Guia de início rápido para aplicativos RAG usando LangChain e LlamaIndex

1
db.embedded_movies.deleteMany({ plot: { $exists: false } } )
2
3
{ acknowledged: true, deletedCount: 80 }

1
import os
2
from dotenv import load_dotenv
3
from langchain_community.document_loaders.mongodb import MongodbLoader
4
import nest_asyncio
5
6
nest_asyncio.apply()
7
load_dotenv()
8
loader = MongodbLoader(
9
    connection_string=os.environ['MONGODB_URI'],
10
    db_name=os.environ['MONGODB_DB'],
11
    collection_name=os.environ['MONGODB_COLL'],
12
    filter_criteria={},
13
    field_names=["title", "plot"]
14
)
15
docs = loader.load()
16
17
print(len(docs))
18
docs[0]

1
# MongoDB URI
2
# Replace <username> and <password> with your MongoDB credentials
3
MONGODB_URI='mongodb+srv://<username>:<password>@kushagracluster.abcedf.mongodb.net/'
4
5
# Name of the MongoDB database to use
6
MONGODB_DB='sample_mflix'
7
8
# Name of the MongoDB collection to use within the specified database
9
MONGODB_COLL='embedded_movies'
10
11
# API key for OpenAI.
12
OPENAI_API_KEY="<OPENAI_API_KEY>"
13
14
# Name of the index to use for vector storage
15
MONGODB_VECTOR_INDEX='vector_index'
16
17
# Name of the collection in MongoDB where embedding data is stored
18
MONGODB_VECTOR_COLL_LANGCHAIN='langchain_coll'

1
from pymongo import MongoClient
2
from langchain.vectorstores import MongoDBAtlasVectorSearch
3
from langchain_community.embeddings import OpenAIEmbeddings
4
from langchain.llms import OpenAI
5
from langchain_community.chat_models import ChatOpenAI
6
from langchain.prompts import ChatPromptTemplate
7
from langchain.chains import LLMChain
8
9
client = MongoClient(os.environ['MONGODB_URI'], appname="devrel.content.langchain_llamaIndex.python")
10
collection = client.get_database(os.environ['MONGODB_DB']).get_collection(os.environ['MONGODB_VECTOR_COLL_LANGCHAIN'])
11
12
vector_search = MongoDBAtlasVectorSearch.from_documents(
13
  documents=docs,
14
  embedding=OpenAIEmbeddings(openai_api_key=os.environ['OPENAI_API_KEY']),
15
  collection=collection,
16
  index_name=os.environ['MONGODB_VECTOR_INDEX'])

1
{
2
 "_id": {
3
   "$oid": "66375d1ff77b69f7b5f0c7ec"
4
 },
5
 "text": "The Black Pirate Seeking revenge, an athletic young man joins the pirate band responsible for his father's death....",
6
 "embedding": [
7
   -0.003676240383024406,
8
   -0.02721614428071944,
9
   ...
10
   0.02543453162153851,
11
]
12
],
13
 "database": "sample_mflix",
14
 "collection": "embedded_movies"
15
}

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

1
# perform a similarity search on the ingested documents
2
prompt='What is the best horror movie to watch?'
3
docs_with_score = vector_search.similarity_search_with_score(query=prompt,k=1)
4
5
llm = ChatOpenAI(openai_api_key=os.environ['OPENAI_API_KEY'])
6
7
prompt_template = ChatPromptTemplate.from_messages([
8
    ("system", "You are a movie recommendation engine which posts a concise and short summary on relevant movies."),
9
    ("user", "List of movies: {input}")
10
])
11
12
# Create an LLMChain
13
chain = LLMChain(
14
    llm=llm,
15
    prompt=prompt_template
16
)
17
18
# Prepare the input for the chat model
19
input_docs = "\n".join([doc.page_content for doc, _ in docs_with_score])
20
21
# Invoke the chain with the input documents
22
response = chain.invoke({"input": input_docs})
23
print(response['text'])

1
import os
2
from llama_index.readers.mongodb import SimpleMongoReader
3
4
reader = SimpleMongoReader(uri=os.environ["MONGODB_URI"])
5
documents = reader.load_data(
6
    os.environ["MONGODB_DB"],
7
    os.environ["MONGODB_COLL"], 
8
    field_names=["title","plot"], # these is a list of the top-level fields in your objects that will be indexed
9
    query_dict={} # this is a mongo query dict that will filter your data if you don't want to index everything
10
)
11
print(documents[0])

1
import pymongo
2
import openai
3
from llama_index.core import VectorStoreIndex,StorageContext
4
from llama_index.vector_stores.mongodb import MongoDBAtlasVectorSearch
5
6
# Create a new client and connect to the server
7
client = pymongo.MongoClient(os.environ["MONGODB_URI"], appname="devrel.content.langchain_llamaIndex.python")
8
9
store = MongoDBAtlasVectorSearch(
10
    client,
11
    db_name=os.environ['MONGODB_DB'],
12
    collection_name=os.environ['MONGODB_VECTOR_COLL_LLAMAINDEX'], # this is where your embeddings will be stored
13
    index_name=os.environ['MONGODB_VECTOR_INDEX'] 
14
)
15
16
storage_context = StorageContext.from_defaults(vector_store=store)
17
index = VectorStoreIndex.from_documents(
18
    documents, storage_context=storage_context,
19
    show_progress=True, # this will show you a progress bar as the embeddings are created
20
)

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

1
prompt="What is the best horror movie to watch?"
2
response = index.as_query_engine(similarity_top_k=1).query(prompt)
3
4
print(response)