Meta just released the next iteration of their open-access Llama language models. The Llama 3 AI model is now publicly available and provides state-of-the-art performance among LLMs.
We don’t have a paper for Llama 3 yet. Still, there is a lot of information scattered across the web, so this video summarised the most important details about the Llama 3 models we currently know. I did my best to present all the information in a structured way and compare everything with the previous version (Llama 2) and other state-of-the-art LLMs.
⭐ SUPPORT ⭐──────────────────
- Subscribe!
- ☕️ Coffee - www.buymeacoffee.com/botsknow...
🎥 CHAPTERS──────────────────
00:00 - Intro
00:15 - Model Sizes
01:14 - Model Variants
02:12 - Maze Generation
02:07 - Pretraining Data
03:40 - Finetuning Data
04:27 - DPO
04:48 - Tokenizer
06:01 - Context Window
06:33 - Benchmarks
📄 PAPERS──────────────────
👉 LIMA: tinyurl.com/lima-paper
👉 DPO: tinyurl.com/dpo-paper
👉 Llama 2: Open Foundation and Fine-Tuned Chat Models - arxiv.org/pdf/2307.09288.pdf
👉 Llama Guard: tinyurl.com/llama-guard-paper
🔗 USEFUL LINKS ──────────────────
👉 Llama 3 Model Card: github.com/meta-llama/llama3/...
👉 Meta’s blog post: ai.meta.com/blog/meta-llama-3/
👉 HuggingFace article: huggingface.co/blog/llama3
👉 Llama3 vs Llama2 benchmarks: tinyurl.com/reddit-benchmarks
👉 Chatbot Arena: chat.lmsys.org/
👉 Tiktoken Tokenizer: github.com/openai/tiktoken
#llama3 #llama #ai #meta
Негізгі бет Ғылым және технология LLAMA 3 : Explained and Summarised Under 8 Minutes (Compared to Llama 2, Meta AI)
Пікірлер: 10