ci: remove not slow filter (#2944)

Signed-off-by: alec-flowers <aflowers@nvidia.com>

.github/workflows/container-validation-backends.yml

@@ -22,7 +22,7 @@ jobs:
             pytest_marks: "e2e and vllm and gpu_1 and not slow"
           - framework: sglang
             target: runtime
-            pytest_marks: "e2e and sglang and gpu_1"
+            pytest_marks: "e2e and sglang and gpu_1 and not slow"
 
     # Do not cancel main branch runs
     concurrency:

pyproject.toml

@@ -178,7 +178,8 @@ markers = [
     "sglang: marks tests as requiring sglang",
     "slow: marks tests as known to be slow",
     "h100: marks tests to run on H100",
-    "kvbm: marks tests for KV behavior and model determinism"
+    "kvbm: marks tests for KV behavior and model determinism",
+    "model: model id used by a test or parameter"
 ]
 
 # Linting/formatting

tests/conftest.py

@@ -20,26 +20,25 @@ import tempfile
 
 import pytest
 
+from tests.utils.constants import TEST_MODELS
 from tests.utils.managed_process import ManagedProcess
 
-# Custom format inspired by your example
+
+def pytest_configure(config):
+    # Defining model morker to avoid `'model' not found in `markers` configuration option`
+    # error when pyproject.toml is not available in the container
+    config.addinivalue_line("markers", "model: model id used by a test or parameter")
+
+
 LOG_FORMAT = "[TEST] %(asctime)s %(levelname)s %(name)s: %(message)s"
 DATE_FORMAT = "%Y-%m-%dT%H:%M:%S"
 
-# Configure logging
 logging.basicConfig(
     level=logging.INFO,
     format=LOG_FORMAT,
     datefmt=DATE_FORMAT,  # ISO 8601 UTC format
 )
 
-# List of models used in tests
-TEST_MODELS = [
-    "Qwen/Qwen3-0.6B",
-    "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
-    "llava-hf/llava-1.5-7b-hf",
-]
-
 
 def download_models(model_list=None, ignore_weights=False):
     """Download models - can be called directly or via fixture
@@ -107,15 +106,33 @@ def download_models(model_list=None, ignore_weights=False):
 
 
 @pytest.fixture(scope="session")
-def predownload_models():
-    """Fixture wrapper around download_models for all TEST_MODELS"""
+def predownload_models(pytestconfig):
+    """Fixture wrapper around download_models for models used in collected tests"""
+    # Get models from pytest config if available, otherwise fall back to TEST_MODELS
+    models = getattr(pytestconfig, "models_to_download", None)
+    if models:
+        logging.info(
+            f"Downloading {len(models)} models needed for collected tests\nModels: {models}"
+        )
+        download_models(model_list=list(models))
+    else:
+        # Fallback to original behavior if extraction failed
         download_models()
     yield
 
 
 @pytest.fixture(scope="session")
-def predownload_tokenizers():
-    """Fixture wrapper around download_models for all TEST_MODELS"""
+def predownload_tokenizers(pytestconfig):
+    """Fixture wrapper around download_models for tokenizers used in collected tests"""
+    # Get models from pytest config if available, otherwise fall back to TEST_MODELS
+    models = getattr(pytestconfig, "models_to_download", None)
+    if models:
+        logging.info(
+            f"Downloading tokenizers for {len(models)} models needed for collected tests\nModels: {models}"
+        )
+        download_models(model_list=list(models), ignore_weights=True)
+    else:
+        # Fallback to original behavior if extraction failed
         download_models(ignore_weights=True)
     yield
 
@@ -135,42 +152,26 @@ def logger(request):
     logger.removeHandler(handler)
 
 
+@pytest.hookimpl(trylast=True)
 def pytest_collection_modifyitems(config, items):
     """
     This function is called to modify the list of tests to run.
-    It is used to skip tests that are not supported on all environments.
     """
-
-    # Tests marked with trtllm requires specific environment with tensorrtllm
-    # installed. Hence, we skip them if the user did not explicitly ask for them.
-    if config.getoption("-m") and "trtllm_marker" in config.getoption("-m"):
-        return
-    skip_trtllm = pytest.mark.skip(reason="need -m trtllm_marker to run")
+    # Collect models via explicit pytest mark from final filtered items only
+    models_to_download = set()
     for item in items:
-        if "trtllm_marker" in item.keywords:
-            item.add_marker(skip_trtllm)
-
-        # Auto-inject predownload_models fixture for serve tests only (not router tests)
-        # Skip items that don't have fixturenames (like MypyFileItem)
-        if hasattr(item, "fixturenames"):
-            # Guard clause: skip if already has the fixtures
-            if (
-                "predownload_models" in item.fixturenames
-                or "predownload_tokenizers" in item.fixturenames
+        # Only collect from items that are not skipped
+        if any(
+            getattr(m, "name", "") == "skip" for m in getattr(item, "own_markers", [])
         ):
             continue
+        model_mark = item.get_closest_marker("model")
+        if model_mark and model_mark.args:
+            models_to_download.add(model_mark.args[0])
 
-            # Guard clause: skip if marked with skip_model_download
-            if item.get_closest_marker("skip_model_download"):
-                continue
-
-            # Add appropriate fixture based on test path
-            if "serve" in str(item.path):
-                item.fixturenames = list(item.fixturenames)
-                item.fixturenames.append("predownload_models")
-            elif "router" in str(item.path):
-                item.fixturenames = list(item.fixturenames)
-                item.fixturenames.append("predownload_tokenizers")
+    # Store models to download in pytest config for fixtures to access
+    if models_to_download:
+        config.models_to_download = models_to_download
 
 
 class EtcdServer(ManagedProcess):

tests/fault_tolerance/test_request_cancellation.py

@@ -9,8 +9,8 @@ import time
 
 import pytest
 import requests
-from huggingface_hub import snapshot_download
 
+from tests.utils.constants import FAULT_TOLERANCE_MODEL_NAME
 from tests.utils.engine_process import FRONTEND_PORT
 from tests.utils.managed_process import ManagedProcess
 from tests.utils.payloads import check_health_generate, check_models_api
@@ -56,7 +56,7 @@ class DynamoWorkerProcess(ManagedProcess):
             "-m",
             "dynamo.vllm",
             "--model",
-            "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
+            FAULT_TOLERANCE_MODEL_NAME,
             "--enforce-eager",
             "--gpu-memory-utilization",
             "0.45",
@@ -137,47 +137,12 @@ class DynamoWorkerProcess(ManagedProcess):
         return False
 
 
-def download_model() -> None:
-    """
-    Download the DeepSeek-R1-Distill-Llama-8B model from HuggingFace Hub if not already cached.
-    """
-    model_id = "deepseek-ai/DeepSeek-R1-Distill-Llama-8B"
-    logger.info(f"Caching model {model_id}...")
-
-    max_retries = 5
-    retry_delay = 30  # seconds
-
-    for attempt in range(max_retries):
-        try:
-            # Download the model to the default cache directory
-            # This will skip download if the model is already cached
-            snapshot_download(
-                repo_id="deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
-                repo_type="model",
-                local_files_only=False,
-            )
-            logger.info(f"Model {model_id} is ready for use")
-            return  # Success, exit the function
-        except Exception as e:
-            if attempt < max_retries - 1:  # Not the last attempt
-                logger.warning(
-                    f"Failed to download model {model_id} (attempt {attempt + 1}/{max_retries}): {e}"
-                )
-                logger.info(f"Retrying in {retry_delay} seconds...")
-                time.sleep(retry_delay)
-            else:  # Last attempt failed
-                logger.error(
-                    f"Failed to download model {model_id} after {max_retries} attempts: {e}"
-                )
-                raise
-
-
 def send_completion_request(
     prompt: str, max_tokens: int, timeout: int = 120
 ) -> requests.Response:
     """Send a completion request to the frontend"""
     payload = {
-        "model": "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
+        "model": FAULT_TOLERANCE_MODEL_NAME,
         "prompt": prompt,
         "max_tokens": max_tokens,
     }
@@ -211,7 +176,7 @@ def send_chat_completion_request(
 ) -> requests.Response:
     """Send a chat completion request to the frontend"""
     payload = {
-        "model": "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
+        "model": FAULT_TOLERANCE_MODEL_NAME,
         "messages": [{"role": "user", "content": prompt}],
         "max_tokens": max_tokens,
         "stream": stream,
@@ -383,8 +348,8 @@ def verify_request_cancelled(
 @pytest.mark.vllm
 @pytest.mark.gpu_1
 @pytest.mark.e2e
-@pytest.mark.slow
-def test_request_cancellation_vllm(request, runtime_services):
+@pytest.mark.model(FAULT_TOLERANCE_MODEL_NAME)
+def test_request_cancellation_vllm(request, runtime_services, predownload_models):
     """
     End-to-end test for request cancellation functionality.
 
@@ -395,8 +360,6 @@ def test_request_cancellation_vllm(request, runtime_services):
     2. Chat completion request (non-streaming)
     3. Chat completion request (streaming)
     """
-    # Step 0: Download the model from HuggingFace if not already cached
-    download_model()
 
     # Step 1: Start the frontend
     with DynamoFrontendProcess(request) as frontend:
@@ -446,8 +409,10 @@ def test_request_cancellation_vllm(request, runtime_services):
 @pytest.mark.vllm
 @pytest.mark.gpu_1
 @pytest.mark.e2e
-@pytest.mark.slow
-def test_request_cancellation_vllm_decode(request, runtime_services):
+@pytest.mark.model(FAULT_TOLERANCE_MODEL_NAME)
+def test_request_cancellation_vllm_decode(
+    request, runtime_services, predownload_models
+):
     """
     End-to-end test for request cancellation functionality with remote prefill.
 
@@ -455,8 +420,6 @@ def test_request_cancellation_vllm_decode(request, runtime_services):
     the system properly handles the cancellation and cleans up resources
     on the decode worker side in a disaggregated setup.
     """
-    # Step 0: Download the model from HuggingFace if not already cached
-    download_model()
 
     # Step 1: Start the frontend
     with DynamoFrontendProcess(request) as frontend:
@@ -501,7 +464,6 @@ def test_request_cancellation_vllm_decode(request, runtime_services):
 @pytest.mark.vllm
 @pytest.mark.gpu_1
 @pytest.mark.e2e
-@pytest.mark.slow
 def test_request_cancellation_vllm_prefill(request, runtime_services):
     """
     End-to-end test for request cancellation on remote prefill.

tests/fault_tolerance/test_request_migration.py

@@ -10,8 +10,8 @@ import time
 
 import pytest
 import requests
-from huggingface_hub import snapshot_download
 
+from tests.utils.constants import FAULT_TOLERANCE_MODEL_NAME
 from tests.utils.engine_process import FRONTEND_PORT
 from tests.utils.managed_process import ManagedProcess, terminate_process_tree
 from tests.utils.payloads import check_models_api
@@ -54,7 +54,7 @@ class DynamoWorkerProcess(ManagedProcess):
             "-m",
             "dynamo.vllm",
             "--model",
-            "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
+            FAULT_TOLERANCE_MODEL_NAME,
             "--enforce-eager",
             "--gpu-memory-utilization",
             "0.45",
@@ -117,47 +117,12 @@ class DynamoWorkerProcess(ManagedProcess):
         return False
 
 
-def download_model() -> None:
-    """
-    Download the DeepSeek-R1-Distill-Llama-8B model from HuggingFace Hub if not already cached.
-    """
-    model_id = "deepseek-ai/DeepSeek-R1-Distill-Llama-8B"
-    logger.info(f"Caching model {model_id}...")
-
-    max_retries = 5
-    retry_delay = 30  # seconds
-
-    for attempt in range(max_retries):
-        try:
-            # Download the model to the default cache directory
-            # This will skip download if the model is already cached
-            snapshot_download(
-                repo_id="deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
-                repo_type="model",
-                local_files_only=False,
-            )
-            logger.info(f"Model {model_id} is ready for use")
-            return  # Success, exit the function
-        except Exception as e:
-            if attempt < max_retries - 1:  # Not the last attempt
-                logger.warning(
-                    f"Failed to download model {model_id} (attempt {attempt + 1}/{max_retries}): {e}"
-                )
-                logger.info(f"Retrying in {retry_delay} seconds...")
-                time.sleep(retry_delay)
-            else:  # Last attempt failed
-                logger.error(
-                    f"Failed to download model {model_id} after {max_retries} attempts: {e}"
-                )
-                raise
-
-
 def send_completion_request(
     prompt: str, max_tokens: int, timeout: int = 120
 ) -> requests.Response:
     """Send a completion request to the frontend"""
     payload = {
-        "model": "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
+        "model": FAULT_TOLERANCE_MODEL_NAME,
         "prompt": prompt,
         "max_tokens": max_tokens,
     }
@@ -324,8 +289,8 @@ def verify_migration_occurred(frontend_process: DynamoFrontendProcess) -> None:
 @pytest.mark.vllm
 @pytest.mark.gpu_1
 @pytest.mark.e2e
-@pytest.mark.slow
-def test_request_migration_vllm(request, runtime_services):
+@pytest.mark.model(FAULT_TOLERANCE_MODEL_NAME)
+def test_request_migration_vllm(request, runtime_services, predownload_models):
     """
     End-to-end test for worker fault tolerance with migration support.
 
@@ -333,8 +298,6 @@ def test_request_migration_vllm(request, runtime_services):
     the system can handle the failure gracefully and migrate the request to
     another worker.
     """
-    # Step 0: Download the model from HuggingFace if not already cached
-    download_model()
 
     # Step 1: Start the frontend
     with DynamoFrontendProcess(request) as frontend:

tests/fault_tolerance/test_vllm_health_check.py

@@ -8,8 +8,8 @@ import time
 
 import pytest
 import requests
-from huggingface_hub import snapshot_download
 
+from tests.utils.constants import FAULT_TOLERANCE_MODEL_NAME
 from tests.utils.engine_process import FRONTEND_PORT
 from tests.utils.managed_process import ManagedProcess
 from tests.utils.payloads import check_models_api, completions_response_handler
@@ -56,10 +56,8 @@ class DynamoWorkerProcess(ManagedProcess):
             "-m",
             "dynamo.vllm",
             "--model",
-            "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
+            FAULT_TOLERANCE_MODEL_NAME,
             "--enforce-eager",
-            "--gpu-memory-utilization",
-            "0.45",
             "--max-model-len",
             "8192",
             "--migration-limit",
@@ -123,47 +121,12 @@ class DynamoWorkerProcess(ManagedProcess):
         return False
 
 
-def download_model() -> None:
-    """
-    Download the DeepSeek-R1-Distill-Llama-8B model from HuggingFace Hub if not already cached.
-    """
-    model_id = "deepseek-ai/DeepSeek-R1-Distill-Llama-8B"
-    logger.info(f"Caching model {model_id}...")
-
-    max_retries = 5
-    retry_delay = 30  # seconds
-
-    for attempt in range(max_retries):
-        try:
-            # Download the model to the default cache directory
-            # This will skip download if the model is already cached
-            snapshot_download(
-                repo_id="deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
-                repo_type="model",
-                local_files_only=False,
-            )
-            logger.info(f"Model {model_id} is ready for use")
-            return  # Success, exit the function
-        except Exception as e:
-            if attempt < max_retries - 1:  # Not the last attempt
-                logger.warning(
-                    f"Failed to download model {model_id} (attempt {attempt + 1}/{max_retries}): {e}"
-                )
-                logger.info(f"Retrying in {retry_delay} seconds...")
-                time.sleep(retry_delay)
-            else:  # Last attempt failed
-                logger.error(
-                    f"Failed to download model {model_id} after {max_retries} attempts: {e}"
-                )
-                raise
-
-
 def send_completion_request(
     prompt: str, max_tokens: int, timeout: int = 120
 ) -> requests.Response:
     """Send a completion request to the frontend"""
     payload = {
-        "model": "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
+        "model": FAULT_TOLERANCE_MODEL_NAME,
         "prompt": prompt,
         "max_tokens": max_tokens,
     }
@@ -194,7 +157,7 @@ def send_completion_request(
 @pytest.mark.vllm
 @pytest.mark.gpu_1
 @pytest.mark.e2e
-@pytest.mark.slow
+@pytest.mark.model(FAULT_TOLERANCE_MODEL_NAME)
 def test_vllm_health_check_active(request, runtime_services):
     """
     End-to-end test for worker fault tolerance with migration support.
@@ -203,8 +166,6 @@ def test_vllm_health_check_active(request, runtime_services):
     the system can handle the failure gracefully and migrate the request to
     another worker.
     """
-    # Step 0: Download the model from HuggingFace if not already cached
-    download_model()
 
     # Step 1: Start the frontend
     logger.info("Starting frontend...")
@@ -251,8 +212,8 @@ def test_vllm_health_check_active(request, runtime_services):
 @pytest.mark.vllm
 @pytest.mark.gpu_1
 @pytest.mark.e2e
-@pytest.mark.slow
-def test_vllm_health_check_passive(request, runtime_services):
+@pytest.mark.model(FAULT_TOLERANCE_MODEL_NAME)
+def test_vllm_health_check_passive(request, runtime_services, predownload_models):
     """
     End-to-end test for worker fault tolerance with migration support.
 
@@ -260,8 +221,6 @@ def test_vllm_health_check_passive(request, runtime_services):
     the system can handle the failure gracefully and migrate the request to
     another worker.
     """
-    # Step 0: Download the model from HuggingFace if not already cached
-    download_model()
 
     # Step 1: Start the frontend
     logger.info("Starting frontend...")

tests/kvbm/test_determinism.py

@@ -117,6 +117,8 @@ class LLMServerManager:
             "--kv-transfer-config",
             '{"kv_connector":"DynamoConnector","kv_role":"kv_both", "kv_connector_module_path": "dynamo.llm.vllm_integration.connector"}',
             os.environ.get("KVBM_MODEL_ID", "deepseek-ai/DeepSeek-R1-Distill-Llama-8B"),
+            "--max-seq-len",
+            "8000",  # required to fit on L4 GPU when using 8b model
         ]
 
         # GPU blocks override

tests/router/test_router_e2e_with_mockers.py

@@ -13,13 +13,14 @@ import aiohttp
 import pytest
 
 from dynamo._core import DistributedRuntime, KvPushRouter, KvRouterConfig
+from tests.utils.constants import ROUTER_MODEL_NAME
 from tests.utils.managed_process import ManagedProcess
 
 pytestmark = pytest.mark.pre_merge
 
 logger = logging.getLogger(__name__)
 
-MODEL_NAME = "Qwen/Qwen3-0.6B"
+MODEL_NAME = ROUTER_MODEL_NAME
 NUM_MOCKERS = 2
 BLOCK_SIZE = 16
 SPEEDUP_RATIO = 10.0
@@ -322,7 +323,8 @@ async def send_inflight_requests(urls: list, payload: dict, num_requests: int):
 
 
 @pytest.mark.pre_merge
-def test_mocker_kv_router(request, runtime_services):
+@pytest.mark.model(MODEL_NAME)
+def test_mocker_kv_router(request, runtime_services, predownload_tokenizers):
     """
     Test KV router with multiple mocker engine instances.
     This test doesn't require GPUs and runs quickly for pre-merge validation.
@@ -379,7 +381,8 @@ def test_mocker_kv_router(request, runtime_services):
 
 
 @pytest.mark.pre_merge
-def test_mocker_two_kv_router(request, runtime_services):
+@pytest.mark.model(MODEL_NAME)
+def test_mocker_two_kv_router(request, runtime_services, predownload_tokenizers):
     """
     Test with two KV routers and multiple mocker engine instances.
     Alternates requests between the two routers to test load distribution.
@@ -446,8 +449,11 @@ def test_mocker_two_kv_router(request, runtime_services):
 
 
 @pytest.mark.pre_merge
+@pytest.mark.model(MODEL_NAME)
 @pytest.mark.skip(reason="Flaky, temporarily disabled")
-def test_mocker_kv_router_overload_503(request, runtime_services):
+def test_mocker_kv_router_overload_503(
+    request, runtime_services, predownload_tokenizers
+):
     """
     Test that KV router returns 503 when all workers are busy.
     This test uses limited resources to intentionally trigger the overload condition.
@@ -612,7 +618,8 @@ def test_mocker_kv_router_overload_503(request, runtime_services):
 
 
 @pytest.mark.pre_merge
-def test_kv_push_router_bindings(request, runtime_services):
+@pytest.mark.model(MODEL_NAME)
+def test_kv_push_router_bindings(request, runtime_services, predownload_tokenizers):
     """
     Test KvPushRouter Python bindings with mocker engines.
     This test creates KvPushRouter as a Python object and verifies
@@ -839,7 +846,8 @@ def test_kv_push_router_bindings(request, runtime_services):
 
 
 @pytest.mark.pre_merge
-def test_indexers_sync(request, runtime_services):
+@pytest.mark.model(MODEL_NAME)
+def test_indexers_sync(request, runtime_services, predownload_tokenizers):
     """
     Test that two KV routers have synchronized indexer states after processing requests.
     This test verifies that both routers converge to the same internal state.
@@ -1066,7 +1074,10 @@ def test_indexers_sync(request, runtime_services):
 
 
 @pytest.mark.pre_merge
-def test_query_instance_id_returns_worker_and_tokens(request, runtime_services):
+@pytest.mark.model(MODEL_NAME)
+def test_query_instance_id_returns_worker_and_tokens(
+    request, runtime_services, predownload_tokenizers
+):
     """
     Test that the KV router correctly handles query_instance_id annotation.
 

tests/serve/common.py

@@ -4,12 +4,16 @@
 """Common base classes and utilities for engine tests (vLLM, TRT-LLM, etc.)"""
 
 import logging
+from collections.abc import Mapping
 from typing import Any, Dict, Optional
 
+import pytest
+
 from tests.utils.client import send_request
 from tests.utils.engine_process import EngineConfig, EngineProcess
 
 DEFAULT_TIMEOUT = 10
+SERVE_TEST_DIR = "/workspace/tests/serve"
 
 
 def run_serve_deployment(
@@ -58,3 +62,16 @@ def run_serve_deployment(
                     method=payload_item.method,
                 )
                 server_process.check_response(payload_item, response)
+
+
+def params_with_model_mark(configs: Mapping[str, EngineConfig]):
+    """Return pytest params for a config dict, adding a model marker per param.
+
+    This enables simple model collection after pytest filtering.
+    """
+    params = []
+    for config_name, cfg in configs.items():
+        marks = list(getattr(cfg, "marks", []))
+        marks.append(pytest.mark.model(cfg.model))
+        params.append(pytest.param(config_name, marks=marks))
+    return params

tests/serve/test_sglang.py

@@ -7,7 +7,11 @@ from dataclasses import dataclass, field
 
 import pytest
 
-from tests.serve.common import run_serve_deployment
+from tests.serve.common import (
+    SERVE_TEST_DIR,
+    params_with_model_mark,
+    run_serve_deployment,
+)
 from tests.utils.engine_process import EngineConfig
 from tests.utils.payload_builder import chat_payload_default, completion_payload_default
 
@@ -26,7 +30,7 @@ sglang_dir = os.environ.get("SGLANG_DIR", "/workspace/components/backends/sglang
 sglang_configs = {
     "aggregated": SGLangConfig(
         name="aggregated",
-        directory="/workspace/tests/serve",
+        directory=SERVE_TEST_DIR,
         script_name="sglang_agg.sh",
         marks=[pytest.mark.gpu_1],
         model="deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
@@ -67,13 +71,7 @@ sglang_configs = {
 }
 
 
-@pytest.fixture(
-    params=[
-        pytest.param("aggregated", marks=[pytest.mark.gpu_1]),
-        pytest.param("disaggregated", marks=[pytest.mark.gpu_2]),
-        pytest.param("kv_events", marks=[pytest.mark.gpu_2]),
-    ]
-)
+@pytest.fixture(params=params_with_model_mark(sglang_configs))
 def sglang_config_test(request):
     """Fixture that provides different SGLang test configurations"""
     return sglang_configs[request.param]
@@ -81,7 +79,9 @@ def sglang_config_test(request):
 
 @pytest.mark.e2e
 @pytest.mark.sglang
-def test_sglang_deployment(sglang_config_test, request, runtime_services):
+def test_sglang_deployment(
+    sglang_config_test, request, runtime_services, predownload_models
+):
     """Test SGLang deployment scenarios using common helpers"""
     config = sglang_config_test
     run_serve_deployment(config, request)
@@ -90,7 +90,7 @@ def test_sglang_deployment(sglang_config_test, request, runtime_services):
 @pytest.mark.skip(
     reason="Requires 4 GPUs - enable when hardware is consistently available"
 )
-def test_sglang_disagg_dp_attention(request, runtime_services):
+def test_sglang_disagg_dp_attention(request, runtime_services, predownload_models):
     """Test sglang disaggregated with DP attention (requires 4 GPUs)"""
 
     # Kept for reference; this test uses a different launch path and is skipped

tests/serve/test_trtllm.py

@@ -7,7 +7,7 @@ from dataclasses import dataclass, field
 
 import pytest
 
-from tests.serve.common import run_serve_deployment
+from tests.serve.common import params_with_model_mark, run_serve_deployment
 from tests.utils.engine_process import EngineConfig
 from tests.utils.payload_builder import chat_payload_default, completion_payload_default
 
@@ -84,12 +84,7 @@ trtllm_configs = {
 }
 
 
-@pytest.fixture(
-    params=[
-        pytest.param(config_name, marks=config.marks)
-        for config_name, config in trtllm_configs.items()
-    ]
-)
+@pytest.fixture(params=params_with_model_mark(trtllm_configs))
 def trtllm_config_test(request):
     """Fixture that provides different trtllm test configurations"""
     return trtllm_configs[request.param]
@@ -97,7 +92,7 @@ def trtllm_config_test(request):
 
 @pytest.mark.trtllm_marker
 @pytest.mark.e2e
-def test_deployment(trtllm_config_test, request, runtime_services):
+def test_deployment(trtllm_config_test, request, runtime_services, predownload_models):
     """
     Test dynamo deployments with different configurations.
     """
@@ -110,9 +105,8 @@ def test_deployment(trtllm_config_test, request, runtime_services):
 @pytest.mark.e2e
 @pytest.mark.gpu_1
 @pytest.mark.trtllm_marker
-@pytest.mark.slow
 def test_chat_only_aggregated_with_test_logits_processor(
-    request, runtime_services, monkeypatch
+    request, runtime_services, predownload_models, monkeypatch
 ):
     """
     Run a single aggregated chat-completions test using Qwen 0.6B with the

tests/serve/test_vllm.py

@@ -7,7 +7,7 @@ from dataclasses import dataclass, field
 
 import pytest
 
-from tests.serve.common import run_serve_deployment
+from tests.serve.common import params_with_model_mark, run_serve_deployment
 from tests.utils.engine_process import EngineConfig
 from tests.utils.payload_builder import (
     chat_payload,
@@ -185,12 +185,7 @@ vllm_configs = {
 }
 
 
-@pytest.fixture(
-    params=[
-        pytest.param(config_name, marks=config.marks)
-        for config_name, config in vllm_configs.items()
-    ]
-)
+@pytest.fixture(params=params_with_model_mark(vllm_configs))
 def vllm_config_test(request):
     """Fixture that provides different vLLM test configurations"""
     return vllm_configs[request.param]
@@ -198,7 +193,9 @@ def vllm_config_test(request):
 
 @pytest.mark.vllm
 @pytest.mark.e2e
-def test_serve_deployment(vllm_config_test, request, runtime_services):
+def test_serve_deployment(
+    vllm_config_test, request, runtime_services, predownload_models
+):
     """
     Test dynamo serve deployments with different graph configurations.
     """

tests/utils/constants.py

+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Shared test constants.
+
+Centralize model identifiers and other shared constants for tests to
+avoid importing from conftest and to keep values consistent.
+"""
+
+import os
+
+QWEN = "Qwen/Qwen3-0.6B"
+LLAMA = "deepseek-ai/DeepSeek-R1-Distill-Llama-8B"  # on an l4 gpu, must limit --max-seq-len, otherwise it will not fit
+
+TEST_MODELS = [
+    QWEN,
+    LLAMA,
+]
+
+# Env-driven defaults for specific test groups
+# Allows overriding via environment variables
+ROUTER_MODEL_NAME = os.environ.get("ROUTER_MODEL_NAME", QWEN)
+FAULT_TOLERANCE_MODEL_NAME = os.environ.get("FAULT_TOLERANCE_MODEL_NAME", QWEN)