NN.Floats.NeuralFloat

This is the “rounding-on-ℝ” side of TorchLean’s float model stack. We use it when we want to talk about formats and rounding generically, without committing to a concrete IEEE bit encoding:

• generic format/rounding infrastructure,
• the rounded scalar type NF,
• ULP-style error bounds and small calculation libraries.

Most files that are not bit-level IEEE execution live under NN/Floats/NeuralFloat/.