pub static FP8_E4M3_LUT: [f32; 256]Expand description
FP8 E4M3 -> f32 lookup table (256 entries, one per byte value).
OCP FP8 E4M3FN: sign(1) | exponent(4) | mantissa(3), bias = 7. There
are no infinities; 0x7F / 0xFF are NaN and max finite is +/-448.0
(exp = 15, mant = 6).
NaN entries decode to 0.0. A NaN weight should not exist in a
checkpoint, and zero stops one bad byte from poisoning an entire
dequanted tensor — which is what propagating NaN through the loader
would do.
Built at compile time so the hot dequant loop is a single indexed load with no branches.