Launch Open-Source Apps with LangChain

1%%writefile requirements.txt2langchain==0.0.3393openai==1.3.34pdf2image==1.17.05pdfminer==201911256pdfminer.six==202211057pillow_heif==0.13.18tabulate==0.9.09tiktoken==0.5.110unstructured==0.11.011opencv-contrib-python-headless==4.8.1.7812unstructured.pytesseract==0.3.1213unstructured.inference==0.7.15

1%conda install -y --quiet poppler tesseract

1%pip install -r requirements.txt --quiet

1import nltk2nltk.download('punkt_tab')3nltk.download('averaged_perceptron_tagger_eng')

1from langchain.document_loaders import OnlinePDFLoader2
3loader = OnlinePDFLoader("http://leavcom.com/pdf/DBpdf.pdf")4
5data = loader.load()

1from langchain.text_splitter import RecursiveCharacterTextSplitter2
3print (f"You have {len(data)} document(s) in your data")4print (f"There are {len(data[0].page_content)} characters in your document")

1text_splitter = RecursiveCharacterTextSplitter(chunk_size=2000, chunk_overlap=0)2texts = text_splitter.split_documents(data)3
4print (f"You have {len(texts)} pages")

1%%sql2DROP DATABASE IF EXISTS pdf_db;3CREATE DATABASE IF NOT EXISTS pdf_db;

1%%sql2DROP TABLE IF EXISTS pdf_docs1;3CREATE TABLE IF NOT EXISTS pdf_docs1 (4    id INT PRIMARY KEY,5    content TEXT CHARACTER SET utf8mb4 COLLATE utf8mb4_general_ci,6    vector BLOB7);

1import os2import getpass3
4os.environ["OPENAI_API_KEY"] = getpass.getpass("OpenAI API Key:")

1import json2import sqlalchemy as sa3from langchain.embeddings import OpenAIEmbeddings4from singlestoredb import create_engine5
6conn = create_engine().connect()7
8embedder = OpenAIEmbeddings()9
10# Fetch all embeddings in one call11embeddings = embedder.embed_documents([doc.page_content for doc in texts])12
13# Build query parameters14params = []15for i, (text_content, embedding) in enumerate(zip(texts, embeddings)):16    params.append(dict(id=i+1, content=text_content, vector=json.dumps(embedding)))17
18stmt = sa.text("""19    INSERT INTO pdf_docs1 (20        id,21        content,22        vector23    )24    VALUES (25        :id,26        :content,27        JSON_ARRAY_PACK_F32(:vector)28    )29""")30
31conn.execute(stmt, params)

1%%sql2SELECT JSON_ARRAY_UNPACK_F32(vector) as vector3FROM pdf_docs14LIMIT 1;

1query_text = "Will object-oriented databases be commercially successful?"2
3query_embedding = embedder.embed_documents([query_text])[0]4
5stmt = sa.text("""6    SELECT7        content,8        DOT_PRODUCT_F32(JSON_ARRAY_PACK_F32(:embedding), vector) AS score9    FROM pdf_docs110    ORDER BY score DESC11    LIMIT 112""")13
14results = conn.execute(stmt, dict(embedding=json.dumps(query_embedding)))15
16for row in results:17    print(row[0])

1import openai2
3client = openai.OpenAI()4
5prompt = f"The user asked: {query_text}. The most similar text from the document is: {row[0]}"6
7response = client.chat.completions.create(8    model="gpt-3.5-turbo",9    messages=[10        {"role": "system", "content": "You are a helpful assistant."},11        {"role": "user", "content": prompt}12    ]13)14
15print(response.choices[0].message.content)

1%%sql2DROP DATABASE IF EXISTS pdf_db