| license: mit | |
| tags: | |
| - harmonic-gpt | |
| - oscillator-dynamics | |
| - clifford-algebra | |
| - byte-level | |
| - from-scratch | |
| - monumental-systems | |
| language: en | |
| pipeline_tag: text-generation | |
| # Wire-9M (H32 L8) | |
| WireNative 9M — n_harmonics=32, 8 layers, best BPB checkpoint | |
| Part of the [Harmonic GPT](https://github.com/DavinciDreams/harmonic-gpt) research into oscillator-based neural computation. | |
| ## Architecture: WireNative | |
| | Property | Value | | |
| |----------|-------| | |
| | Parameters | 8,894,520 | | |
| | BPB | 3.0866 | | |
| | Training step | 5,000 | | |
| | n_harmonics | 64 | | |
| | n_layers | 8 | | |
| | n_groups | 7 | | |
| | d_model | 896 | | |
| | Vocab | 256 (raw bytes) | | |
| ## Usage | |
| ```python | |
| from huggingface_hub import hf_hub_download | |
| from safetensors.torch import load_file | |
| weights = load_file(hf_hub_download("MonumentalSystems/wire-9m-best", "model.safetensors")) | |
| config = json.load(open(hf_hub_download("MonumentalSystems/wire-9m-best", "config.json"))) | |
| ``` | |
| All operations are native Clifford algebra / harmonic oscillator dynamics — no softmax attention, no MLP, no ReLU. | |