diff --git a/tests/inference/tests/integration/test_inference_transformers_real_limit.py b/tests/inference/tests/integration/test_inference_transformers_real_limit.py
new file mode 100644
index 0000000..96be139
--- /dev/null
+++ b/tests/inference/tests/integration/test_inference_transformers_real_limit.py
@@ -0,0 +1,97 @@
+import json
+from pathlib import Path
+import tempfile
+
+import pytest
+
+from llmsql.inference.inference_transformers import inference_transformers
+
+
+def _write_jsonl(data, path: Path):
+    with path.open("w", encoding="utf-8") as f:
+        for row in data:
+            f.write(json.dumps(row) + "\n")
+
+
+@pytest.mark.integration
+def test_real_inference_limit():
+    """
+    Real integration test:
+    - real transformers backend
+    - real HF model (Pythia-14m)
+    - real inference pipeline
+    - verifies limit argument
+    - verifies output format
+    """
+
+    LIMIT = 50
+    MODEL_NAME = "EleutherAI/pythia-14m"
+    TOKENIER_NAME = "EleutherAI/pythia-14m"
+
+    questions = [
+        {
+            "question_id": f"q{i}",
+            "table_id": "t1",
+            "question": f"What is row {i}?",
+        }
+        for i in range(100)
+    ]
+
+    tables = [
+        {
+            "table_id": "t1",
+            "header": ["id", "name"],
+            "types": ["int", "text"],
+            "rows": [[1, "Alice"], [2, "Bob"]],
+        }
+    ]
+
+    with tempfile.TemporaryDirectory() as tmpdir:
+        tmpdir_path = Path(tmpdir)
+
+        questions_file = tmpdir_path / "questions.jsonl"
+        tables_file = tmpdir_path / "tables.jsonl"
+        output_file = tmpdir_path / "outputs.jsonl"
+
+        _write_jsonl(questions, questions_file)
+        _write_jsonl(tables, tables_file)
+
+        try:
+            results = inference_transformers(
+                model_or_model_name_or_path=MODEL_NAME,
+                tokenizer_or_name=TOKENIER_NAME,
+                output_file=str(output_file),
+                workdir_path=str(tmpdir_path),
+                limit=LIMIT,
+                batch_size=4,
+                max_new_tokens=16,
+                temperature=0.0,
+                do_sample=False,
+            )
+        except (OSError, RuntimeError) as exc:
+            pytest.skip(f"Unable to load/run model in CI: {exc}")
+
+        # Verify limit handling
+        assert len(results) == 50
+
+        expected_ids = [f"q{i}" for i in range(50)]
+        actual_ids = [r["question_id"] for r in results]
+
+        assert actual_ids == expected_ids
+
+        # Verify output file was written
+        assert output_file.exists()
+
+        output_lines = output_file.read_text(encoding="utf-8").splitlines()
+
+        assert len(output_lines) == 50
+
+        # Verify output schema
+        for line in output_lines:
+            row = json.loads(line)
+
+            assert "question_id" in row
+            assert "completion" in row
+
+            assert isinstance(row["question_id"], str)
+            assert isinstance(row["completion"], str)
\ No newline at end of file