To install the zerolink package, you can use the following command:

pip install 'zerolink @ git+https://github.com/zerolink-io/zerolink-runner.git'

In addition you will need one of the following dependencies to run the zerolink model inference:

• MLX ( Apple Silicon )
• VLLM ( Nvidia GPUs )
• llama.cpp ( CPU )

To use MLX with your Apple Silicon Mac, you can install the mlx dependency group with one of the following command:

pip install 'zerolink[mlx] @ git+https://github.com/zerolink-io/zerolink-runner.git'
poetry install --with mlx

You will need a system with at least 20GB of memory and 15GB of free disk space to install the model locally.

You can use the to_sql function to convert a natural language question into SQL. Optionally pass the context parameter to provide schema information.

from zerolink import to_sql

sql = to_sql("What are the top selling products?", framework="mlx")
print(sql)

To use VLLM on a server with attached Nvidia GPUs, you can install the vllm dependency group with one of the following command:

pip install 'zerolink[vllm] @ git+https://github.com/zerolink-io/zerolink-runner.git'
poetry install --with vllm

Supported GPUs:

• Nvidia T4
• Nvidia A40
• Nvidia A100 (40GB / 80GB)
• Nvidia H100

To run the model inference on the GPU, you can use the following code:

from zerolink import to_sql

to_sql("What are the top selling products?", framework="vllm")

To use llama.cpp on stock Intel or AMD CPUs, you can install the llamacpp dependency group with one of the following command:

pip install 'zerolink[llamacpp] @ git+https://github.com/zerolink-io/zerolink-runner.git'
poetry install --with llamacpp

To run the model inference on a CPU, you can use the following code:

from zerolink import to_sql

to_sql("What are the top selling products?", framework="llamacpp")

This project is licensed under the Apache License - see LICENSE.md file for details