alpamayo1-sft

What problem does it solve?

This Skill automates the end-to-end supervised fine-tuning process for the Alpamayo-1 Vision-Learning Model (VLM) on the Physical AI Autonomous Vehicles (PAI) dataset. It removes the need for manual fine-tuning steps and speeds up the process.

Core Features & Use Cases

• End-to-End Fine-Tuning: Automates the entire fine-tuning process, including Stage 1 and Stage 2, using Hugging Face, Hydra, and DeepSpeed.
• Data Preparation: Provides commands to download and prepare the PAI dataset, including options for selecting a subset or enabling CoC reasoning labels.
• Model Training: Handles the installation and configuration of the necessary environment, including the uv virtual environment and the required libraries.
• Evaluation: Includes commands for evaluating the trained model on the validation dataset.
• Use Case: Ideal for researchers or developers looking to fine-tune the Alpamayo-1 VLM on the PAI dataset without manually setting up the environment or training pipeline.

Quick Start

Use the alpamayo1-sft skill to fine-tune the Alpamayo-1 VLM on the PAI dataset using the default configuration.