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 #3425: the recommendation says how much of the picked size would be read from the SSD when RAM + VRAM is only just the tier's floor

Niko122118b4affb87c8
diff --git a/tools/strata_mcp.py b/tools/strata_mcp.py
index eab649158..88964766c 100644
--- a/tools/strata_mcp.py
+++ b/tools/strata_mcp.py
@@ -1151,6 +1151,15 @@ class Strata:
"(the smallest model, the Coder, keeps ~23 GB of experts in "
"RAM)"}
ctx = 32768 if vram < 14 else 65536 if vram < 20 else 131072
+ # #3425: the tiers count RAM + VRAM together, but the desktop keeps about 10 GB of the RAM and the card keeps ~6 GB for
+ # its dense weights, so a PC at the floor of its tier reads part of the experts from the SSD at every answer. Said
+ # plainly beside the pick (the same arithmetic as the app's "This computer" card); the pick itself stays the owner's tier.
+ arena = float(models[model].get("arena_gb") or 0)
+ spill = arena - max(0.0, tvram - 6.0) - max(0.0, ram - 10.0)
+ if arena and spill >= 4.0:
+ notes.append(f"about {spill:.0f} GB of {model} would be read from the SSD while answering (this PC's RAM leaves about 10 GB "
+ "for the system and the card holds the rest): expect a few tokens per second on a SATA or USB drive, more "
+ "on a fast NVMe. A smaller size fits better; it is a recommendation, every size stays a choice")
if vram < 11:
notes.append("less than 12 GB of VRAM: it runs, but slowly (most experts stay on the CPU)")
if hw.get("cpu", {}).get("avx2") is False: # #623: a warning, not a stop (setup compiles for it)
diff --git a/tools/test_strata_mcp.py b/tools/test_strata_mcp.py
index df5be39bc..62230b8ef 100644
--- a/tools/test_strata_mcp.py
+++ b/tools/test_strata_mcp.py
@@ -663,6 +663,16 @@ class Helpers(unittest.TestCase):
r = s.recommend({"ram_gb": 64, "gpus": [{**gpu[0], "vram_gb": 12.0}], "cpu": {"avx2": True}})
self.assertEqual(r["context"], 32768)
+ def test_a_pick_that_reads_from_the_ssd_says_so(self):
+ # #3425: 32 GB RAM + 16 GB VRAM is the 48 GB tier; about 11 GB of IQ3_XXS would come from the SSD
+ s = M.Strata(HERE.parent)
+ gpu = [{"index": 0, "name": "x", "vram_gb": 15.9, "usable": True, "vendor": "nvidia"}]
+ r = s.recommend({"ram_gb": 31.9, "gpus": gpu, "cpu": {"avx2": True}})
+ self.assertEqual((r["family"], r["model"]), ("qwen", "IQ3_XXS"))
+ self.assertTrue(any("read from the SSD" in n for n in r["notes"]), r["notes"])
+ roomy = s.recommend({"ram_gb": 64, "gpus": gpu, "cpu": {"avx2": True}})
+ self.assertFalse(any("read from the SSD" in n for n in roomy["notes"]), roomy["notes"])
+
class PhysicalCores(unittest.TestCase):
"""#3417: the hardware report's cpu.cores is the physical cores; the Work profile takes 80% of it for --pool-workers."""