egora-diagnostics 0.2.1

EgoRA Diagnostics — Rotation geometry analysis and visualization for fine-tuned language models. Companion package for egora.

EgoRA Diagnostics

Companion package for egora.

This package installs egora with diagnostics dependencies (matplotlib, scipy) and re-exports the diagnostics API for convenience.

Installation

pip install egora-diagnostics

This is equivalent to:

pip install "egora[diagnostics]"

What You Get

• Rotation Geometry Analysis: compute_head_geometry() — per-head rotation angles, learning mode classification
• Interhead Diversity: compute_interhead_diversity() — similarity matrix across attention heads
• Publication-Quality Plots: plot_rotation_report() — combined 4-panel report figure
• CLI Tool: egora diagnose base_model tuned_model --plot

Quick Start

from egora_diagnostics import (
    compute_head_geometry,
    compute_interhead_diversity,
    save_geometry,
    plot_rotation_report,
    plot_rotation_heatmap,
    plot_layer_profile,
    plot_mode_distribution,
)

# Analyze rotation between base and fine-tuned model
results = compute_head_geometry(base_model, tuned_model)
print(f"Mean rotation: {results['theta_bar_deg']:.2f}°")

# Generate report
fig = plot_rotation_report(results)
fig.savefig("rotation_report.png", dpi=150)

Main Package

For the full EgoRA library (adapters, entropy governor, PEFT compatibility), install the main package:

pip install egora

See github.com/ArsSocratica/EgoRA for full documentation.

Citation

@article{dillerop2026rotation,
  title={The Rotation-Retention Law: Knowledge Loss Is Proportional to 
         Representational Rotation in Fine-Tuned Language Models.
         With EgoRA: Entropy-Governed Orthogonality Regularization 
         for Adaptation},
  author={Dillerop, Mark},
  year={2026},
  note={Preprint}
}

License

AGPL-3.0-or-later with Academic Additional Permission. See LICENSE-ACADEMIC.