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++

Merge f-dsv41-r97 fb321d62c into f-dsv41-rel (layer-major chunked prompt as the engine default)

Niko1221d6f18babe9cf
diff --cc src/dsv41/ds4_main.cpp
index 3378403b0,9fedb893a..73389d71f
--- a/src/dsv41/ds4_main.cpp
+++ b/src/dsv41/ds4_main.cpp
@@@ -847,8 -834,10 +847,10 @@@ int main(int argc, char** argv)
}
if (!o.vision_only) std::fprintf(stderr, "%s", m.describe().c_str());
if (serve_mode) return serve(m, eos, spec, o.dspark_conf);
+ const int window_given = window;
+ if (window == 0) window = 16; // (the scoring / profiling runs; the plain run below takes the model's feed unit)
if (!prof_ids.empty()) return profile_run(m, prof_ids, window);
- if (!spec_ids.empty()) return spec_eval(m, spec_ids, n, window, o.dspark_conf);
+ if (!spec_ids.empty()) return spec_eval(m, spec_ids, n, window, o.dspark_conf, spec);
if (!score_ids.empty()) return score(m, score_ids, score_out.empty() ? score_ids + ".tsv" : score_out, window, std::atoi(eos.c_str()));
if (!enc_rgb.empty()) { // the vision tower alone: a picture's span rows (BF16 [rows][dim]) into --dump-image