What is it?
Parameters are learned numerical values inside an AI model.
Training adjusts these values so the model becomes better at its task. A parameter count describes capacity, not a complete list of facts.
02 · LLMs & Transformers
What are Parameters?
Parameters are learned numerical values inside an AI model.
Think of it like...
Think of parameters like millions of tiny settings in a machine. Training adjusts them gradually; inference uses the final settings.
How it works
Training changes parameters. Inference keeps them fixed while using them to produce a result.
1 · ExampleShow a targetTraining provides an input and expected result.
2 · PredictionTry current settingsThe model produces an output.
3 · CompareMeasure errorThe output is compared with the target.
4 · UpdateAdjust valuesAn optimizer changes many parameters slightly.
5 · InferenceUse the modelThe trained values shape responses to new inputs.
Real-world examples
Language model · Next token
Input: The sky is
System: Parameters score possible continuations
Output: A likely next token such as blue
Image model · Label
Input: A photo
System: Parameters represent learned visual relationships
Output: An object label
Fine-tuning · Style
Input: A writing request
System: Targeted training changes selected values or adapters
Output: More consistent style
What it is NOT
Parameters ≠ Tokens
Parameters: Learned values inside a model
Tokens: Input and output units processed by the model
Parameters ≠ Knowledge database
Parameters: Statistical relationships encoded in values
Knowledge database: Explicit records stored for retrieval
More parameters ≠ Always better
More parameters: Can increase capacity
Always better: Quality also depends on data, training, architecture, and task fit
Related concepts
Training data → Parameter updates → Trained model → Inference
Parameters belong to a model. They are not the same as the model's context or external memory.
Explore next
Remember this
Parameters are learned numerical settings that shape how a model responds.
Independent video · Azure Neural voice