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++
Gyro: the MCP recommendation asks 4 GB of room beside the memory figure; tests for the Gyro catalog
Niko12215e33c4e85d7d
diff --git a/tools/strata_mcp.py b/tools/strata_mcp.pyindex dc37773c9..7612ef26c 100644--- a/tools/strata_mcp.py+++ b/tools/strata_mcp.py@@ -108,6 +108,7 @@ FALLBACK_FAMILIES = {"~80 GB of RAM to keep every expert in RAM","tag": "nvfp4-"},}+GYRO_ROOM_GB = 4.0 # recommend() suggests a Gyro size only when VRAM + RAM exceed its memory figure by this muchFALLBACK_CONTEXTS = [8192, 32768, 65536, 131072, 204800, 262144, 393216, 524288]BENCH_PROMPT = ("Write a short story (about 300 words) about a lighthouse keeper who finds a message in a bottle. ""Plain prose, no title.")@@ -707,10 +708,11 @@ class Strata:elif low_fits("IQ1_M"):fam, model, why = "coder", "IQ1_M", (f"{ram:.0f} GB of RAM is below the Coder's 32 GB, but the GPU's "f"{vram:.0f} GB make up for it (setup's low-RAM mode: slower)")- elif "Gyro-S" in models and ram + vram >= models["Gyro-S"]["ram_gb"]:- # Gyro (agentionai): only where the sizes above do not fit; the smaller memory total, the 27.6 GB of Gyro-S,- # is VRAM + RAM together. Gyro-M (35 GB) when this PC has that much, else Gyro-S.- model = "Gyro-M" if ram + vram >= models["Gyro-M"]["ram_gb"] else "Gyro-S"+ elif "Gyro-S" in models and ram + vram >= models["Gyro-S"]["ram_gb"] + GYRO_ROOM_GB:+ # Gyro (agentionai): only where the sizes above do not fit. Their memory figure (Gyro-S 27.6 GB, Gyro-M 35 GB)+ # is VRAM + RAM together and the system needs room beside it, so GYRO_ROOM_GB more is asked for here.+ # Gyro-M when this PC has that much, else Gyro-S.+ model = "Gyro-M" if ram + vram >= models["Gyro-M"]["ram_gb"] + GYRO_ROOM_GB else "Gyro-S"fam, why = "gyro", (f"{ram:.0f} GB of RAM is below the Coder's 32 GB, but RAM + VRAM = {ram + vram:.0f} GB "f"holds {model} ({models[model]['ram_gb']} GB of memory in total; the quality is lower ""than the 2-3-bit sizes, and it needs the whole download on a fast SSD)")diff --git a/tools/test_setup_gyro.py b/tools/test_setup_gyro.pynew file mode 100644index 000000000..e4bcf6278--- /dev/null+++ b/tools/test_setup_gyro.py@@ -0,0 +1,80 @@+"""The Gyro entries of setup's catalog (agentionai/Qwen3.8-Flash-Next-Gyro-GGUF): sizes, pins, the VRAM + RAM memory rule,+never a default, and the MCP server's recommendation naming them only where nothing else fits. Offline."""+import pathlib+import sys+import unittest++ROOT = pathlib.Path(__file__).resolve().parent.parent+sys.path.insert(0, str(ROOT))+sys.path.insert(0, str(ROOT / "tools"))+import setup # noqa: E402+import strata_mcp as M # noqa: E402+++def hardware(ram, vram):+ return {"ram_gb": ram, "gpus": [{"usable": True, "vram_gb": vram, "vendor": "nvidia", "index": 0, "name": "RTX"}],+ "cpu": {"avx2": True}}+++class GyroCatalog(unittest.TestCase):+ def test_the_two_sizes_and_their_numbers(self):+ self.assertEqual(setup.FAMILIES["gyro"]["file"].format(q="Gyro-S"), "Qwen3.8-Flash-Next-Gyro-S.gguf")+ for model, name, download, memory in (("Gyro-S", "Qwen3.8-Flash-Next-Gyro-S-TQ1_0.gguf", 58.5, 27.6),+ ("Gyro-M", "Qwen3.8-Flash-Next-Gyro-M-TQ2_0.gguf", 92.0, 35.0)):+ d = setup.MODELS[model]+ self.assertEqual(setup.model_file(setup.FAMILIES["gyro"], model, 1), name)+ self.assertEqual((d["download_gb"], d["ram_gb"]), (download, memory))+ size, sha = setup.GYRO_FILES[name]+ self.assertAlmostEqual(size / 1e9, download, delta=0.05)+ self.assertEqual(len(sha), 64)+ self.assertEqual(setup.model_shards(setup.FAMILIES["gyro"], model), 1)+ self.assertIs(d.get("vision"), False)+ self.assertTrue(d["mem_total"])+ self.assertEqual(d["families"], ("gyro",))++ def test_pinned_revision_and_license(self):+ repo = "agentionai/Qwen3.8-Flash-Next-Gyro-GGUF"+ self.assertEqual(len(setup.HF_REVISIONS[repo]), 40)+ self.assertIn(setup.HF_REVISIONS[repo], setup.FAMILIES["gyro"]["hf"])+ self.assertIn("Qwen Community License 1.0", setup.FAMILIES["gyro"]["license"])++ def test_never_the_default_family(self):+ self.assertEqual(setup.menu_families(False)[0], "qwen")+ self.assertNotIn("gyro", setup.RECOMMENDED_SIZES)++ def test_memory_is_vram_plus_ram(self):+ self.assertEqual(setup.mem_have("Gyro-S", 16.0, 12.0), 28.0)+ self.assertEqual(setup.mem_have("Q2_0", 16.0, 12.0), 16.0)+ names = ["Gyro-S", "Gyro-M"]+ self.assertEqual(setup.recommended_size("gyro", names, 24.0, 12.0), "Gyro-M") # 36 GB+ self.assertEqual(setup.recommended_size("gyro", names, 16.0, 12.0), "Gyro-S") # 28 GB+ self.assertEqual(setup.recommended_size("gyro", names, 8.0, 8.0), "Gyro-S") # nothing fits: the first++ def test_mcp_table_matches(self):+ for m in ("Gyro-S", "Gyro-M"):+ for k in ("download_gb", "ram_gb", "arena_gb", "mem_total", "vision"):+ self.assertEqual(setup.MODELS[m][k], M.FALLBACK_MODELS[m][k], (m, k))+ self.assertEqual(setup.FAMILIES["gyro"]["tag"], M.FALLBACK_FAMILIES["gyro"]["tag"])++ def test_mcp_recommends_gyro_only_where_nothing_else_fits(self):+ s = M.Strata(ROOT)+ for ram, vram, family in ((64.0, 12.0, "qwen"), (48.0, 12.0, "qwen"), (32.0, 12.0, "coder")):+ self.assertEqual(s.recommend(hardware(ram, vram))["family"], family, (ram, vram))+ got = s.recommend(hardware(20.0, 12.0)) # 32 GB in total, the Coder's 32 GB of RAM missing+ self.assertEqual((got["family"], got["model"]), ("gyro", "Gyro-S"))+ got = s.recommend(hardware(24.0, 16.0)) # 40 GB in total: the Coder fits in the low-RAM mode, so it wins+ self.assertEqual((got["family"], got["model"]), ("coder", "IQ1_M"))+ self.assertIsNone(s.recommend(hardware(12.0, 8.0))["family"])+ self.assertIsNone(s.recommend(hardware(16.0, 11.0))["family"]) # 27 GB: below Gyro-S's 27.6 + room++ def test_mcp_table_verdict_counts_vram(self):+ s = M.Strata(ROOT)+ fam = next(f for f in s.model_table(hardware(16.0, 16.0)) if f["family"] == "gyro")+ by = {x["model"]: x for x in fam["sizes"]}+ self.assertEqual(by["Gyro-S"]["on_this_pc"], "fits")+ self.assertEqual(by["Gyro-M"]["on_this_pc"], "tight")+ self.assertFalse(by["Gyro-S"]["images"])+++if __name__ == "__main__":+ unittest.main()