WFGY/specs

WFGY – Public API Specs & ONNX Graphs

Version 1.0 · last updated 2025-06-12

WFGY is open by design.
To guarantee that each module can be inspected, audited, or re-implemented, we publish (1) a concise API spec and (2) a lightweight ONNX graph for every core component.
Download the .onnx files and open them in Netron or any ONNX viewer to see the exact computation graph—no black box, no hidden layers.

Module	API spec	ONNX graph
BBMC Semantic Residue	bbmc_api.md	onnx/bbmc.onnx
BBPF Progression Flow	bbpf_api.md	onnx/bbpf.onnx
BBCR Collapse–Rebirth	bbcr_api.md	onnx/bbcr.onnx
BBAM Attention Modulator	bbam_api.md	onnx/bbam.onnx

---

Re-exporting the graphs

All four graphs are intentionally tiny (~20 KB each) so they build in seconds and reveal the tensor flow without shipping proprietary weights.

# in repo root
python export_onnx.py

This script recreates the folder structure:

specs/
├── onnx/
│   ├── bbam.onnx
│   ├── bbcr.onnx
│   ├── bbmc.onnx
│   └── bbpf.onnx
└── *.md               # human-readable API specs

Feel free to replace these dummy graphs with full-precision weights if you need exact numerical reproduction.

---

License & Citation

The API descriptions and ONNX graphs are released under the MIT license. If you use WFGY in research, please cite the Zenodo release:

PSBigBig, “WFGY 1.0: A Self-Healing Variance Gate for LLMs,”
Zenodo, 2025, doi:10.5281/zenodo.15630970

⭐ Help us reach 10 000 GitHub stars to unlock WFGY 2.0! Any issue / PR is welcome—open science thrives on transparency.

---

### Will Colab runs change after adding this folder?

No. `specs/` is pure documentation.  
All examples (`example_01`–`08`), tests, and the SDK itself remain identical,
so you can keep running the original Colab notebook without any modification.