Skip to content

Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.

C++

layer-major prefill: declines on STRATA_EXPERT_NIBBLE builds (#3976)

Niko12219e2031906987
diff --git a/src/prefill/prefill.cpp b/src/prefill/prefill.cpp
index 9f7b1d29d..448a4ff06 100644
--- a/src/prefill/prefill.cpp
+++ b/src/prefill/prefill.cpp
@@ -5951,6 +5951,10 @@ int64_t Prefill::lm_prepare(int64_t n, std::string& why, int64_t cap) {
if (kva_config().on) { why = "STRATA_KVA is on"; return 0; }
if (stop_at_chunk) { why = "a batch read that yields at chunk boundaries"; return 0; }
if (force_chunk_ > 0) { why = "a --prefill-pipe plan sets the chunks of this read"; return 0; } // (left set for run_impl)
+#if STRATA_EXPERT_NIBBLE_BUILT // the 4-bit expert slots (STRATA_EXPERT_NIBBLE) are gathered for MMQ only by run_impl
+ for (int64_t l = 0; l < g.n_layers; ++l)
+ if (core::nibble::layer(l).on) { why = "STRATA_EXPERT_NIBBLE (the experts are 4-bit slots)"; return 0; }
+#endif
// run_impl's debug hooks on the routing and the rows have no layer-major form
if (std::getenv("STRATA_DBG_FEAT") || std::getenv("STRATA_DBG_ROUTE_OVERRIDE") || std::getenv("STRATA_PREFILL_DUMP_R_ALL")) {
why = "a routing or row debug dump";