You can deploy this model and start earning money today!

arturoornelasb/
triadic-gptneo-125m-v9

Name: triadic-gptneo-125m-v9
Availability: InStock

ontological-primitives

phase-transition

emergent-duality

representation-learning

triadic-duality

wave-fep

feature-extraction

Skylion007/openwebtext

EleutherAI/gpt-neo-125m

finetune

apache-2.0

Views

No views yet

Triadic GPT-Neo 125M (V9)

A 72-bit triadic projection head trained on top of frozen GPT-Neo 125M.

Paper DOI

Repo DOI

What is this?

This model adds a triadic head to GPT-Neo 125M that projects language model embeddings into 72-bit binary codes representing ontological primitives from the Emergent Duality framework.

Each bit corresponds to one of 72 semantic primitives organized in 6 algebraic layers (Boolean, Fuzzy, Ordinal, Modal, Trivalent, Probabilistic). The head learns to map continuous embeddings to discrete binary codes that preserve ontological structure.

Files

File	Description
`best.pt`	Final trained weights (best checkpoint at 50K steps)
`step_5000.pt`	Base checkpoint before phase transition (for X-ray reproducibility)
`run_config.json`	Full training configuration
`gold_extended_v7.json`	Gold standard: 72 primitives + extended concept targets

X-ray Reproducibility

step_5000.pt is the base checkpoint used for the telescopic X-ray analysis that revealed the cascade phase transition. To reproduce:

Resume training from step_5000.pt with save-every-5-steps
Run reptimeline on the resulting checkpoints
Observe the cascade activation at step ~5060 (31/72 bits change in waves over ~45 steps)
6 seed bits (L3-L5) trigger cascading activation to ontological extremes (L1, L6)

See train_v9_xray.bat for exact commands.

Training

Base model: GPT-Neo 125M (frozen)
Head: Deep triadic head (561K trainable params)
Activation: IFSQ (Integer Finite Scalar Quantization)
Dataset: OpenWebText (100K docs)
Steps: 50,000
Bit accuracy: 83.0% (test)
Subsumption: 97.0% (test)
Dead bits: 28/72
GPU: NVIDIA RTX 4060 Ti 16 GB

Key findings

Phase transition (cascade) at step ~5060 with wave propagation over ~45 steps
6 seed bits (L3-L5) trigger cascading activation to ontological extremes (L1, L6)
Wave-FEP fit: R²=0.734, underdamped regime (d/w=0.70), dAIC=+44.5 (wave preferred over exponential)
Cross-architecture cascade pattern confirmed with v8 (GPT-2 Medium)

Cross-architecture comparison

Metric	V8 (GPT-2 Medium)	V9 (GPT-Neo 125M)
R² (wave)	0.943	0.734
Regime	Overdamped (d/w=1.34)	Underdamped (d/w=0.70)
dAIC	+92.2	+44.5
Cascade duration	~145 steps	~45 steps
Bits involved	65/72 (90%)	31/72 (43%)
Preferred model	Wave	Wave

Usage

python

import torch
from huggingface_hub import hf_hub_download

# Download checkpoint
ckpt_path = hf_hub_download("arturoornelasb/triadic-gptneo-125m-v9", "best.pt")
checkpoint = torch.load(ckpt_path, map_location="cpu")

# Or use with triadic_extractor (pip install reptimeline)
from triadic_extractor import TriadicExtractor
extractor = TriadicExtractor(n_bits=72)
snap = extractor.extract(ckpt_path, concepts, device='cuda')

Citation

bibtex

@misc{ornelas2026triadic,
  author       = {Ornelas Brand, J. Arturo},
  title        = {Triadic Duality: Ontological Primitives as Emergent Algebraic Structure in Neural Representations},
  year         = {2026},
  doi          = {10.5281/zenodo.19375167},
  publisher    = {Zenodo},
}

arturoornelasb/triadic-gptneo-125m-v9