LLM - Day 1 - Intro

An LLM, or Large Language Model, is a type of model that functions as an equation. To understand equations, one must grasp the concept of a mathematical expression like y = mx + c or y = m1x2 + m2x3 + m3x2. In the realm of language models, these equations are represented by weights. For instance, the GPT-OSS-120B model utilizes the Transformer Architecture and produces outputs upon receiving data.

These weights can be adjusted to generate various desired outputs. The term "model" refers to a floating point storage unit containing multiple parameters, with 120B parameters indicating the number of weights within the model. The more context provided, the higher the likelihood of accurate predictions. Running a 120B parameter model on a machine requires substantial memory and processing power, which are supplied by RAM and CPU.

Thus, the choice of model depends on the available resources. Open weights models, such as ChatGPT or Gemini, are not the models themselves but rather tools used to run them. LLMs operate as next word predictors, generating a range of possible continuations for a given input. The model selects the most probable option based on a scoring system, typically ranging from 0 to 1.

Control parameters like temperature, top-k, and top-p allow users to manipulate the randomness and creativity of the generated text.