model.decode(document_embed)

for i in range(len(decoded_query)):
    query_token, query_score = decoded_query[i]
    doc_score = next((score for token, score in decoded_document if token == query_token), 0)
    if doc_score != 0:
        print(f"Token: {query_token}, Query score: {query_score:.4f}, Document score: {doc_score:.4f}")

# Similarity: tensor([[17.5307]], device='cuda:0')
# Token: ny, Query score: 5.7729, Document score: 1.4109
# Token: weather, Query score: 4.5684, Document score: 1.4673
# Token: now, Query score: 3.5895, Document score: 0.7473

Usage (HuggingFace)

This model is supposed to run inside OpenSearch cluster. But you can also use it outside the cluster, with HuggingFace models API.

import json
import itertools
import torch

from transformers import AutoModelForMaskedLM, AutoTokenizer

# get sparse vector from dense vectors with shape batch_size * seq_len * vocab_size
def get_sparse_vector(feature, output):
    values, _ = torch.max(output*feature["attention_mask"].unsqueeze(-1), dim=1)
    values = torch.log(1 + torch.relu(values))
    values[:,special_token_ids] = 0
    return values

# transform the sparse vector to a dict of (token, weight)
def transform_sparse_vector_to_dict(sparse_vector):
    sample_indices,token_indices=torch.nonzero(sparse_vector,as_tuple=True)
    non_zero_values = sparse_vector[(sample_indices,token_indices)].tolist()
    number_of_tokens_for_each_sample = torch.bincount(sample_indices).cpu().tolist()
    tokens = [transform_sparse_vector_to_dict.id_to_token[_id] for _id in token_indices.tolist()]

    output = []
    end_idxs = list(itertools.accumulate([0]+number_of_tokens_for_each_sample))
    for i in range(len(end_idxs)-1):
        token_strings = tokens[end_idxs[i]:end_idxs[i+1]]
        weights = non_zero_values[end_idxs[i]:end_idxs[i+1]]
        output.append(dict(zip(token_strings, weights)))
    return output

# download the idf file from model hub. idf is used to give weights for query tokens
def get_tokenizer_idf(tokenizer):
    from huggingface_hub import hf_hub_download
    local_cached_path = hf_hub_down