Why it matters
FP8 is next precision frontier. Understanding shapes modern training.
Advertisement
The architecture
E4M3: 4-bit exponent, 3-bit mantissa.
E5M2: 5-bit exponent, 2-bit mantissa.
Per-tensor scaling.
Advertisement
How it works end to end
Transformer Engine (TE) library.
Delayed scaling: use previous batch stats.
Amax history tracking.
Master weights in fp32.