For this page ml will be used as an acronym for machine learning
Below is an example of an LLM called ollama which has been imported into the website
This is an example of a ml algorithm and there are many different types of ml algorithms
The example is the type LLM(Large Language Model) and is also the model that is used by chatbots like chatgpt, gemini and claude.
These LLM were all engineered from one paper published by google called Attention is all you need. Within this paper they gave a design for a transformer which is the basis of all LLMs!
The goal of the Ai was to travel to the goal cube and it could only turn left, right and move forwards
It would be rewarded 2 points for touching the goal cube and would be penalised 1 point for touching the walls or not reaching the goal in the 30 second time frame(roughly 5000 steps) and it would aim to maximise the amount of points it could get
In order to speed up the training I had 65 models running in parallel, the 50 minutes training in the video would be equivilant to roughly 3250 minutes/54 hours of training with one iteration running
This demonstraights just how useful it is to have a powerful GPU when training machine learning algorithms
A diagram below, A popular concept frequently used in ml is neural networks, they work by having lots of nodes connected together in a network, each node has a bias and each connection has a weight, the inputs are put into the front layer of the neural network, from there the data is passed into a node on the next layer, multiplying the value by the weight and then adding the bias of the node it goes to, this happens from every node in the fornt layer to every node in the next layer(hidden layer) and happens for all layers, in the end the final layer is a list of the possible outputs it should give, the values that are in these final output nodes are added up and then transfered into probabilities of each output depending on what it wants, it then compares those probabilities to the actual output does a technique known as back propogation. This happens for every piece of training data per epoch amount of times, Making the Neural Network proficient at that task/reducing loss(the measure of how far off the correct answer they are)
Below is an interactive example of a linear regresion algorithm that draws a line of best fit through your points