Bartholomew Gallacher
Well-known member
- Joined
- Sep 26, 2018
- Messages
- 6,991
- SL Rez
- 2002
Let's talk about Llama and Alpaca for a moment. First about LLama; this is Facebook's attempt on a chat bot like ChatGPT.
ai.facebook.com
The code for this was shared, however the model and trained data not. But soon enough the training data was leaked mid of march this year.
The AI guys at Stanford then took the model, and fine tuned it, naming it Alpaca - meaning removed a lot of parameters.
As a result Alpaca is runnable on normal consumer grade hardware, even a Raspberry Pi will do, although a little bit slowly. It's not covering as much stuff as ChatGPT, but in certain areas this effort is on par with ChatGPT. The video shows Alpaca running on a Raspberry Pi in real time in the end at 4:20, and for such low end hardware it's running decently fast.
According to the scientists Llama is comparable with text-davinci-003, which was a stepping stone from ChatGPT 3 to 3.5. The fine tuning of the model did cost around 600 US$. The data model is around 4 GB.
Introducing LLaMA: A foundational, 65-billion-parameter language model
Today, we’re releasing our LLaMA (Large Language Model Meta AI) foundational model with a gated release. LLaMA is more efficient and competitive with previously published models of a similar size on existing benchmarks.
ai.facebook.com
The code for this was shared, however the model and trained data not. But soon enough the training data was leaked mid of march this year.
The AI guys at Stanford then took the model, and fine tuned it, naming it Alpaca - meaning removed a lot of parameters.
As a result Alpaca is runnable on normal consumer grade hardware, even a Raspberry Pi will do, although a little bit slowly. It's not covering as much stuff as ChatGPT, but in certain areas this effort is on par with ChatGPT. The video shows Alpaca running on a Raspberry Pi in real time in the end at 4:20, and for such low end hardware it's running decently fast.
According to the scientists Llama is comparable with text-davinci-003, which was a stepping stone from ChatGPT 3 to 3.5. The fine tuning of the model did cost around 600 US$. The data model is around 4 GB.
Last edited:





















