Add backend support to LiteRT-LM end-to-end tests.

This change allows specifying the backend (e.g., cpu, gpu) for each test case in the JSON data, enabling testing across different hardware configurations. The `run_engine` fixture now accepts and uses the backend parameter.

LiteRT-LM-PiperOrigin-RevId: 881666398

Author: ai-edge-bot

tools/test/conftest.py

@@ -29,6 +29,7 @@
 
 def pytest_addoption(parser: pytest.Parser) -> None:
   """Adds custom command line arguments to pytest."""
+
   parser.addoption(
       "--model-path",
       action="store",
@@ -108,10 +109,10 @@ def run_engine(
     model_path: Path to the model file.
   """
 
-  def _run(prompt: str, timeout: int = 120) -> str:
+  def _run(backend: str, prompt: str, timeout: int = 120) -> str:
     cmd = [
         engine_binary,
-        "--backend=cpu",
+        f"--backend={backend}",
         f"--model_path={model_path}",
         f"--input_prompt={prompt}",
     ]

tools/test/test_data/test_e2e_sanity_checks.json

@@ -1,7 +1,14 @@
 {
   "test_model_sanity": [
     {
-      "id": "basic_question",
+      "id": "basic_question_cpu",
+      "backend": "cpu",
+      "prompt": "What's the tallest building in the world?",
+      "response": "Burj Khalifa"
+    },
+    {
+      "id": "basic_question_gpu",
+      "backend": "gpu",
       "prompt": "What's the tallest building in the world?",
       "response": "Burj Khalifa"
     }

tools/test/test_e2e_sanity_checks.py

@@ -47,11 +47,12 @@ def test_model_sanity(
       "response".
   """
 
+  backend = test_case["backend"]
   prompt = test_case["prompt"]
   expected_response = test_case["response"]
 
   # Execute
-  output = run_engine(prompt)
+  output = run_engine(backend, prompt)
   clean_output = output.replace("\n", " ")
 
   # Validate