File size: 5,177 Bytes
22d91a3
 
 
 
 
 
0bb2d4f
22d91a3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b21e4ba
22d91a3
 
 
 
 
 
 
 
 
302fc29
22d91a3
302fc29
22d91a3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b21e4ba
 
 
22d91a3
 
 
 
191d00b
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
156
157
158
159
160
161
162
import logging
import os
from concurrent.futures import ProcessPoolExecutor, ThreadPoolExecutor, as_completed

import click
import torch

from langchain.docstore.document import Document
from langchain.embeddings import HuggingFaceInstructEmbeddings
from langchain.text_splitter import Language, RecursiveCharacterTextSplitter
from langchain.vectorstores import Chroma

from constants import (
    CHROMA_SETTINGS,
    DOCUMENT_MAP,
    EMBEDDING_MODEL_NAME,
    INGEST_THREADS,
    PERSIST_DIRECTORY,
    SOURCE_DIRECTORY,
)

def load_single_document(file_path: str) -> Document:
    # Loads a single document from a file path
    file_extension = os.path.splitext(file_path)[1]
    loader_class = DOCUMENT_MAP.get(file_extension)
    if loader_class:
        loader = loader_class(file_path)
    else:
        raise ValueError("Document type is undefined")
    return loader.load()[0]


def load_document_batch(filepaths):
    logging.info("Loading document batch")
    # create a thread pool
    with ThreadPoolExecutor(len(filepaths)) as exe:
        # load files
        futures = [exe.submit(load_single_document, name) for name in filepaths]
        # collect data
        data_list = [future.result() for future in futures]
        # return data and file paths
        return (data_list, filepaths)


def load_documents(source_dir: str) -> list[Document]:
    # Loads all documents from the source documents directory, including nested folders
    paths = []
    for root, _, files in os.walk(source_dir):
        for file_name in files:
            file_extension = os.path.splitext(file_name)[1]
            source_file_path = os.path.join(root, file_name)
            if file_extension in DOCUMENT_MAP.keys():
                paths.append(source_file_path)

    # Have at least one worker and at most INGEST_THREADS workers
    n_workers = min(INGEST_THREADS, max(len(paths), 1))
    chunksize = round(len(paths) / n_workers)
    docs = []
    with ProcessPoolExecutor(n_workers) as executor:
        futures = []
        # split the load operations into chunks
        for i in range(0, len(paths), chunksize):
            # select a chunk of filenames
            filepaths = paths[i : (i + chunksize)]
            # submit the task
            future = executor.submit(load_document_batch, filepaths)
            futures.append(future)
        # process all results
        for future in as_completed(futures):
            # open the file and load the data
            contents, _ = future.result()
            docs.extend(contents)

    return docs


def split_documents(documents: list[Document]) -> tuple[list[Document], list[Document]]:
    # Splits documents for correct Text Splitter
    text_docs, python_docs = [], []
    for doc in documents:
        file_extension = os.path.splitext(doc.metadata["source"])[1]
        if file_extension == ".py":
            python_docs.append(doc)
        else:
            text_docs.append(doc)

    return text_docs, python_docs


@click.command()
@click.option(
    "--device_type",
    default="cuda" if torch.cuda.is_available() else "cpu",
    type=click.Choice(
        [
            "cpu",
            "cuda",
            "ipu",
            "xpu",
            "mkldnn",
            "opengl",
            "opencl",
            "ideep",
            "hip",
            "ve",
            "fpga",
            "ort",
            "xla",
            "lazy",
            "vulkan",
            "mps",
            "meta",
            "hpu",
            "mtia",
        ],
    ),
    help="Device to run on. (Default is cuda)",
)
def main(device_type):
    # Load documents and split in chunks
    logging.info(f"Loading documents from {SOURCE_DIRECTORY}")
    documents = load_documents(SOURCE_DIRECTORY)
    text_documents, python_documents = split_documents(documents)
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
    python_splitter = RecursiveCharacterTextSplitter.from_language(
        language=Language.PYTHON, chunk_size=1000, chunk_overlap=200
    )
    texts = text_splitter.split_documents(text_documents)
    texts.extend(python_splitter.split_documents(python_documents))
    logging.info(f"Loaded {len(documents)} documents from {SOURCE_DIRECTORY}")
    logging.info(f"Split into {len(texts)} chunks of text")

    # Create embeddings
    embeddings = HuggingFaceInstructEmbeddings(
        model_name=EMBEDDING_MODEL_NAME,
        model_kwargs={"device": "cuda"},
    )

    # change the embedding type here if you are running into issues.
    # These are much smaller embeddings and will work for most appications
    # If you use HuggingFaceEmbeddings, make sure to also use the same in the
    # run_localGPT.py file.

    # embeddings = HuggingFaceEmbeddings(model_name=EMBEDDING_MODEL_NAME)

    db = Chroma.from_documents(
        texts,
        embeddings,
        persist_directory=PERSIST_DIRECTORY,
        client_settings=CHROMA_SETTINGS,

    )

    db.persist()
    db = None

if __name__ == "__main__":
    logging.basicConfig(
        format="%(asctime)s - %(levelname)s - %(filename)s:%(lineno)s - %(message)s", level=logging.INFO
    )
    main()