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++

fix: list the pool's own worker cores for the start-up line and --aux-cpus under the NUMA split

Baptiste Rajaut2614b82c9a87
diff --git a/include/strata/kernels/cpu/pool.hpp b/include/strata/kernels/cpu/pool.hpp
index 9dcb349bd..98320a925 100644
--- a/include/strata/kernels/cpu/pool.hpp
+++ b/include/strata/kernels/cpu/pool.hpp
@@ -209,6 +209,8 @@ public:
bool host_works() const { return host_works_; }
bool is_hybrid() const { return topo_.is_hybrid; }
+ /// The CPUs worker i is pinned to, in worker order (STRATA_NUMA_SPLIT reorders them: interleave_packages).
+ const std::vector<int>& worker_cores() const { return topo_.worker_cores; }
int p_cores() const { return topo_.p_cores; }
int p_threads() const { return topo_.p_threads; }
int e_cores() const { return topo_.e_cores; }
diff --git a/src/program/generate.cpp b/src/program/generate.cpp
index 427957b9d..7366427a0 100644
--- a/src/program/generate.cpp
+++ b/src/program/generate.cpp
@@ -5835,9 +5835,11 @@ int main(int argc, char** argv) {
}
{ // where the threads run (--host-core)
const strata::kernels::cpu::CpuTopology ht = strata::kernels::cpu::detect_cpu_topology(true, o.pool_affinity);
+ // the pool's own order, not a fresh detection's: STRATA_NUMA_SPLIT interleaves the sockets
+ const std::vector<int>& wc = pool.worker_cores();
std::string on;
- for (int i = 0; i < pool.workers() && i < (int) ht.worker_cores.size(); ++i)
- on += (i > 0 ? "," : "") + std::to_string(ht.worker_cores[(size_t) i]);
+ for (int i = 0; i < pool.workers() && i < (int) wc.size(); ++i)
+ on += (i > 0 ? "," : "") + std::to_string(wc[(size_t) i]);
std::fprintf(stderr, "strata generate: %d pool workers on logical processors %s, the host thread on %d%s "
"(--host-core %s)\n", pool.workers(), on.c_str(), ht.host_core,
pool.host_works() ? " (draining too)" : "",
@@ -5846,10 +5848,9 @@ int main(int argc, char** argv) {
: "first");
// --aux-cpus: the plan, now the workers' CPUs are known; the threads that exist already (the CUDA driver's) move
// now when it is on, the ones created later pin themselves (strata/platform/aux_cpus.hpp)
- const size_t nw = std::min<size_t>((size_t) pool.workers(), ht.worker_cores.size());
+ const size_t nw = std::min<size_t>((size_t) pool.workers(), wc.size());
std::fprintf(stderr, "%s\n",
- strata::aux_cpus::set_topology(ht.host_core, std::vector<int>(ht.worker_cores.begin(),
- ht.worker_cores.begin() + (long) nw)).c_str());
+ strata::aux_cpus::set_topology(ht.host_core, std::vector<int>(wc.begin(), wc.begin() + (long) nw)).c_str());
}
if (o.no_ple_prefetch) strata::kernels::ple_prefetch_enable(false);
// ---- R4's slot storage. Allocated AFTER the weights and the session, so `cudaMemGetInfo` inside `open`