Skip to main content

XDNA_TOSA_FP8_TARGET

Constant XDNA_TOSA_FP8_TARGET 

Source
pub const XDNA_TOSA_FP8_TARGET: Target;
Expand description

The FP8 storage tier: graph-visible E4M3/E5M2 inputs explicitly cast to BF16 on the NPU.

This is separate from XDNA_TOSA_TARGET so adding the storage tier does not change the identity of the existing BF16 target. FP8 is not advertised as a compute dtype: every admitted graph has an FP8 block input, one explicit CAST, and a BF16 block output.