Float8_E5M2 and Float8_E4M3FN currently have OVF as the default overflow policy:
which doesn't match many hardware implementations, and is less useful than SAT in practice:
6-bit floats and below are finite-only and therefore SAT by necessity. See also #23.
Float8_E5M2 and Float8_E4M3FN currently have OVF as the default overflow policy:
which doesn't match many hardware implementations, and is less useful than SAT in practice:
6-bit floats and below are finite-only and therefore SAT by necessity. See also #23.