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: the FP16 path needs no MMQ identity table or token upload buffer (#3976)
Norihiro Nishimura5602253e3f3b
diff --git a/src/prefill/prefill.cpp b/src/prefill/prefill.cppindex 8e8f1df89..a6ace0b56 100644--- a/src/prefill/prefill.cpp+++ b/src/prefill/prefill.cpp@@ -5665,7 +5665,7 @@ int64_t Prefill::lm_prepare(int64_t n, std::string& why) {}}if (m.fused_bufs) { why = "the fused expert layout"; return 0; }- if (!m.GU || !m.Dm || !m.ids_identity || !m.tok_dev) { why = "the MoE buffers are missing"; return 0; }+ if (!m.GU || !m.Dm || (n_mmq > 0 && !m.ids_identity)) { why = "the MoE buffers are missing"; return 0; }// (an explicit STRATA_PREFILL_LM_TOKENS may be one chunk: the A/B against the default path, experts aside)const int64_t min_s = lm_tokens_env() > 0 ? m.T : 2 * m.T;if (n < min_s) { why = "a short prompt"; return 0; }