FP8_QUANT_HOPPER_ENTRY

Constant FP8_QUANT_HOPPER_ENTRY 

Source
pub const FP8_QUANT_HOPPER_ENTRY: &str = "per_token_group_quant_fp8_hopper";
Expand description

Entry point of the Hopper twin. A NEW name, not an override of the shared one: both kernels must be in the Hopper image at once, because the gate for the twin is byte equality against the shared kernel ON DEVICE.