Revolucionando a interação de IA: integrando Mistral AI e MongoDB para um aplicativo LLM GenAI personalizado

1
mistralai                                        				  0.0.8
2
pymongo                                           4.3.3
3
gradio                                            4.10.0
4
gradio_client                                     0.7.3
5
langchain                                         0.0.348
6
langchain-core                                    0.0.12
7
pandas                                            2.0.3

1
import gradio as gr
2
import os
3
import pymongo
4
import pandas as pd
5
from mistralai.client import MistralClient
6
from mistralai.models.chat_completion import ChatMessage
7
from langchain.document_loaders import PyPDFLoader
8
from langchain.text_splitter import RecursiveCharacterTextSplitter

1
def data_prep(file):
2
    # Set up Mistral client
3
    api_key = os.environ["MISTRAL_API_KEY"]
4
    client = MistralClient(api_key=api_key)
5
6
    # Process the uploaded file
7
    loader = PyPDFLoader(file.name)
8
    pages = loader.load_and_split()
9
10
    # Split data
11
    text_splitter = RecursiveCharacterTextSplitter(
12
        chunk_size=100,
13
        chunk_overlap=20,
14
        separators=["\n\n", "\n", "(?<=\. )", " ", ""],
15
        length_function=len,
16
    )
17
    docs = text_splitter.split_documents(pages)
18
19
    # Calculate embeddings and store into MongoDB
20
    text_chunks = [text.page_content for text in docs]
21
    df = pd.DataFrame({'text_chunks': text_chunks})
22
    df['embedding'] = df.text_chunks.apply(lambda x: get_embedding(x, client))
23
24
    collection = connect_mongodb()
25
    df_dict = df.to_dict(orient='records')
26
    collection.insert_many(df_dict)
27
28
    return "PDF processed and data stored in MongoDB."

1
def connect_mongodb():
2
    # Your MongoDB connection string
3
    mongo_url = os.environ["MONGO_URI"]
4
    client = pymongo.MongoClient(mongo_url)
5
    db = client["mistralpdf"]
6
    collection = db["pdfRAG"]
7
    return collection

1
export MONGO_URI="Your_cluster_connection_string"

1
def get_embedding(text, client):
2
    text = text.replace("\n", " ")
3
    embeddings_batch_response = client.embeddings(
4
        model="mistral-embed",
5
        input=text,
6
    )
7
    return embeddings_batch_response.data[0].embedding

1
def qna(users_question):
2
    # Set up Mistral client
3
    api_key = os.environ["MISTRAL_API_KEY"]
4
    client = MistralClient(api_key=api_key)
5
6
    question_embedding = get_embedding(users_question, client)
7
    print("-----Here is user question------")
8
    print(users_question)
9
    documents = find_similar_documents(question_embedding)
10
    
11
    print("-----Retrieved documents------")
12
    print(documents)
13
    for doc in documents:
14
        doc['text_chunks'] = doc['text_chunks'].replace('\n', ' ')
15
    
16
    for document in documents:
17
        print(str(document) + "\n")
18
19
    context = " ".join([doc["text_chunks"] for doc in documents])
20
    template = f"""
21
    You are an expert who loves to help people! Given the following context sections, answer the
22
    question using only the given context. If you are unsure and the answer is not
23
    explicitly written in the documentation, say "Sorry, I don't know how to help with that."
24
25
    Context sections:
26
    {context}
27
28
    Question:
29
    {users_question}
30
31
    Answer:
32
    """
33
    messages = [ChatMessage(role="user", content=template)]
34
    chat_response = client.chat(
35
        model="mistral-tiny",
36
        messages=messages,
37
    )
38
    formatted_documents = '\n'.join([doc['text_chunks'] for doc in documents])
39
40
    return chat_response.choices[0].message, formatted_documents

1
{
2
 "type": "vectorSearch",
3
 "fields": [
4
   {
5
     "numDimensions": 1536,
6
     "path": "'embedding'",
7
     "similarity": "cosine",
8
     "type": "vector"
9
   }
10
 ]
11
}

1
def find_similar_documents(embedding):
2
    collection = connect_mongodb()
3
    documents = list(
4
        collection.aggregate([
5
            {
6
                "$vectorSearch": {
7
                    "index": "vector_index",
8
                    "path": "embedding",
9
                    "queryVector": embedding,
10
                    "numCandidates": 20,
11
                    "limit": 10
12
                }
13
            },
14
            {"$project": {"_id": 0, "text_chunks": 1}}
15
        ]))
16
    return documents

1
# Gradio Interface for PDF Upload
2
pdf_upload_interface = gr.Interface(
3
    fn=data_prep,
4
    inputs=gr.File(label="Upload PDF"),
5
    outputs="text",
6
    allow_flagging="never"
7
)
8
9
# Gradio Interface for Chatbot
10
chatbot_interface = gr.Interface(
11
    fn=qna,
12
    inputs=gr.Textbox(label="Enter Your Question"),
13
    outputs=[
14
        gr.Textbox(label="Mistral Answer"),
15
        gr.Textbox(label="Retrieved Documents from MongoDB Atlas")
16
    ],
17
    allow_flagging="never"
18
)
19
20
# Combine interfaces into tabs
21
iface = gr.TabbedInterface(
22
    [pdf_upload_interface, chatbot_interface],
23
    ["Upload PDF", "Chatbot"]
24
)
25
26
# Launch the Gradio app
27
iface.launch()