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++
DeepSeek engine (AMD): the drafter's cumulative confidence cut on by default (0.5) on HIP
Niko1221e83b82afe562
diff --git a/src/dsv41/ds4_main.cpp b/src/dsv41/ds4_main.cppindex 57950f603..fdbd542e3 100644--- a/src/dsv41/ds4_main.cpp+++ b/src/dsv41/ds4_main.cpp@@ -152,11 +152,19 @@ static int g_round_nd = 0;// The drafts' cut on the product of their confidences (DeepInfer's cumulative rule): a draft is verified while the chance// that it and every draft before it are kept stays at least this - a long window whose later drafts are likely lost is// not verified whole. STRATA_DS4_DSPARK_CUM=F, a request's dspark_cum key; 0: off, only the floor on each draft's own-// confidence (dspark_conf). Off by default (not measured faster on code: H100 / owner PC); Project Maya runs 0.5 (a Radeon 8065S: decode +6.5%, the same tokens)+// confidence (dspark_conf). On AMD 0.5 is the default: measured on 4x R9700, DeepSeek 2.25 bpw, six 128-token+// decodes an arm in one load - 71.7 -> 76.9 tok/s (medians), the same ids; Project Maya runs 0.5 on a Radeon 8065S+// too (+6.5% decode, the same tokens), while the CUDA measurements in 0.2.7 had it flat, so CUDA keeps 0.+// STRATA_DS4_DSPARK_CUM overrides either way.float dspark_cum_default() {static const float v = [] {const char* e = std::getenv("STRATA_DS4_DSPARK_CUM");- return e != nullptr ? (float) std::atof(e) : 0.0f;+ if (e != nullptr) return (float) std::atof(e);+#if defined(STRATA_USE_HIP)+ return 0.5f;+#else+ return 0.0f;+#endif}();return v;}