Stanford Alpaca: Instruction-following LLaMA model training
Research project that fine-tunes LLaMA models to follow instructions using self-generated training data.
Learn more about Stanford Alpaca
Stanford Alpaca is a fine-tuned version of Meta's LLaMA model trained specifically for instruction-following tasks. The model uses a 52K instruction dataset generated through a modified Self-Instruct approach, where GPT-3's text-davinci-003 creates diverse instruction-response pairs. The fine-tuning process employs standard Hugging Face training code with specific hyperparameters optimized for 7B and 13B parameter models. The project is designed for research purposes and includes data generation pipelines, training scripts, and weight recovery tools.
Self-Generated Dataset
Uses a modified Self-Instruct pipeline with text-davinci-003 to generate 52K diverse instruction-following examples at reduced cost (under $500).
Complete Training Pipeline
Provides end-to-end code for data generation, model fine-tuning, and weight recovery with documented hyperparameters for reproducible results.
Research-Focused Implementation
Built specifically for academic research with CC BY NC 4.0 licensing and detailed documentation of methodology and limitations.
See how people are using Stanford Alpaca
Related Repositories
Discover similar tools and frameworks used by developers
Pi Mono
Monorepo providing AI agent development tools, unified LLM API, and deployment management for multiple providers.
Open Lovable
AI tool that generates React applications from website clones or user descriptions.
OpenVINO
Convert and deploy deep learning models across Intel hardware.
Ray
Unified framework for scaling AI and Python applications from laptops to clusters with distributed runtime.
Ollama
Go-based CLI for local LLM inference and management.