Parameters
LLM & Language ModelsThe learned numerical values inside a neural network that determine its behavior — more parameters generally means more capability. GPT-4 is estimated to have 1.8 trillion parameters.
Parameters are the numbers that a neural network adjusts during training to capture patterns in data. They're the model's 'knowledge' — encoded not as facts in a database but as billions of numerical weights that together produce intelligent behavior.
Model size is usually measured in parameters: Llama 8B has 8 billion, Llama 70B has 70 billion, GPT-4 is estimated at 1.8 trillion. Generally, more parameters = more capable, but architecture matters too — a well-designed 70B model can outperform a poorly designed 175B model.
Parameter count affects practical considerations: larger models need more GPU memory (VRAM) to run, generate responses slower, and cost more per token. This is why AI providers offer multiple model sizes — you pick the smallest model that handles your task well.
Real-World Example
When you choose between GPT-4 (large) and GPT-4o-mini (small) you're choosing between different parameter counts — more parameters mean more capability but higher cost and slower speed.
Related Terms
More in LLM & Language Models
FAQ
What is Parameters?
The learned numerical values inside a neural network that determine its behavior — more parameters generally means more capability. GPT-4 is estimated to have 1.8 trillion parameters.
How is Parameters used in practice?
When you choose between GPT-4 (large) and GPT-4o-mini (small) you're choosing between different parameter counts — more parameters mean more capability but higher cost and slower speed.
What concepts are related to Parameters?
Key related concepts include Training, VRAM (Video RAM), GPU (Graphics Processing Unit), LLM (Large Language Model), Foundation Model. Understanding these together gives a more complete picture of how Parameters fits into the AI landscape.