FP8_GROUPED_GATE_UP_ROWS_PER_PASS

Constant FP8_GROUPED_GATE_UP_ROWS_PER_PASS 

Source
pub const FP8_GROUPED_GATE_UP_ROWS_PER_PASS: u32 = 4;
Expand description

2026-09-26: Rows per gate+up pass. Must equal GU_GROUP_ROWS in the .cu: the shared expert takes ceil(num_tokens / this) block rows.