README.md

metadata

{}

Dataset Card for hlm-paraphrase-multilingual-mpnet-base-v2

Dataset Summary

Chromadb vectorstore for 红楼梦, created with

import os
from langchain.document_loaders import TextLoader
from langchain.embeddings import SentenceTransformerEmbeddings
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import Chroma

model_name = 'paraphrase-multilingual-mpnet-base-v2'
embedding = SentenceTransformerEmbeddings(model_name=model_name)

url = 'https://raw.githubusercontent.com/ffreemt/multilingual-dokugpt/master/docs/hlm.txt'
os.system(f'wget -c {url}')
doc = TextLoader('hlm.txt').load()

text_splitter = RecursiveCharacterTextSplitter(
    separators=["\n\n", "\n", ".", "!", "?", ",", " ", ""],
    chunk_size=620,
    chunk_overlap=60,
    length_function=len
)

doc_chunks = text_splitter.split_documents(doc)
client_settings = Settings(chroma_db_impl="duckdb+parquet", anonymized_telemetry=False, persist_directory='db')

# takes 8-20 minutes on CPU
vectorstore = Chroma.from_documents(
    documents=doc_chunks,
    embedding=embedding,
    persist_directory='db',
    client_settings=client_settings,
)
vectorstore.persist()

How to use

Download the hlm directory to a local directory, e.g., db, for example

from huggingface_hub import snapshot_download

snapshot_download(
  repo_id="mikeee/chroma-paraphrase-multilingual-mpnet-base-v2",
  repo_type="dataset",
  allow_patterns="hlm/*",
  local_dir="db",
  resume_download=True,
)

Load the vectorestore:

from langchain.embeddings import SentenceTransformerEmbeddings
from langchain.vectorstores import Chroma
from chromadb.config import Settings

model_name = 'paraphrase-multilingual-mpnet-base-v2'
embedding = SentenceTransformerEmbeddings(model_name=model_name)

client_settings = Settings(
  chroma_db_impl="duckdb+parquet",
  anonymized_telemetry=False,
  persist_directory='db/hlm'
)

db = Chroma(
    # persist_directory='docs',
    embedding_function=embedding,
    client_settings=client_settings,
)

res = db.search("红楼梦主线", search_type="similarity", k=2)
print(res)
# [Document(page_content='通灵宝玉正面图式\u3000通灵宝玉反面图式\n\n\n\n玉宝灵通\u3000\u3000\u3000\u3000\u3000三二一\n\n仙莫\u3000\u3000\u3000\u3000\u3000\u3000知疗除\n\n寿失\u3000\u3000\u3000\u3000\u3000\u3000祸冤邪\n\n恒莫\u3000\u3000\u3000\u3000\u3000\u3000福疾崇\n\n昌忘\n\n\n\n宝钗看毕，【甲戌双行。。。