Do you need 7B+ parameters to get great performance from your Language Models? Discover how Microsoft Research's Phi-2, a 2.7 billion-parameter language model, challenges this norm by outperforming models up to 25x its size (according to Microsoft Research). We'll delve into the training methods behind Phi-2, from 'textbook-quality' training data to scaled knowledge transfer techniques. We'll load the model into a Google Colab and try it out in coding, math, reasoning, and data extraction.
Blog Post: [ Ссылка ]
Phi-2 on HF Hub: [ Ссылка ]
AlpacaEval: [ Ссылка ]
AI Bootcamp (preview drops on Christmas): [ Ссылка ]
Discord: [ Ссылка ]
Subscribe: [ Ссылка ]
GitHub repository: [ Ссылка ]
Join this channel to get access to the perks and support my work:
[ Ссылка ]
00:00 - Intro
00:32 - AI Bootcamp on MLExpert.io
01:25 - What is Phi-2?
07:25 - Phi-2 vs Mistral vs Llama 2
08:42 - Phi-2 on HuggingFace Hub
10:45 - Google Colab Setup
13:35 - Prompt Format
14:25 - Text Generation
20:17 - Math
21:15 - Coding
24:30 - Text Analysis
26:46 - Conclusion
#artificialintelligence #chatgpt #gpt4 #python #chatbot #llama2 #llm
![](https://i.ytimg.com/vi/ZustIyKzbAU/maxresdefault.jpg)