Skip to content

Quickstart

Five working examples to get you productive in under five minutes.

1. Generate an Image (No GPU Needed)

If you have a Kaggle account (free), you can offload image generation to their GPU:

# One-time setup: connect your Kaggle account
vllama login --service kaggle --username YOUR_USERNAME --key YOUR_API_KEY

# Generate an image — runs on Kaggle's free T4 GPU, downloads to your machine
vllama run stabilityai/sd-turbo --service kaggle --prompt "A neon-lit city street at night"

The image is saved as vllama_kaggle_<timestamp>.png in your current directory.

Where to find your Kaggle API key

Go to kaggle.com → Account → API → Create New Token. This downloads kaggle.json. Your username and key are in that file.

2. AutoML: Train Models on Your CSV

Two commands to go from a raw CSV to a trained model leaderboard:

# Step 1: Preprocess your dataset
vllama data --path my_data.csv --target target_column

# Step 2: Train all models on the preprocessed data
# (replace the folder name with the one created in Step 1)
vllama train --path ./output_folder_20240101_120000 --target target_column

After vllama train, open results/report.html in your browser for a full interactive leaderboard.

Don't have a dataset handy? Grab the Titanic dataset:

# Download the Titanic dataset
curl -O https://raw.githubusercontent.com/datasciencedojo/datasets/master/titanic.csv

vllama data --path titanic.csv --target Survived
vllama train --path ./output_folder_* --target Survived

3. Run a Local LLM and Chat With It

# Terminal 1: Start a local LLM server (downloads model on first run)
vllama run_llm Qwen/Qwen2.5-Coder-0.5B-Instruct

# Terminal 2: Chat with it
vllama chat_llm

You> Write a Python function to reverse a linked list
Assistant> Here's a function...

The model runs entirely on your machine. No internet after the initial download, no API keys, no usage costs.

4. Detect Objects in an Image

# From a local file
vllama detect_image --path photo.jpg

# From a URL
vllama detect_image --url https://ultralytics.com/images/bus.jpg

Results are saved to the current directory with bounding boxes drawn on the image.

5. Text-to-Speech

# Speak text directly
vllama tts --text "Hello from Vllama"

# Interactive mode — enter multiple lines
vllama tts

What's Next?