[Show & Tell] SpiralTorch — Rust-first ML stack (WebGPU/WASM/MPS/CUDA), trains in Z-space

Show and Tell
1

SpiralTorch — a Rust-first ML stack that trains in Z-space.
WGPU/WebGPU, MPS, CUDA, HIP; no NumPy/Torch shims.
Planner (A/B/C roundtable) self-rewrites SpiralK heuristics.

Model card: RyoSpiralArchitect/SpiralTorch · Hugging Face

Hello SpiralSession (Python):

from spiraltorch import SpiralSession, Tensor
s = SpiralSession(device="auto", curvature=-0.95)
x = Tensor(1,4,[0.1,-0.2,0.3,-0.4]); y = Tensor(1,2,[0.0,1.0])
print(s.train_epoch(x, y).average_loss)
1 Like
2

Happy to answer questions, esp. WGPU/WebGPU vs MPS/CUDA perf.

1 Like