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++
setup: a NVIDIA driver older than CUDA 13 needs (528 to 579) runs the CUDA 12 engine by itself and says so in one line instead of stopping and naming a flag strata launch does not have (#3843); tests
Niko12211736b558bd04
diff --git a/setup.py b/setup.pyindex 83a09dcfe..47e3fa48f 100644--- a/setup.py+++ b/setup.py@@ -1580,7 +1580,7 @@ def named_gpus(gpu, gpus) -> list:return []-def cuda_choice(archs, cuda=None):+def cuda_choice(archs, cuda=None, driver=None):"""The CUDA toolkit of one model's engine: (12 or 13, why). 13 (the ready-made engine) unless a card is olderthan CUDA 13 supports (Pascal / Volta: CUDA 13 cannot compile for them) - one engine runs per model, so its oldestcard decides. `cuda` (--cuda 12|13) overrides it; setup recommends, it does not refuse (the caller warns)."""@@ -1595,6 +1595,9 @@ def cuda_choice(archs, cuda=None):if old:return 12, (f"sm_{old[0]} is older than CUDA 13 supports (it dropped Pascal and Volta): this model runs the ""experimental CUDA 12 engine")+ if driver and CUDA12_MIN_DRIVER <= int(driver) < MIN_DRIVER: # #3843: a driver too old for CUDA 13 still runs the CUDA 12 engine+ return 12, (f"the NVIDIA driver {driver} is older than CUDA 13 needs ({MIN_DRIVER}): this model runs the CUDA 12 engine "+ f"(it works with driver {CUDA12_MIN_DRIVER} or newer; a newer driver from https://www.nvidia.com/drivers gives the CUDA 13 engine)")return 13, None@@ -8801,15 +8804,13 @@ def main() -> int:if multi:ok("GPUs: " + " + ".join(gpu_name(x) for x in chosen) + " together (the model's layers are split across them)")ok(f"GPU: {gpu['name']}, {gpu['vram_gb']:.1f} GB VRAM, compute capability {cc(gpu)}, driver {gpu['driver']}")- cuda_tk, why = cuda_choice(gpu["archs"], a.cuda) # one engine per model: its oldest card decides+ cuda_tk, why = cuda_choice(gpu["archs"], a.cuda, driver_major(gpu)) # one engine per model: its oldest card decidesif why:(warn if cuda_tk == 13 or str(a.cuda) == "12" else ok)(f"CUDA {cuda_tk}: {why}")min_driver = CUDA12_MIN_DRIVER if cuda_tk == 12 else MIN_DRIVERif driver_major(gpu) < min_driver:fail(f"the NVIDIA driver is too old ({gpu['driver']}; {min_driver} or newer is needed)",- "update it with the NVIDIA App or from https://www.nvidia.com/drivers, restart, and run this again" +- ("" if cuda_tk == 12 else f" (or --cuda 12: the experimental CUDA 12 engine runs with driver "- f"{CUDA12_MIN_DRIVER} or newer, docs/OLDER_GPUS.md)"))+ "update it with the NVIDIA App or from https://www.nvidia.com/drivers, restart, and run this again")if gpu["vram_gb"] < 11:warn("less than 12 GB of " + ("GPU memory (this APU's carve-out + shared memory)" if gpu.get("uma") else "VRAM")+ ": Strata will run, but most experts stay on the CPU and it will be slow")diff --git a/tools/test_setup_older_gpus.py b/tools/test_setup_older_gpus.pyindex 72d9a97cd..d11419e15 100644--- a/tools/test_setup_older_gpus.py+++ b/tools/test_setup_older_gpus.py@@ -39,6 +39,14 @@ class Choice(unittest.TestCase):self.assertEqual(tk, 12, archs)self.assertIn(f"sm_{min(archs)} is older than CUDA 13 supports", why)+ def test_a_driver_too_old_for_cuda13_gets_the_cuda12_engine(self): # #3843+ self.assertEqual(setup.cuda_choice([86], None, 576)[0], 12)+ self.assertIn("older than CUDA 13 needs", setup.cuda_choice([86], None, 576)[1])+ self.assertEqual(setup.cuda_choice([86], None, 580), (13, None))+ self.assertEqual(setup.cuda_choice([86], None, 0), (13, None)) # driver unknown: as before+ self.assertEqual(setup.cuda_choice([86], None, 500), (13, None)) # too old for both: setup stops with the driver message+ self.assertEqual(setup.cuda_choice([86], "13", 576)[0], 13) # chosen by hand: kept+def test_the_override_is_honored_with_a_warning(self):tk, why = setup.cuda_choice([70], "13")self.assertEqual(tk, 13) # recommend, never force: kept, and said@@ -225,12 +233,17 @@ class Install(unittest.TestCase):self.assertEqual((calls, cfg["cuda"]), ([12], 12))self.assertIn("--cuda 12 (as you chose", out)- def test_an_old_driver_is_told_about_cuda12(self):- code, out, _, _, _ = self.run_setup(127.8, [card(0, "NVIDIA GeForce RTX 4090", 24.0, "89", "550.54")],+ def test_an_old_driver_gets_cuda12_by_itself(self): # #3843: said in one line, not a stop+ code, out, cfg, calls, _ = self.run_setup(127.8, [card(0, "NVIDIA GeForce RTX 4090", 24.0, "89", "550.54")],+ argv_for("qwen", "IQ3_XXS"))+ self.assertEqual(code, 0, out[-3000:])+ self.assertEqual((calls, cfg["cuda"]), ([12], 12))+ self.assertIn("older than CUDA 13 needs (580): this model runs the CUDA 12 engine", out)+ code, out, _, _, _ = self.run_setup(127.8, [card(0, "NVIDIA GeForce RTX 4090", 24.0, "89", "470.10")],argv_for("qwen", "IQ3_XXS"))- self.assertEqual(code, 1)- self.assertIn("580 or newer is needed", out)- self.assertIn("--cuda 12", out)+ self.assertEqual(code, 1) # too old for both: the stop stays+ self.assertIn("is too old", out)+ self.assertNotIn("--cuda 12", out)code, out, cfg, calls, _ = self.run_setup(127.8, [card(0, "NVIDIA GeForce RTX 4090", 24.0, "89", "550.54")],argv_for("qwen", "IQ3_XXS") + ["--cuda", "12"])self.assertEqual(code, 0, out[-3000:])