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++
tests: setup no longer tips about the conversation cache (the engine sizes it, #3493)
Niko12215de8d2694959
diff --git a/tools/test_setup_bench_tips.py b/tools/test_setup_bench_tips.pyindex 141f377b9..f549ed112 100644--- a/tools/test_setup_bench_tips.py+++ b/tools/test_setup_bench_tips.py@@ -47,9 +47,10 @@ class BenchTips(unittest.TestCase):self.assertEqual([t for t in tips(res, {"STRATA_RESIDENT_HEADROOM_GIB": "8"}, ram=32) if "HEADROOM" in t], [])self.assertEqual([t for t in tips(BASE, ram=32) if "HEADROOM" in t], []) # not a resident config- def test_conversation_cache_for_agents(self):- self.assertTrue(any("--conversation-cache-mib 8192" in t for t in tips(ram=64, model=40)))- self.assertEqual([t for t in tips(ram=64, model=41) if "conversation" in t], []) # < 24 GB left+ def test_no_conversation_cache_tip_the_engine_sizes_it_itself(self):+ # 0.2.3 (#3493): the engine sizes the conversation cache from the free RAM when no value is given, so setup no longer tips about it+ for ram, model in ((64, 40), (64, 41), (128, 40)):+ self.assertEqual([t for t in tips(ram=ram, model=model) if "conversation" in t], [])self.assertEqual([t for t in tips([*BASE, "--conversation-cache-mib", "4096"], ram=128) if "conversation" in t], [])def test_vision_cpu_on_small_cards(self):