tests/integration/defs/examples/serve/test_configs/Qwen3-30B-A3B-FP8.yml (1)

21-21: Add newline at EOF.

Minor formatting nit for consistency with repo style.

-print_iter_log: true
+print_iter_log: true
+

tests/integration/defs/conftest.py (4)

77-82: Docstring is inaccurate and missing closing punctuation; clarify default behavior.

Function now defaults to /scratch.trt_llm_data/llm-models and no longer asserts validity. Update the docstring and end the first line with a period.

-def llm_models_root() -> str:
-    """return LLM_MODELS_ROOT path if it is set in env, assert when it's set but not a valid path"""
+def llm_models_root() -> str:
+    """Return the LLM models root path.
+
+    If LLM_MODELS_ROOT is unset, defaults to '/scratch.trt_llm_data/llm-models'.
+    This function does not validate path existence.
+    """

---

149-151: Avoid assert False; raise an explicit exception.

Using assert False can be stripped with -O and obscures intent.

-                    assert (
-                        False
-                    ), f"{cached_dir} does not exist, and fail_if_path_is_invalid is True, please check the cache directory"
+                    raise AssertionError(
+                        f"{cached_dir} does not exist, and fail_if_path_is_invalid is True, please check the cache directory"
+                    )

---

2001-2009: Log fallback path for GPU memory detection.

When nvidia-smi and /proc/meminfo parsing fail, consider logging a warning before defaulting to 8192 MiB to aid debugging in CI.

-            except:
-                memory = 8192  # Default 8GB if all else fails
+            except Exception:
+                print_warning("Falling back to default GPU memory=8192 MiB; nvidia-smi and /proc/meminfo unavailable.")
+                memory = 8192  # Default 8GB if all else fails

---

2389-2395: Docstring punctuation and wording.

Add a period and fix wording.

-@pytest.fixture(scope="function")
-def serve_test_root(llm_root):
-    "Get servetest root"
+@pytest.fixture(scope="function")
+def serve_test_root(llm_root):
+    "Get serve test root."

tests/integration/defs/examples/serve/test_serve.py (3)

95-122: Consider using a free ephemeral port to avoid flakiness.

Hard-coding 8000 risks collisions in CI. Optional: pick a free port at runtime and pass it to both the server and checks.

+    # Optionally pick a free port to avoid collisions
+    # import socket
+    # with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+    #     s.bind(("127.0.0.1", 0))
+    #     http_port = s.getsockname()[1]
+    http_port = "8000"
     cmd = [
         "trtllm-serve",
         "serve",
         model_path,
         "--host",
         "0.0.0.0",
         "--port",
-        "8000",
+        str(http_port),
         "--backend",
         "pytorch",
         "--extra_llm_api_options",
         config_file,
     ]
@@
-        check_server_ready()
+        check_server_ready(http_port=str(http_port))

---

126-131: Use os.path.basename to derive model id; safer across platforms.

Path splitting on “/” breaks on Windows.

-        model_name = model_path.split('/')[-1]  # "Qwen3-30B-A3B-FP8"
+        model_name = os.path.basename(model_path)  # "Qwen3-30B-A3B-FP8"

---

54-66: Optional: verify accepted model id before chatting.

Listing /v1/models and asserting the chosen model exists would make the test more robust.

-        # Make a chat completion request
+        # Optional: verify model is available
+        # models = client.models.list()
+        # assert any(m.id == model_name for m in models.data), f"Model {model_name} not found in server."
+        # Make a chat completion request

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

💡 Knowledge Base configuration:

• MCP integration is disabled by default for public repositories
• Jira integration is disabled by default for public repositories
• Linear integration is disabled by default for public repositories

You can enable these sources in your CodeRabbit configuration.

Learnt from: moraxu
PR: NVIDIA/TensorRT-LLM#6303
File: tests/integration/test_lists/qa/examples_test_list.txt:494-494
Timestamp: 2025-07-28T17:06:08.621Z
Learning: In TensorRT-LLM testing, it's common to have both CLI flow tests (test_cli_flow.py) and PyTorch API tests (test_llm_api_pytorch.py) for the same model. These serve different purposes: CLI flow tests validate the traditional command-line workflow, while PyTorch API tests validate the newer LLM API backend. Both are legitimate and should coexist.