| --- |
| title: Monumental Systems - Model Showcase |
| emoji: "🌊" |
| colorFrom: indigo |
| colorTo: purple |
| sdk: static |
| app_file: index.html |
| pinned: true |
| short_description: Explore spectral, rotor, oscillator, and byte models. |
| license: mit |
| tags: |
| - geometric-deep-learning |
| - language-models |
| - harmonic-gpt |
| models: |
| - MonumentalSystems/harmonic-gpt-spherical-kan-128m-clean |
| - MonumentalSystems/harmonic-gpt-spherical-kan-128m-maxine |
| - MonumentalSystems/rotor-25m-annealed |
| - MonumentalSystems/wire-9m-best |
| - MonumentalSystems/harmonic-gpt-128m-byte-agent-multiparty |
| - MonumentalSystems/harmonic-gpt-128m-byte-chat |
| --- |
| |
| # Monumental Systems Model Showcase |
|
|
| A lightweight, static front door to our architecture research and interactive demos. |
| Model descriptions are grounded in the linked public model cards. Runnable demos and |
| downloadable research checkpoints have separate labels; archives and conventional |
| Transformer controls are identified explicitly. |
|
|
| - SpectralKAN Lab: native Clean and Maxine recurrent inference on free CPU hardware. |
| - Rotor 25M: the existing demo corrected to load its advertised Annealed checkpoint. |
| - WIRE 9M: weights and configuration for research inspection; no hosted runtime bundled. |
| - Harmonic Byte Chat: CPU frontend backed by Apple Metal on the Mac. |
| - Multiparty dialogue: model weights with participant-aware native byte framing. |
|
|
| No paid hardware or ZeroGPU quota is used by the showcase. |
|
|
