page

Curriculum Learning

What It Is

Ordering training data from simple to complex, so the model learns foundational concepts before advanced ones.

How It Accelerates Training

  • Model doesn't waste compute on complex examples it can't understand yet
  • Gradients are more stable in early training
  • Research shows 20-50% training time reduction for some tasks

Implementation for LARS

  1. Sort datasets by complexity:
  2. Short prompts/responses first
  3. Single-concept examples before multi-concept
  4. Nexus basics before advanced workflows

  5. Stage the training:

  6. Stage 1: Identity (Who is LARS?)
  7. Stage 2: Basic tasks (simple commands)
  8. Stage 3: Complex reasoning (3D dataset)
  9. Stage 4: Multi-step workflows

  10. Use difficulty scoring:

  11. Count tokens, nested concepts, required context
  12. Auto-sort datasets by difficulty score

Research References

  • Bengio et al. 'Curriculum Learning' (2009)
  • Self-paced learning variants
  • Competence-based curriculum (measure mastery before advancing)
ID: 34e7d2f6
Path: Accelerated AI Training > Training Acceleration Methods > Curriculum Learning
Updated: 2026-01-01T19:27:28