Skip to content

Welcome to yasp Compile

yasp Compile is an agentic platform designed to optimize, compile, and deploy PyTorch models at lightning speed. By leveraging intelligent runner agents, yasp bridges the gap between raw research code and production-ready inference artifacts.

🚀 Quick Resources

  • Getting Started Repository Clone the official repository to access deployment scripts, sample models, and local configuration tools.
  • API Reference Comprehensive documentation for the yasp Python client and platform API endpoints.

Core Workflow

yasp Compile simplifies the optimization lifecycle into three distinct phases.

1. Configure the Client

Set up your local Python environment to interact with the platform. Our tools automatically verify strict dependencies—including Python 3.11, PyTorch 2.7.1, and CUDA 12.8—to ensure compatibility with the compilation engine.

2. Deploy the Runner

The Evaluation Runner is a Dockerized agent that handles the heavy lifting of model compilation and benchmarking. It is deployed in your infrastructure (or locally) to keep your model weights secure while communicating securely with the yasp control plane.

  • Setup: Download and load the runner image.
  • Run: Launch the runner agent with GPU access.

3. Optimize & Benchmarking

Submit your standard PyTorch nn.Module code. The platform analyzes, compiles, and benchmarks the model against baselines (like Inductor) to verify correctness and measure speedups.

Basics

  • Get your API Token Learn how to generate and securely save the API token required to interact with the yasp platform.

  • Evaluation Runners Set up the worker nodes that perform on-device measurements. This guide covers downloading and running Docker images for both NVIDIA and AMD hardware.

  • Environment Setup Instructions for advanced users to install the Client SDK and Python dependencies directly on a local machine.

Ready to start?

Check out the Getting Started Repo to spin up a local demo environment and compile your first model in minutes.