Starter Tutorial (Local Models)#
Tip
Make sure you’ve followed the installation steps first.
This is our famous “5 lines of code” starter example with local LLM and embedding models. We will use BAAI/bge-m3 as our embedding model and Mistral-7B served through Ollama as our LLM.
Download data#
This example uses the text of Paul Graham’s essay, “What I Worked On”. This and many other examples can be found in the examples folder of our repo.
The easiest way to get it is to download it via this link and save it in a folder called data.
Setup#
Ollama is a tool to help you get setup with LLMs locally (currently supported on OSX and Linux. You can install Ollama on Windows through WSL 2).
Follow the README to learn how to install it.
To load in a Mistral-7B model just do ollama pull mistral
NOTE: You will need a machine with at least 32GB of RAM.
Load data and build an index#
In the same folder where you created the data folder, create a file called starter.py file with the following:
from llama_index import VectorStoreIndex, SimpleDirectoryReader, ServiceContext
from llama_index.embeddings import resolve_embed_model
from llama_index.llms import Ollama
documents = SimpleDirectoryReader("data").load_data()
# bge-m3 embedding model
embed_model = resolve_embed_model("local:BAAI/bge-small-en-v1.5")
# ollama
llm = Ollama(model="mistral", request_timeout=30.0)
service_context = ServiceContext.from_defaults(
embed_model=embed_model, llm=llm
)
index = VectorStoreIndex.from_documents(
documents, service_context=service_context
)
This builds an index over the documents in the data folder (which in this case just consists of the essay text, but could contain many documents).
Your directory structure should look like this:
├── starter.py └── data └── paul_graham_essay.txt
We use the BAAI/bge-small-en-v1.5 model through resolve_embed_model, which resolves to our HuggingFaceEmbedding class. We also use our Ollama LLM wrapper to load in the mistral model.
Query your data#
Add the following lines to starter.py
query_engine = index.as_query_engine()
response = query_engine.query("What did the author do growing up?")
print(response)
This creates an engine for Q&A over your index and asks a simple question. You should get back a response similar to the following: The author wrote short stories and tried to program on an IBM 1401.
You can view logs, persist/load the index similar to our starter example.