type
Post
Created date
May 18, 2024 05:15 AM
category
LLM
tags
status
Published
Language
English
From
summary
slug
password
Author
Priority
Featured
Featured
Cover
Origin
Type
URL
Youtube
Youtube
icon
The LLM community has been preaching about local LLM so I tried to test it out - it is indeed magical! With the advent of Phi-3, I am more more convinced that the world is marching towards a scenario where everyone can own their own model. It is not perfect but the fact that a Phi 3 model with 3.8 billion parameters can keep up with Mistral (7 Billions) is indeed jaw-dropping!
ย
I specifically highlighted benchmarks of MMLU, SocialQA, and GSM8K as they are most relatable to me. Both models have an intelligence in a wide range of knowledge ( e.g., different languages and history), math reasoning, and sense of humour.
Benchmark is not everything - such a lightweight model speeds up loading time and inference time. Its user experience is easy to setup, straightforward, and smart. One of the engineering techniques (chain of thought) was especially intuitive.
ย
Kudos to Microsoft for investing in these projects and make AI alive! It is impressive how life has changed since the advent of ChatGPT, reviving the disciplines of GPT technology. Looking forward to an upcoming trend in computer vision too :)
ย
University teached me how to think but the WHY part has always been ambiguous. However, I am grateful to live this era where technology is ever-changing. I am keen on learning more use cases with my peers!
- Author:Jason Siu
- URL:https://jason-siu.com/article%2F9cef8ee9-e0cb-4c9c-a593-eb338d41e57e
- Copyright:All articles in this blog, except for special statements, adopt BY-NC-SA agreement. Please indicate the source!