diff --git a/README.md b/README.md
index d4b1b80..a074aee 100644
--- a/README.md
+++ b/README.md
@@ -129,6 +129,15 @@ python kernel-coder/nano_r1_script.py --nproc 8  --max_response_tokens 2048
 
 ```
 
+### 🧪 Running tests
+
+Install `pytest` and run the test suite from the project root:
+
+```bash
+pip install pytest
+pytest
+```
+
 
 ## 🙏 Acknowledgements
 
diff --git a/tests/test_utils.py b/tests/test_utils.py
new file mode 100644
index 0000000..c4bfdf4
--- /dev/null
+++ b/tests/test_utils.py
@@ -0,0 +1,56 @@
+import sys
+import types
+import importlib.util
+from pathlib import Path
+import pytest
+
+torch = pytest.importorskip("torch")
+
+# Provide dummy modules for optional dependencies
+_dummy_modules = {
+    "wandb": {},
+    "datasets": {"Dataset": object},
+    "deepspeed": {"DeepSpeedEngine": object},
+    "transformers": {"AutoTokenizer": object, "PreTrainedModel": object},
+    "vllm": {"LLM": object, "SamplingParams": object},
+}
+for name, attrs in _dummy_modules.items():
+    if name not in sys.modules:
+        module = types.ModuleType(name)
+        for attr, val in attrs.items():
+            setattr(module, attr, val)
+        sys.modules[name] = module
+
+# Dynamically load utils.py since the folder name contains a dash
+utils_path = Path(__file__).resolve().parents[1] / "kernel-coder" / "utils.py"
+spec = importlib.util.spec_from_file_location("kernel_utils", utils_path)
+utils = importlib.util.module_from_spec(spec)
+spec.loader.exec_module(utils)
+
+prepare_model_inputs = utils.prepare_model_inputs
+
+def test_prepare_model_inputs_basic():
+    query_token_ids = [[1, 2], [3]]
+    response_token_ids = [[4, 5, 6], [7, 8]]
+    advantages = [[0.1, 0.2, 0.3], [0.4, 0.5]]
+    device = torch.device("cpu")
+
+    outputs = prepare_model_inputs(query_token_ids, response_token_ids, advantages, device)
+
+    for key in ["input_ids", "attention_mask", "labels", "advantages", "labels_mask"]:
+        assert key in outputs
+        assert outputs[key].shape == (2, 5)
+
+    # First sequence has no padding
+    assert torch.equal(outputs["input_ids"][0], torch.tensor([1, 2, 4, 5, 6]))
+    assert torch.equal(outputs["attention_mask"][0], torch.tensor([1, 1, 1, 1, 1]))
+    assert torch.equal(outputs["labels"][0], torch.tensor([-100, -100, 4, 5, 6]))
+    assert torch.allclose(outputs["advantages"][0], torch.tensor([0.0, 0.0, 0.1, 0.2, 0.3]))
+    assert torch.equal(outputs["labels_mask"][0], torch.tensor([0, 0, 1, 1, 1]))
+
+    # Second sequence padded to length 5
+    assert torch.equal(outputs["input_ids"][1], torch.tensor([3, 7, 8, 0, 0]))
+    assert torch.equal(outputs["attention_mask"][1], torch.tensor([1, 1, 1, 0, 0]))
+    assert torch.equal(outputs["labels"][1], torch.tensor([-100, 7, 8, -100, -100]))
+    assert torch.allclose(outputs["advantages"][1], torch.tensor([0.0, 0.4, 0.5, 0.0, 0.0]))
+    assert torch.equal(outputs["labels_mask"][1], torch.tensor([0, 1, 1, 0, 0]))