From feda78fbbbda6e868580ad2063d4b254b5c2250e Mon Sep 17 00:00:00 2001
From: Joseph Magly <1159087+jmagly@users.noreply.github.com>
Date: Sat, 22 Aug 2026 10:09:41 -0400
Subject: [PATCH] test(loader): isolate optional quantization dependency

---
 tests/test_loader_boundaries.py | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/tests/test_loader_boundaries.py b/tests/test_loader_boundaries.py
index 0d60b43..06b0e3f 100644
--- a/tests/test_loader_boundaries.py
+++ b/tests/test_loader_boundaries.py
@@ -651,6 +651,7 @@ def test_cuda_quantization_builds_explicit_config(loader_boundary, monkeypatch,
 
     monkeypatch.setattr(loader.dev, "get_device", lambda _preference="auto": "cuda")
     monkeypatch.setattr(loader.dev, "supports_bitsandbytes", lambda _device=None: True)
+    monkeypatch.setitem(sys.modules, "bitsandbytes", ModuleType("bitsandbytes"))
     monkeypatch.setattr("transformers.BitsAndBytesConfig", fake_bnb)
     handle = loader.load_model("x", quantization=quantization, skip_snapshot=True)
     kwargs = loader_boundary.model_class.from_pretrained.call_args.kwargs
@@ -743,6 +744,7 @@ def _enable_cuda_quantization(monkeypatch, estimate_gb: float) -> None:
         "get_device_properties",
         lambda _index: SimpleNamespace(total_memory=16 * gib),
     )
+    monkeypatch.setitem(sys.modules, "bitsandbytes", ModuleType("bitsandbytes"))
     monkeypatch.setattr("transformers.BitsAndBytesConfig", lambda **kwargs: kwargs)