EngineersofAI LogoCoursesBlogResearch LabAI LettersThe LabCode BankInteractive 3DKodrEarnestJobs
Skip to main content
Interactive 3D/SLERP - Spherical Interpolation for Model Merging
Weight Space Visualization (2D projection of hypersphere)
ABSLERP (on sphere)LERP (chord, shrinks)
Model A (coding, reasoning, math)
Model B (creative, dialogue, roleplay)
Angle between: 45°
t = 0.50
Norm Preservation
SLERP norm
1.000
Preserved ✓
LERP norm
0.924
Shrinks ✗
SLERP interpolates along the great circle - preserving the vector norm. LERP takes the chord, shrinking the vector. Shrunk weight vectors underfit. SLERP avoids this by staying on the hypersphere surface.
Interpolated Model Capabilities (t=0.50)
coding
reasoning
math
creative
dialogue
roleplay
Model A dominantModel B dominant
Controls
Interpolation t
t0.50
AB
Model Pair
A+B
A+C
B+C
Path Mode
SLERP (Spherical Linear Interpolation): interpolates along the great circle on the unit sphere. Preserves weight norms and produces smoother capability blending than naive LERP.

SLERP - Spherical Interpolation for Model Merging - Interactive Visualization

SLERP (Spherical Linear Interpolation) interpolates between two weight vectors along the great circle of the unit sphere, preserving vector norms throughout. LERP (linear interpolation) takes the chord through the sphere, causing the intermediate vector to shrink - shrunk weight vectors underfit because the model's effective scale is reduced. SLERP prevents this by keeping all interpolated points on the same hypersphere surface as the original models.

  • SLERP path: great circle arc on unit sphere - norm always 1.0 throughout
  • LERP path: straight chord - norm shrinks to minimum at t=0.5 (max by cos(θ/2) factor)
  • At angle=90°, LERP norm drops to 0.707 at midpoint - significant capability degradation
  • SLERP capability blending: at t=0.5, model has roughly equal mix of both model capabilities
  • MergeKit and HuggingFace merge tools implement SLERP for practical model merging

Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.