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 split auto: STRATA_SPLIT_UNIFORM=1 restores the uniform miss cost
mesb88445169721
diff --git a/src/program/generate.cpp b/src/program/generate.cppindex 403fe9312..62be0cffe 100644--- a/src/program/generate.cpp+++ b/src/program/generate.cpp@@ -4073,11 +4073,15 @@ int main(int argc, char** argv) {}// #2007: relative to the fastest stage's link and only from a 4x gap on (split_rules::link_miss_scale): a// symmetric rig, or a moderate gap, keeps the fitted formula. STRATA_SPLIT_LINK_ABS=1: 0.1.41's 20 GB/s rule.+ // STRATA_SPLIT_UNIFORM=1: no per-link scaling at all (the pre-0.1.41 uniform miss cost), for A/B on links+ // where any scaling moves auto off the balanced placement.static const bool link_abs = [] { const char* v = std::getenv("STRATA_SPLIT_LINK_ABS"); return v && v[0] == '1'; }();+ static const bool split_uniform = [] { const char* v = std::getenv("STRATA_SPLIT_UNIFORM"); return v && v[0] == '1'; }();std::vector<double> link_scale = strata::program::split_rules::link_miss_scale(link_gbps);for (int i = 0; i < ns; ++i) {const double bw = link_gbps[(size_t) i];- if (link_abs) link_scale[(size_t) i] = bw > 0.0 ? std::clamp(20.0 / bw, 1.0, 64.0) : 1.0;+ if (split_uniform) link_scale[(size_t) i] = 1.0;+ else if (link_abs) link_scale[(size_t) i] = bw > 0.0 ? std::clamp(20.0 / bw, 1.0, 64.0) : 1.0;std::fprintf(stderr, "strata generate: layer split auto: CUDA%d link %.1f GB/s -> miss cost x%.2f\n",i == 0 ? 0 : stages[(size_t) i - 1]->dev, bw, link_scale[(size_t) i]);}