Argo1-OOAS/TinyLlama-1.1B-V3-EdgeTravel

Recommended Usage

For the easiest way to use this model with all required settings and its full functionality, use the TrajectoryLM application.

Screenshots

TinyLlama-1.1B V3 — edge/travel trigger

This model scales the contextual edge/travel trigger branch to a frozen TinyLlama-1.1B chat core. Only 16,781,313 branch parameters were optimized; the official Transformer weights remained unchanged. A trigger can activate the branch for a configurable number of following tokens, optionally conditioned on user text.

Property	Value
Total parameters	1,116,829,697
Trained branch parameters	16,781,313
Layers / width	22 / 2,048
Context length	2,048
Completed branch updates	2,000
Token presentations	262,144,000
Terminal sampled validation loss	0.5318 (PPL 1.7)

The validation figure is a terminal training-loop estimate for the selected branch lesson objective, not ordinary all-token TinyLlama perplexity.

UI usage

git clone https://github.com/Argo1-OOAS/TrajectoryLM.git
cd TrajectoryLM

Run start_windows.ps1 on Windows or start_macos.command on macOS, following the TrajectoryLM README. The cross-platform launcher can fetch missing files with download progress. The UI highlights triggers and affected tokens and audits the branch-off versus final token choices. See research_paper.pdf for the complete report.

Limitations

Strong context guidance can damage fluency or cause trigger loops. The model is experimental, has no broad safety/capability evaluation, and must not be used as a reliable source for high-impact decisions.

Citation

@misc{argo1ooas2026tinyllamav3,
 title={Scaling Trigger-Scoped Edge/Travel Adaptation to TinyLlama-1.1B},
 author={Argo1-OOAS}, year={2026},
 url={https://huggingface.co/Argo1-OOAS/TinyLlama-1.1B-V3-EdgeTravel}
}