ParametersConcepts

Model parameters

3 min readbeginnerUpdated 28 Sept 2026
1 · In one line

Parameters are the model values that training can change to improve how inputs map to outputs.

1 · What it is

TensorFlow uses parameter as a broad name for model values learned from data. In its dense-layer example, the kernel and bias are parameters of the affine mapping.

Training decides how much to change each learnable parameter. PyTorch places learnable weights and biases in a model’s parameters.

The architecture specifies which layers a model contains and how they are connected.

2 · Why it exists

TensorFlow describes a machine learning model as a function with learnable parameters.

Store learningTensorFlow says optimal parameters are obtained by training the model on data.
Shape mattersDense-layer parameters include a kernel and a bias.
Count parametersModel summaries can report the total number of parameters.
3 · How it works

Follow one dense layer from parameter shapes to an output.

A dense layer uses a kernel and a bias as parameters of its affine mapping.
  1. 1 · shapeThe layer defines a parameter tensor with a known shape.
  2. 2 · inspectInspect the framework's reported parameter total.
  3. 3 · computeThe layer multiplies the input by the kernel and adds the bias.
  4. 4 · updateTraining decides how much to change each learnable parameter.

Frameworks expose parameter tensor shapes.

4 · Where it's used
WhoWhat they askWhat it works with
Model engineer“Which tensors will this layer learn?”Parameter names, shapes and trainable status
Training engineer“Which values should the optimizer update?”The trainable parameter collection
Deployment engineer“Which model state must be restored?”Saved parameter tensors
Researcher“Did model capacity change between experiments?”Architecture and parameter count
5 · What it solves, and what it doesn't
solves
  • Parameters give training values it can adjust from data.
  • A dense layer's kernel and bias are parameters of its affine mapping.
  • A framework can mark a layer weight as trainable.
  • A saved parameter dictionary can restore learned model state.
doesn't solve
  • Not every item in model state is trainable.
6 · Go deeper

Sources used

This explainer is written in original language. The links below support its factual claims.

  1. docsTraining models, TensorFlow · read 28 Sept 2026
  2. docsIntroduction to modules, layers, and models, TensorFlow · read 28 Sept 2026
  3. docsMaking new layers and models via subclassing, Keras · read 28 Sept 2026
  4. docsSaving and Loading Models, PyTorch · read 28 Sept 2026
  5. docsSave, serialize, and export models, TensorFlow · read 28 Sept 2026