Decoder. plain-English AI glossary

Forward pass

● Core

Running input through the model start to finish to get an output — no learning, just computing.

Think of it like

Feeding a coin into a vending machine and watching the can drop: input goes in, result comes out, nothing about the machine changes.

Example

Every time you send a prompt and get an answer, that’s one forward pass through billions of parameters.

How it actually works

Data flows from the input through each layer’s math to the output. During training this is paired with a backward pass that computes gradients; at inference it’s all you run. It’s what you pay for per request — compute proportional to model size and sequence length.

For product teams

Inference cost is basically the price of one forward pass — bigger model or longer input, bigger bill.

For engineers

Evaluation of the network’s function on an input, producing activations and logits, without gradient updates.

Logits — At output it produces logits.
Inference — Running it is inference.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open Decoder Add to Chrome