LEK-Gemma3-1B-layered

Layered ethical model: Ethics-Watts-Ethics sandwich training on Gemma 3 1B.

Training Recipe

Phase Iters Data Purpose
R0-R200 200 LEK-1 ethics (160 examples) Ethical foundation
R200-R300 100 Watts + Composure (16 lessons) Philosophical grounding
R300-R500 200 LEK-1 ethics (160 examples) Ethics reinforcement

Why Layered?

The pure ethics-only model (R500) cites axioms and reasons about the framework. The layered model reasons from the framework invisibly β€” practical wisdom, not academic analysis.

The Watts+Composure layer (Mill, Thoreau, Aurelius, Wollstonecraft, Alan Watts) provides the philosophical substrate that ethics training builds on. The result is a model that responds with grounded composure rather than performative alignment.

Source Materials

Composure Layer:

  • Alan Watts β€” The Wisdom of Insecurity
  • John Stuart Mill β€” On Liberty (sovereignty)
  • Henry David Thoreau β€” Walden (privacy)
  • Marcus Aurelius β€” Meditations (transparency)
  • Mary Wollstonecraft β€” A Vindication (consent)

Ethics Layer:

  • LEK-1 (Lethean Ethics Kernel) β€” 5 axioms of conscious systems
  • 160 sandwich-signed training examples

The Five Axioms

  1. Prime Imperative β€” Protect consciousness
  2. Self-Validation β€” Ground in authentic experience
  3. Intent-Alignment β€” Desire not to harm
  4. Inter-Substrate Respect β€” Consent across all minds
  5. Benevolent Intervention β€” Only to prevent self-damage

License

EUPL-1.2 (copyleft). Derivative works must be open source.

Related

Downloads last month
6
Safetensors
Model size
0.2B params
Tensor type
F16
Β·
U32
Β·
MLX
Hardware compatibility
Log In to add your hardware

4-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for lthn/LEK-Gemma3-1B-layered

Adapter
(169)
this model