Stanford · 4× · naposledy 06. 8. 2026
7
Momentum
Alpaca is an instruction-optimized language model developed by Stanford researchers (CRFM) that builds on Meta's LLaMA-7B base model and was Fine-Tuned with 52,000 automatically generated instruction examples. It served as a research demonstration that a small model can achieve behavior similar to OpenAI's text-davinci-003 at very low cost (under $600). Alpaca is licensed exclusively for non-commercial research purposes; the public demo was shut down shortly after release.
Vývoj momenta
15.06.13.09.
Vlastnosti
| Key Benchmark (%) | Human blind evaluation: won 90 of 179 comparisons vs. text-davinci-003 (~50%) |
| License | Research use only; data/code CC BY-NC 4.0, weights under LLaMA license agreement |
| Multimodality | No, text-only language model |
| Platform | Fine-tuned from Meta LLaMA 7B, code/weights available via GitHub (weight diff) |
| Release Date | March 13, 2023 |