| license: mit | |
| # Northstar Aurora | |
| NorthStar Aurora is a touch-first Christmas light show engine — it composes, sequences, and streams pixel-mapped lighting effects to Falcon F16 controllers over E1.31/DDP. The AI side of the project (Aurora AI) fine-tunes large language models to compose original light shows from audio input. | |
| **What lives here on Hugging Face:** | |
| - Benchmark and reproducibility data from training/inference workloads on consumer-grade AMD hardware (Strix Halo / gfx1151) | |
| - Eventually: open-sourced fine-tunes once their training pipelines are fully reproducible from public data | |
| **Related work:** | |
| - **[strix-halo-llm-finetune-guide](https://github.com/h34v3nzc0dex/strix-halo-llm-finetune-guide)** — the production guide for the hardware platform we train on. Everything from kernel patches through training scripts, with the troubleshooting record from six months of daily-driving the box | |
| - **[Paul Durkin (@H34v3nzC0d3x)](https://huggingface.co/H34v3nzC0d3x)** — the human behind the work | |
| Most of our model and dataset repos are kept private; the public ones are the pieces directly useful to the broader Strix Halo and open-LLM communities. | |