Deployment & Inference · Level 3 of 5
Distillation
Training a student model to reproduce selected behavior of a teacher.
A student need not match all of the teacher's capabilities.
Example
A compact network learns from a larger model's probability outputs.
Listen to the definition and example
Audio transcript
Distillation. Training a student model to reproduce selected behavior of a teacher. A student need not match all of the teacher's capabilities. For example: A compact network learns from a larger model's probability outputs.
Explore this concept
Why it matters
This helps you balance prediction quality with memory, latency, and throughput.
Start with
Related concepts
Quick recall question
Try answering before looking back at the definition.