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++
Metal: Gyro engine message fix, expert_parity prints the pipelines' limits (PIPE_INFO)
Niko1221a64f1dd99b0c
diff --git a/metal/src/engine/window.cpp b/metal/src/engine/window.cppindex a792aa5f8..25a892907 100644--- a/metal/src/engine/window.cpp+++ b/metal/src/engine/window.cpp@@ -521,8 +521,7 @@ bool Window::init(const strata::core::ModelGeometry& g, strata::core::SessionSta};upload(N, hsx_);upload(NFF, hsh_);- std::fprintf(stderr, "strata metal: Hadamard-folded experts (block %d): the gate/up and down inputs are rotated on the GPU-",+ std::fprintf(stderr, "strata metal: Hadamard-folded experts (block %d): the gate/up and down inputs are rotated on the GPU\n",had_block_);}}diff --git a/metal/tests/expert_parity.cpp b/metal/tests/expert_parity.cppindex 7c101c0f1..492855447 100644--- a/metal/tests/expert_parity.cpp+++ b/metal/tests/expert_parity.cpp@@ -365,6 +365,9 @@ int main(int argc, char** argv) {K.dn = cx.pipeline(lib, kname("moe_down", dn == ref::Q2_0 && !g_dn_prep ? 142 : dn, b));K.cgu = a;K.cdn = b;+ if (std::getenv("PIPE_INFO"))+ std::printf("pipelines: gu %s max threads %d tg mem %d | dn %s max threads %d tg mem %d\n", K.gu.name.c_str(), K.gu.max_threads,+ K.gu.static_tg_mem, K.dn.name.c_str(), K.dn.max_threads, K.dn.static_tg_mem);return K;};auto args_for = [&](const Kernels& K, int nt, int n_expert) {