Fine-tuned Llama-3.2-1B — pure text continuationchecking…
Type anything (a title, some MIDI lines, whatever) and click Generate.
The model appends tokens right into the same box; keep clicking Generate to extend.
▶ Advanced settings
higher = more chaos
nucleus sampling cutoff
How the GPU status light is inferred
🔊 Audiohover to refresh from workspace
(no notes yet — generate some, then hover here)
🎹 Piano rollhover to refresh from workspace
(no notes yet — generate some, then hover here)
Load an example (replaces current text):
Backend: Modal L4 (mistral.rs, ≈91 tok/s greedy / 77 tok/s sampling) with a 170s warmth window.
Falls back to CPU (transformers, ≈0.5 tok/s) if the GPU has cooled down. See Advanced settings above.