Alpaca 7B

Alpaca is the instruction-following language model released by Stanford's CRFM on March 13, 2023, created by fine-tuning LLaMA 7B on 52,000 instruction demonstrations generated from OpenAI's text-davinci-003 via the Self-Instruct method. Total cost was under $600 — under $500 for data generation and under $100 for three hours of fine-tuning on eight A100s — proving that GPT-3.5-class instruction following was reproducible on an academic budget and igniting the wave of open instruction-tuned models.

Hard limitations: the dataset and weights are CC BY-NC 4.0, strictly non-commercial and research-only, and Stanford pulled the public demo over hosting costs and safety. Alpaca hallucinates, was never safety-aligned, and is historically important rather than practically useful today; consider it a museum piece that changed the industry.