COMPUTERS

You Can Now Play With Source Code From AlexNet, The Model That Started The LLM Craze

A Look At The Code That Led Us To Where We Are Now

Google and the Computer History Museum have released the source code of AlexNet, the first convolutional neural network to identify objects in pictures with a level accuracy rate high enough to be useful.  AlexNet was developed back in 2012 on NVIDIA GPUs using CUDA, and is generally considered to be the first example of the LLMs that are now all around us.  It was proof that deep learning was a far better way to train AI models than the processes which were currently being developed.  Those models were more traditional but abysmal when it came to accurately identifying the contents of images.

AlexNet, developed out of the University of Toronto, didn’t succeed because it introduced brand new technology but rather because it incorporated existing AI techniques in novel ways.   The traditional way to train an AI involved writing huge volumes of code to try to define certain objects so the model could identify them in a picture.  AlexNet did something completely different, which anyone who follows LLMs will immediately recognize.  It used deep neural networks, massive image datasets, and NVIDIA GPUs to teach itself to recognize “patterns at different levels of abstraction—from simple edges and textures in early layers to complex object parts in deeper layers.“

While AlexNet was the first of the modern LLM architectures, it has been superseded the Transformer model based LLMs that have been flooding the market lately.  If you are curious about how we got where we are now check out the GitHub page and put your CUDA skills to work.


Source link

Related Articles

Back to top button