Update test_gpu_examples.py

Added a GPU test with a 4-bit model in test_gpu_examples.py using @pytest.mark.single_gpu_tests and @require_bitsandbytes. It verifies that find_kappa_target_modules runs successfully and returns target modules.

Author: oswaldoludwig

tests/test_gpu_examples.py

@@ -6652,3 +6652,35 @@ def test_load_adapter_save(self, tmp_path):
         tmp_dir_reference = tmp_path / "reference"
         tmp_dir_tp = tmp_path / "tp"
         self._spawn(_test_load_adapter_save, tmp_dir_reference, tmp_dir_tp, port_offset=6)
+
+@pytest.mark.single_gpu_tests
+@require_bitsandbytes
+def test_kappatune_with_4bit_model():
+    """Test that KappaTune works with 4-bit quantized models on GPU."""
+    from peft.helpers import find_kappa_target_modules
+    from transformers import AutoModelForCausalLM, BitsAndBytesConfig
+    import torch
+
+    # Use a very small model for faster testing
+    quantization_config = BitsAndBytesConfig(
+        load_in_4bit=True,
+        bnb_4bit_compute_dtype=torch.float16,
+        bnb_4bit_quant_type="nf4",
+        bnb_4bit_use_double_quant=True,
+    )
+
+    model = AutoModelForCausalLM.from_pretrained(
+        "hf-internal-testing/tiny-random-LlamaForCausalLM",
+        quantization_config=quantization_config,
+        device_map="cuda",
+        torch_dtype=torch.float16,
+    )
+
+    # Run KappaTune
+    targets = find_kappa_target_modules(model, top_p=0.3)
+
+    # Basic assertions
+    assert isinstance(targets, dict)
+    assert "target_modules" in targets
+    assert isinstance(targets["target_modules"], list)
+    assert len(targets["target_modules"]) > 0, "Should return at least some target modules"