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++
sycl: STRATA_PF_XMX=2 is the default on Arc Alchemist (opt-out STRATA_PF_XMX=0); Battlemage unchanged
Niko122112fd8cff6562
diff --git a/docs/INTEL.md b/docs/INTEL.mdindex 41967e5ef..1cbe0c902 100644--- a/docs/INTEL.md+++ b/docs/INTEL.md@@ -759,6 +759,17 @@ Measured on an Arc A750 (8 GB, `i915`, PCIe 4.0) with the Flash-Next IQ3_XXS in8192 --prefill-repeat 5` for the time of both kernels. Not taken: the same kernel with its state in SLM (no registerspills, the same bits, ~4x slower: 13.7 vs 3.1-3.3 ms at T=1000).+### Prompt speed on Alchemist: the grouped XMX expert launch is the default (0.1.48, `STRATA_PF_XMX`)++The A750's prompt path decodes and multiplies a group of experts in one fused launch on the XMX units (`STRATA_PF_XMX=2`, `sycl/src/prefill/xmx_moe.dp.cpp`)+instead of dequantizing each expert to FP16 and calling oneMKL. It is the default on Alchemist cards; `STRATA_PF_XMX=0` turns it off and `=1` is the+variant that dequantizes first and runs one grouped XMX GEMM per product. Battlemage keeps the old path (the same kernels are 7-11% slower there).+Measured on the A750 (IQ3_XXS, default flags, interleaved rounds against the run before): a 1K-token prompt 35.1 -> 46.4 tok/s, 2K 61.7 -> 83.1, 4K 79.3 -> 110.4+(+39%, 5 of 5 rounds), 6K 66.3 -> 114.8; decode unchanged (18.7 tok/s), the same number of cache slots. It rounds differently from the FP16 path:+teacher-forced KL against the default over 6,006 forced tokens 0.0002 (the default against itself: 0.0001; with the default PCIe share the A750's own run-to-run noise+is 0.012), first-token KL 0.0017 (floor 0.0005), top-1 agreement 99.7%. HumanEval (every 2nd, 82) 73, 69 and 74 for the mode and two default runs, reasoning 21 / 23 / 22 of+30, a 35-question Chinese/Japanese/Korean set 35/35 in all, needles at 2K-6K 9/9, an 8K-context tool-call agent 6/6.+## Not done- Images, on both Intel engines. Strata's vision path encodes with `strata-vision` into embeddings the CUDAdiff --git a/sycl/src/prefill/xmx_moe.dp.cpp b/sycl/src/prefill/xmx_moe.dp.cppindex b25a058c2..8cbdecf72 100644--- a/sycl/src/prefill/xmx_moe.dp.cpp+++ b/sycl/src/prefill/xmx_moe.dp.cpp@@ -150,7 +150,11 @@ int mode() {// sub-group 16 (Xe2, Battlemage).static const int v = [] {const char* e = std::getenv("STRATA_PF_XMX");- const int want = e ? std::atoi(e) : 0;+ // 0.1.48: mode 2 (the fused grouped XMX launch) is the default on Arc Alchemist (A-series): A750 prompts +30..39% (4K), +32..35% (1-2K tokens),+ // first-token KL 0.002 against a 0.0005 floor, HumanEval / reasoning / agent / CJK / needles inside the run-to-run spread; Battlemage is+ // 7-11% slower with it, so it stays off there. STRATA_PF_XMX=0 turns it off, 1 is the dequantize + grouped XMX variant.+ const bool alchemist = strata::intel_gpu_gen(strata::q_of(nullptr)->get_device()) == strata::IntelGpuGen::Alchemist;+ const int want = e ? std::atoi(e) : alchemist ? 2 : 0;if (want == 0) return 0;#ifdef STRATA_NO_DG2_XMXstd::fprintf(stderr, "STRATA_PF_XMX ignored: this build is AOT for a device without the sub-group 8 XMX kernels\n");