| --- |
| language: |
| - en |
| license: mit |
| tags: |
| - synthetic |
| - demo |
| - text-generation |
| - educational |
| pipeline_tag: text-generation |
| --- |
| |
| # NebulaText-125M-Demo |
|
|
| **A fictional model concept with an executable toy text generator.** The name “125M” is purely branding: this repository **does not contain a 125-million-parameter model**, pretrained weights, or benchmark results. |
|
|
| ## Overview |
|
|
| NebulaText is a speculative concept for a lightweight language model with a fictional *Orbit Memory Routing* architecture. This repository provides a small, working word-level Markov-chain demonstration trained on a short synthetic example corpus. It is meant for Hugging Face repository setup and experimentation—not production inference. |
|
|
| ### Concept (not implemented) |
|
|
| - Orbit Memory Routing (OMR): fictional recurrent context router |
| - Adaptive Token Folding (ATF): fictional compute optimization |
| - Hypothetical architecture only; no empirical performance claims |
|
|
| ## Run the included demo |
|
|
| Python 3.9+; no third-party dependencies: |
|
|
| ```bash |
| python generate.py --prompt "The future of AI" --max-new-words 35 --seed 7 |
| ``` |
|
|
| You can also train the tiny word-transition model on your own plaintext corpus: |
|
|
| ```bash |
| python train.py --input corpus.txt --output transitions.json |
| python generate.py --model transitions.json --prompt "AI systems" --max-new-words 30 |
| ``` |
|
|
| The included `transitions.json` is a count table, **not neural weights**. Outputs may be repetitive or nonsensical and shouldn't be treated as facts. |
|
|
| ## Model details |
|
|
| | Field | Value | |
| |---|---| |
| | Model type | Toy word-level Markov generator | |
| | Neural parameter count | N/A | |
| | Training source | Included synthetic demonstration corpus | |
| | Intended use | Software demos, tutorials, repository experiments | |
| | Not intended for | Production generation, factual QA, medical or financial decisions | |
|
|
| ## Limitations and ethics |
|
|
| This is a deterministic seeded sampling demo, not an LLM. It has no verified safety, accuracy, or robustness properties. Any future experimental research results should be independently measured before publication. Not affiliated with NVIDIA, Hugging Face, or any AI laboratory. |
|
|
| ## Repository layout |
|
|
| - `generate.py`: sample text using a small transition table |
| - `train.py`: build transitions from a local text file |
| - `transitions.json`: synthetic corpus transition counts |
| - `corpus.txt`: example source sentences |
| - `config.json`: demo metadata |
| - `LICENSE`: MIT |
|
|
