LLaMA 3
LLaMA 3 (short for Large Language Model Meta AI) is a family of large language models created by Meta (Facebook’s parent company).
It’s the third generation after LLaMA 1 (2023) and LLaMA 2 (2023), and was released in April 2024.
It’s open-weight which means the model parameters are freely downloadable and usable under Meta’s license for research and commercial work.
It is a powerful GPT-3.5-class model that we can download, run locally, or fine-tune for our own chatbot, assistant, or research.
_________________________________________
Features:
* We can download and run it locally
* There are 2 versions:
- 8 B (fast, laptop-friendly)
- 70 B (large, powerful).
* Versions fine-tuned to follow human-style instructions (“LLaMA 3-Instruct”).
* Strong performance in English and major European languages.
* Uses advanced attention and tokenizer improvements for faster inference.
* Performs near ChatGPT-3.5 level in reasoning, coding, and text generation.
_________________________________________
Like all large language models, LLaMA 3 is trained on a massive text dataset (web data, books, code, etc.)
It learns to predict the next word given previous words over billions of examples.
Once trained, it can:
* Write essays or code
* Summarize or translate text
* Summarize or translate text
No API key, no internet once downloaded.
It runs on Linux.
Enjoy #linux
Well, that was exciting. See you in the next one!