It is intentionally not a public trained-weights release. The core trained implementation, model weights, checkpoints, CUDA kernels, datasets, tokenizer files, and training recipes remain proprietary and are not released here.

Repository contents

This repository contains a public Hugging Face compatibility / architecture surface for the Waveformer line:

README.md - public evidence map and validation-status summary
config.json - public Waveformer configuration metadata
configuration_waveformer.py - public custom Transformers config class
modeling_waveformer.py - public architecture-surface wrapper

This repository does not contain:

trained weights
tokenizer files
checkpoints
CUDA kernels
datasets
exact data curation recipes
training recipes
the proprietary trained Waveformer/System Omega engine

Because trained weights are not included, this repository should not be read as a normal downloadable model checkpoint. It is a public proof/provenance and architecture landing surface for the Waveformer/System Omega research line.

Public evidence chain

Date	Record	DOI	Role
2026-05-27	Continuous Oscillator Intelligence Runtime - Timestamp Proof	https://doi.org/10.5281/zenodo.20412409	Timestamp and priority proof using SHA256 hashes of source/checkpoint artifacts.
2026-06-16	Deterministic Layer Freezing via Continuous Phase Coherence Analysis	https://doi.org/10.5281/zenodo.20720827	Phase-coherence layer analysis across GPT-2, Qwen2.5, and Mistral scales.
2026-06-18	Emergence Over Attention: Continuous-Time Phase Synchronization as a Computational Primitive	https://doi.org/10.5281/zenodo.20741536	Waveformer architecture and training report.
2026-06-23	Non-Autoregressive Semantic Steering of Discrete Language Models via Continuous-Time Phase Manifolds	https://doi.org/10.5281/zenodo.20819119	System Omega semantic-steering report and raw validation metrics.
2026-06-29	Waveformer / System Omega: Public Evidence Map and Current Validation Status	https://doi.org/10.5281/zenodo.21044769	Consolidates the public DOI trail, private IP boundary, current validation status, and explicit non-claims.

How to read this repository

The public records above are different proof classes:

The May 27 record is a timestamp/provenance anchor.
The June 16 record is a phase-coherence method and empirical report.
The June 18 record is the Waveformer architecture and training report.
The June 23 record is the System Omega semantic-steering report.
The June 29 record is a consolidation/status technical note that maps the public proof trail, private IP boundary, current validation status, and explicit non-claims.
This Hugging Face repository is the AI-community landing surface tying those records together.

Public reproducibility would require releasing trained weights, source code, kernels, checkpoints, datasets, and recipes. That is intentionally outside the current public release boundary.

Current validation status

The current strongest result is not chatbot replacement or product-grade token generation. It is runtime-state event forecasting.

Private audit summary, 2026-06-29:

System Omega runtime state predicts future internal error spikes on real telemetry with AUC around 0.95 to 0.99 across tested horizons.
Fixed-state online operation was verified over a 1,048,576-token stream with flat allocated tensor memory.
The load-bearing mechanism is saved phase/velocity plus bidirectional cross-coupling; ablation destroys the low-mode fossil when velocity or cross-coupling is removed.
Predictive-maintenance/degradation streams are the strongest external use-path, especially when Omega/Waveformer state is combined with native features.

Current non-claims

This repository does not claim that:

Waveformer currently beats frontier LLMs.
The current public repository is a runnable trained model release.
The current token-generation UX is product-grade.
Markets, earthquakes, or Wikipedia attention streams are standalone product-grade wins.
Current Fractal/Phi-4 LoRA bridge artifacts are crystallized.
Public reproducibility is provided.

This repository also does not release proprietary source code, trained weights, checkpoints, CUDA kernels, datasets, tokenizer files, or training recipes.

Usage

This is not a public trained-weights release.

Do not expect the following to load a trained Waveformer checkpoint from this repository:

python

from transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(
    "wannaq/Waveformer",
    trust_remote_code=True,
)

The trained weights are intentionally not included. The repository is maintained as a public proof/provenance and architecture surface. See the Zenodo DOI chain above for the formal public records.

Research direction

Based on the strongest surviving evidence, the most credible current direction is stateful event forecasting and monitoring: detecting future instability, degradation, failure, or regime change in continuous streams.

Relevant private application areas include:

local AI telemetry monitoring
experiment-governor systems
predictive maintenance and degradation monitoring
GPU/training health monitoring
autonomous verification pipelines
closed-source research engineering
private continuous-state event-prediction systems

Citation anchors

bibtex

@misc{bajddi2026runtime,
  author = {Bajddi, Nossair},
  title = {Continuous Oscillator Intelligence Runtime - Timestamp Proof},
  year = {2026},
  publisher = {Zenodo},
  doi = {10.5281/zenodo.20412409},
  url = {https://doi.org/10.5281/zenodo.20412409}
}

@misc{bajddi2026phasecoherence,
  author = {Bajddi, Nossair},
  title = {Deterministic Layer Freezing in Autoregressive Language Models via Continuous Phase Coherence Analysis},
  year = {2026},
  publisher = {Zenodo},
  doi = {10.5281/zenodo.20720827},
  url = {https://doi.org/10.5281/zenodo.20720827}
}

@misc{bajddi2026waveformer,
  author = {Bajddi, Nossair},
  title = {Emergence Over Attention: Continuous-Time Phase Synchronization as a Computational Primitive},
  year = {2026},
  publisher = {Zenodo},
  doi = {10.5281/zenodo.20741536},
  url = {https://doi.org/10.5281/zenodo.20741536}
}

@misc{bajddi2026omega,
  author = {Bajddi, Nossair},
  title = {Non-Autoregressive Semantic Steering of Discrete Language Models via Continuous-Time Phase Manifolds},
  year = {2026},
  publisher = {Zenodo},
  doi = {10.5281/zenodo.20819119},
  url = {https://doi.org/10.5281/zenodo.20819119}
}

@misc{bajddi2026evidencemap,
  author = {Bajddi, Nossair},
  title = {Waveformer / System Omega: Public Evidence Map and Current Validation Status},
  year = {2026},
  publisher = {Zenodo},
  doi = {10.5281/zenodo.21044769},
  url = {https://doi.org/10.5281/zenodo.21044769}
}