File size: 4,390 Bytes
1536dad
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
# ## Files used:
# conf.files.context
# conf.files.index
# conf.files.embeddings

import logging
import csv

import hydra
from omegaconf import DictConfig, OmegaConf
import faiss
from datasets import load_dataset, IterableDataset
from tqdm.auto import tqdm

from langchain_community.docstore.in_memory import InMemoryDocstore
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain_core.documents import Document

log = logging.getLogger(__name__)


def get_hnsw_pq_index(conf: DictConfig) -> faiss.IndexHNSWPQ:
    index = faiss.index_factory(
        conf.embeddings.dim,
        f"HNSW{conf.indexes.hnsw_pq.hnsw_m},PQ{conf.indexes.hnsw_pq.pq_m}x{conf.indexes.hnsw_pq.bits}",
    )
    index.hnsw.efConstruction = conf.indexes.hnsw_pq.ef_construction
    index.hnsw.efSearch = conf.indexes.hnsw_pq.ef_search
    return index


def get_hnsw_index(conf: DictConfig) -> faiss.IndexHNSW:
    index = faiss.index_factory(
        conf.embeddings.dim,
        f"HNSW{conf.indexes.hnsw_pq.hnsw_m}",
    )
    index.hnsw.efConstruction = conf.indexes.hnsw_pq.ef_construction
    index.hnsw.efSearch = conf.indexes.hnsw_pq.ef_search
    return index


def get_index(conf: DictConfig) -> faiss.Index:
    if conf.indexing.index_type == "hnsw_pq":
        return get_hnsw_pq_index(conf)
    elif conf.indexing.index_type == "hnsw":
        return get_hnsw_index(conf)
    else:
        raise ValueError(f"Unknown index type: {conf.indexing.index_type}")


def get_embedding_model(conf: DictConfig) -> HuggingFaceEmbeddings:
    return HuggingFaceEmbeddings(
        model_name=conf.embeddings.model,
        model_kwargs={"trust_remote_code": True, "device": conf.embeddings.device},
    )


def get_faiss_store(conf: DictConfig, embedding_model: HuggingFaceEmbeddings) -> FAISS:
    index = get_index(conf)
    return FAISS(
        embedding_function=embedding_model,
        index=index,
        docstore=InMemoryDocstore(),
        index_to_docstore_id={},
    )


def get_dataset(
    conf,
) -> IterableDataset:

    with open(conf.files.context, "r") as f:
        num_lines = sum(1 for _ in f)

    dataset = load_dataset(
        "json",
        data_files={"train": [conf.files.context]},
        streaming=True,
        split="train",
    )

    batched_dataset = dataset.batch(batch_size=conf.indexing.batch_size)
    return batched_dataset, num_lines


def add_batch(
    faiss_store: FAISS,
    embeddings: list[list[float]],
    batch: dict,
) -> None:

    min_size = min(len(batch["text_content"]), len(embeddings))
    if min_size == 0:
        return

    text_and_embeddings = zip(batch["text_content"][:min_size], embeddings[:min_size])
    metadatas = [
        {
            "source_name": source_name,
            "associated_dates": associated_date,
            "chunk_id": chunk_id,
        }
        for source_name, associated_date, chunk_id in zip(
            batch["source_name"][:min_size],
            batch["associated_dates"][:min_size],
            batch["chunk_id"][:min_size],
        )
    ]

    faiss_store.add_embeddings(
        text_embeddings=text_and_embeddings, metadatas=metadatas, ids=batch["chunk_id"]
    )


def batched_read_embeddings(
    conf: DictConfig, csv_reader: csv.reader
) -> list[list[float]]:
    embeddings = []
    for _ in range(conf.indexing.batch_size):
        try:
            row = next(csv_reader)
            embeddings.append([float(x) for x in row])
        except StopIteration:
            break
    return embeddings


def save(conf: DictConfig, faiss_store: FAISS) -> None:
    faiss_store.save_local(conf.files.index)


@hydra.main(version_base=None, config_path="conf", config_name="config")
def main(conf: DictConfig) -> None:
    log.info("Config:\n%s", OmegaConf.to_yaml(conf))

    embedding_model = get_embedding_model(conf)
    faiss_store = get_faiss_store(conf, embedding_model)
    dataset, num_lines = get_dataset(conf)

    with open(conf.files.embeddings, "r", newline="") as f:
        csv_reader = csv.reader(f, delimiter="\t")

        for i, batch in tqdm(
            enumerate(dataset), total=num_lines // conf.indexing.batch_size
        ):
            embeddings = batched_read_embeddings(conf, csv_reader)
            add_batch(faiss_store, embeddings, batch)

    save(conf, faiss_store)


if __name__ == "__main__":
    main()