Launch Open-Source Apps with LangChain

1
%%writefile requirements.txt
2
langchain==0.0.339
3
openai==1.3.3
4
pdf2image==1.17.0
5
pdfminer==20191125
6
pdfminer.six==20221105
7
pillow_heif==0.13.1
8
tabulate==0.9.0
9
tiktoken==0.5.1
10
unstructured==0.11.0
11
opencv-contrib-python-headless==4.8.1.78
12
unstructured.pytesseract==0.3.12
13
unstructured.inference==0.7.15

1
%conda install -y --quiet poppler tesseract

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

1
import nltk
2
nltk.download('punkt_tab')
3
nltk.download('averaged_perceptron_tagger_eng')

1
from langchain.document_loaders import OnlinePDFLoader
2

3
loader = OnlinePDFLoader("http://leavcom.com/pdf/DBpdf.pdf")
4

5
data = loader.load()

1
from langchain.text_splitter import RecursiveCharacterTextSplitter
2

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

1
text_splitter = RecursiveCharacterTextSplitter(chunk_size=2000, chunk_overlap=0)
2
texts = text_splitter.split_documents(data)
3

4
print (f"You have {len(texts)} pages")

1
%%sql
2
DROP DATABASE IF EXISTS pdf_db;
3
CREATE DATABASE IF NOT EXISTS pdf_db;

1
%%sql
2
DROP TABLE IF EXISTS pdf_docs1;
3
CREATE TABLE IF NOT EXISTS pdf_docs1 (
4
    id INT PRIMARY KEY,
5
    content TEXT CHARACTER SET utf8mb4 COLLATE utf8mb4_general_ci,
6
    vector BLOB
7
);

1
import os
2
import getpass
3

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

1
import json
2
import sqlalchemy as sa
3
from langchain.embeddings import OpenAIEmbeddings
4
from singlestoredb import create_engine
5

6
conn = create_engine().connect()
7

8
embedder = OpenAIEmbeddings()
9

10
# Fetch all embeddings in one call
11
embeddings = embedder.embed_documents([doc.page_content for doc in texts])
12

13
# Build query parameters
14
params = []
15
for i, (text_content, embedding) in enumerate(zip(texts, embeddings)):
16
    params.append(dict(id=i+1, content=text_content, vector=json.dumps(embedding)))
17

18
stmt = sa.text("""
19
    INSERT INTO pdf_docs1 (
20
        id,
21
        content,
22
        vector
23
    )
24
    VALUES (
25
        :id,
26
        :content,
27
        JSON_ARRAY_PACK_F32(:vector)
28
    )
29
""")
30

31
conn.execute(stmt, params)

1
%%sql
2
SELECT JSON_ARRAY_UNPACK_F32(vector) as vector
3
FROM pdf_docs1
4
LIMIT 1;

1
query_text = "Will object-oriented databases be commercially successful?"
2

3
query_embedding = embedder.embed_documents([query_text])[0]
4

5
stmt = sa.text("""
6
    SELECT
7
        content,
8
        DOT_PRODUCT_F32(JSON_ARRAY_PACK_F32(:embedding), vector) AS score
9
    FROM pdf_docs1
10
    ORDER BY score DESC
11
    LIMIT 1
12
""")
13

14
results = conn.execute(stmt, dict(embedding=json.dumps(query_embedding)))
15

16
for row in results:
17
    print(row[0])

1
import openai
2

3
client = openai.OpenAI()
4

5
prompt = f"The user asked: {query_text}. The most similar text from the document is: {row[0]}"
6

7
response = 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

15
print(response.choices[0].message.content)

1
%%sql
2
DROP DATABASE IF EXISTS pdf_db