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++

PLE residency policy: the load log also says the expert arena size the rule counted (the proof that the layout is initialised and the sum is not 0)

Niko1221bc5f84fd62b2
diff --git a/src/program/generate.cpp b/src/program/generate.cpp
index 745c79f87..53736e867 100644
--- a/src/program/generate.cpp
+++ b/src/program/generate.cpp
@@ -4111,9 +4111,10 @@ int main(int argc, char** argv) {
if (ple_table.open(o.ple_gguf, ram_err, pio)) {
ple_auto_ram = true;
std::fprintf(stderr, "strata generate: --ple-io ram chosen by the residency policy (the table is "
- "%.1f GiB and %.1f GiB of RAM is available; STRATA_PLE_AUTO_RAM=0 keeps the "
- "direct reads, --ple-io direct forces them)\n",
- (double) need / (1ull << 30), (double) *available / (1ull << 30));
+ "%.1f GiB, the expert arena still to fill %.1f GiB and %.1f GiB of RAM is available; "
+ "STRATA_PLE_AUTO_RAM=0 keeps the direct reads, --ple-io direct forces them)\n",
+ (double) need / (1ull << 30), (double) experts_ram / (1ull << 30),
+ (double) *available / (1ull << 30));
} else {
// Back to what was asked for; a table that cannot be mapped must not stop the engine.
std::fprintf(stderr, "strata generate: resident PLE table failed (%s); keeping direct reads\n",