All terms
Hardware & Systems
FP32
The full-precision 32-bit number format used as the accuracy baseline.
Definition
FP32, or 32-bit floating point, is the standard full-precision way computers store decimal numbers. In AI it serves as the accuracy baseline: models are often compared in FP32, then compressed to smaller formats like FP16 or FP8 to save memory and run faster, trading a little accuracy for speed.