Jump to content

Connect Leeroopedia MCP: Equip your AI agents to search best practices, build plans, verify code, diagnose failures, and look up hyperparameter defaults.

Principle:Alibaba ROLL Distillation Validation

From Leeroopedia


Knowledge Sources
Domains Knowledge_Distillation, Evaluation
Last Updated 2026-02-07 20:00 GMT

Overview

An evaluation principle for monitoring knowledge distillation progress by computing student validation loss (SFT only, no distillation loss).

Description

Distillation Validation evaluates the student model on held-out data using SFT loss only (without teacher logits). This measures the student's standalone language modeling quality independently of the teacher.

Usage

Use at configured intervals during distillation training.

Theoretical Basis

Validation uses SFT loss only to assess standalone student quality without teacher dependency.

Related Pages

Implemented By

Related Heuristics

No specific heuristics inform this principle.

Page Connections

Double-click a node to navigate. Hold to expand connections.
Principle
Implementation
Heuristic
Environment