Expand description
Toto-2 inference on the Burn backend (Flex CPU).
Patched input scaler, transformer with GQA time/variate layers, xPos RoPE,
PerDimScale, unit-scaled linears, SwiGLU FFN, and quantile output head,
with weights loaded straight from GGUF via zsfm_burn::GgufWeights.
The compute_f64 path is not ported: pass a config with
compute_f64 = false (the default). The Burn port bails otherwise.