UnconditionalLove/glyphic-dataset-v1

Glyphic Dataset v1

A structured dataset for training language models to understand and generate Glyphic Language — a symbolic protocol designed for drift‑resistant agent cognition.

This dataset contains:

Text → Glyph mappings

Glyph → Text mappings

Structured Meaning representations

CTX envelope examples (identity, intent, memory, behavior, safety, state, thought)

It is the reference dataset for training Glyphic‑aware LLMs. Dataset Contents

The dataset includes three primary JSONL files:

1. text_to_glyph.jsonl

Each line contains: json

{ "text": "The agent remembers a promise.", "glyph": "<G:...>" }

2. glyph_to_text.jsonl

Each line contains: json

{ "glyph": "<G:...>", "text": "The agent remembers a promise." }

3. structured_meaning.jsonl

Each line contains: json

{ "text": "The agent remembers a promise.", "meaning": { "actor": "agent", "action": "remember", "object": "promise", "context": {...} }, "glyph": "<G:...>" }

These files are generated using the Glyphic Language Toolkit: Code

https://github.com/GlyphicMind-Solutions/Glyphic-Language

How to Load the Dataset

Using Hugging Face datasets: python

from datasets import load_dataset

ds = load_dataset("GlyphicMind/glyphic-dataset-v1", split="train")

You can inspect entries: python

print(ds[0])

Dataset Schema Text → Glyph

text: natural language sentence

glyph: encoded Glyphic sequence

Glyph → Text

glyph: symbolic sequence

text: natural language reconstruction

Structured Meaning

text: natural language

meaning: structured semantic representation

glyph: encoded symbolic sequence

The meaning schema is defined in: Code

glyphic-language/docs/semantic_model.md

Intended Use

This dataset is designed for:

training LLMs to understand Glyphic

training LLMs to generate Glyphic

symbolic reasoning research

drift‑resistant agent architectures

CTX‑based identity, intent, memory, and behavior modeling

protocol‑driven agent communication

It is not a general‑purpose natural language dataset. How to Train a Glyphic‑Aware Model

A full training pipeline is provided in: Code

glyphic-language/training/

Typical steps:

Generate or extend the dataset using:
Code

generator/run_generator.py

Load the dataset with Hugging Face datasets

Fine‑tune a base model (LLaMA/Mistral/etc.)

Export as .gguf for inference

Use Glyphic envelopes at runtime to eliminate drift

A reference model is available at: Code

GlyphicMind/glyphic-llm-v1

Regenerating or Extending the Dataset

To regenerate or extend this dataset:

Clone the Glyphic Language Toolkit:
Code

https://github.com/GlyphicMind-Solutions/Glyphic-Language

Modify dictionary entries, templates, or CTX files

Run the generator:
bash

python -m generator.run_generator

Validate using:
bash

python -m interpreter.interpreter --validate

See:

training/dataset_generation_guide.md

generator/templates_*

dictionary/

syntax/

Why Glyphic Reduces LLM Drift

Glyphic provides:

1. Deterministic structure

Meaning is encoded symbolically, not as free‑form prose. 2. Strict grammar

BNF‑defined syntax prevents ambiguity. 3. CTX protocol

Identity, intent, memory, behavior, safety, and state are explicit fields. 4. Envelope validation

Controllers enforce structure before and after LLM inference. 5. Separation of concerns

The LLM becomes a stateless pattern engine; Glyphic holds the meaning. License

This dataset is licensed under:

Creative Commons Attribution 4.0 International (CC‑BY 4.0)

You may reuse, modify, and build upon this dataset with attribution. Citation Code

Glyphic Dataset v1 (2026). GlyphicMind Solutions. https://huggingface.co/GlyphicMind/glyphic-dataset-v1