斯坦福微调了 7B LLaMA 模型[1]，只用了 52K 的数据，达到了和 达芬奇003[2] 类似的效果，并且可以跑在消费级设备上，比如树莓派[3]。
https://github.com/tatsu-lab/stanford_alpaca
[1]: 模型大小 12.6 GiB
[2]: https://platform.openai.com/docs/models/gpt-3-5
[3]: https://en.wikipedia.org/wiki/Raspberry_Pi#RAM
https://github.com/tatsu-lab/stanford_alpaca
[1]: 模型大小 12.6 GiB
[2]: https://platform.openai.com/docs/models/gpt-3-5
[3]: https://en.wikipedia.org/wiki/Raspberry_Pi#RAM
GitHub
GitHub - tatsu-lab/stanford_alpaca: Code and documentation to train Stanford's Alpaca models, and generate the data.
Code and documentation to train Stanford's Alpaca models, and generate the data. - tatsu-lab/stanford_alpaca
