Skip to main content

MOE_FUSED_EPI_DISPATCHES

Static MOE_FUSED_EPI_DISPATCHES 

Source
pub static MOE_FUSED_EPI_DISPATCHES: AtomicU64
Expand description

Fused MoE-epilogue dispatches taken since process start (MEMRA_MOE_FUSED_EPI), incremented once per (token, layer) that actually runs moe_fused_epi_token_q8.

This exists because the arm cannot be observed any other way: setting MEMRA_MOE_STATS / MEMRA_MOE_TRACE / MEMRA_MOE_WEIGHT_TRACE / MEMRA_MOE_INPUT_TRACE_DIR sets observe_routes in moe_ffn_inner, which DIVERTS dispatch to the host-routed path — so a gate that tried to prove the fused arm ran by tracing would prove it about a different program. Read it via moe_fused_epilogue_dispatches around a workload.