For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).
Mojo function
e4m3_magnitude_lut
def e4m3_magnitude_lut(scale: Float32) -> SIMD[.uint32, 2]
e4m3(m * scale) for the e2m1 magnitudes {0, .5, 1, 1.5, 2, 3, 4, 6}.
Byte i is the entry for magnitude code i, laid out as the two words
byte_permute indexes. m * scale is exact in f32 and is rounded once to
e4m3 (nearest even, which is sign-symmetric), so
e4m3(-x) == e4m3(x) | 0x80. The product needs up to 6 significant bits, so
it is not always exact in e4m3; a scale from a real quantizer keeps 6 * scale inside the e4m3 range.
Returns: