fix: increase timeout for test_{sglang,vllm,trtllm} model download time (#4792)

Signed-off-by: Keiven Chang <keivenchang@users.noreply.github.com>
Co-authored-by: Keiven Chang <keivenchang@users.noreply.github.com>

tests/serve/test_sglang.py

@@ -50,9 +50,7 @@ sglang_configs = {
         marks=[
             pytest.mark.gpu_1,
             pytest.mark.pre_merge,
-            pytest.mark.timeout(
-                180
-            ),  # ~5x measured time (39s), can take longer on HF donwload
+            pytest.mark.timeout(240),  # 3x measured time (39s) + download time (120s)
         ],
         model="Qwen/Qwen3-0.6B",
         env={},
@@ -133,7 +131,7 @@ sglang_configs = {
             pytest.mark.gpu_1,
             pytest.mark.pre_merge,
             pytest.mark.nightly,
-            pytest.mark.timeout(60),  # 3x measured time (20s)
+            pytest.mark.timeout(240),  # 3x measured time (20s) + download time (180s)
         ],
         model="Qwen/Qwen3-0.6B",
         env={},
@@ -181,7 +179,7 @@ sglang_configs = {
             pytest.mark.gpu_1,
             pytest.mark.pre_merge,
             pytest.mark.nightly,
-            pytest.mark.timeout(90),  # 3x measured time (29s)
+            pytest.mark.timeout(270),  # 3x measured time (29s) + download time (180s)
         ],
         model="Qwen/Qwen3-Embedding-4B",
         delayed_start=0,
@@ -217,8 +215,8 @@ sglang_configs = {
         marks=[
             pytest.mark.gpu_1,
             pytest.mark.timeout(
-                160
-            ),  # Total test timeout: 2x measured average (79.36s)
+                420
+            ),  # Total test timeout: 2x measured average (79.36s) + download time (240s) for 7B model
         ],
         model="deepseek-ai/deepseek-llm-7b-base",
         script_args=[

tests/serve/test_trtllm.py

@@ -46,7 +46,9 @@ trtllm_configs = {
             pytest.mark.gpu_1,
             pytest.mark.pre_merge,
             pytest.mark.trtllm,
-            pytest.mark.timeout(140),  # 3x measured time (44.66s)
+            pytest.mark.timeout(
+                300
+            ),  # 3x measured time (44.66s) + download time (150s)
         ],
         model="Qwen/Qwen3-0.6B",
         models_port=8000,
@@ -76,7 +78,9 @@ trtllm_configs = {
             pytest.mark.gpu_1,
             pytest.mark.pre_merge,
             pytest.mark.trtllm,
-            pytest.mark.timeout(320),  # 3x measured time (103.66s)
+            pytest.mark.timeout(
+                480
+            ),  # 3x measured time (103.66s) + download time (150s)
         ],
         model="Qwen/Qwen3-0.6B",
         models_port=8000,
@@ -95,7 +99,9 @@ trtllm_configs = {
             pytest.mark.gpu_1,
             pytest.mark.pre_merge,
             pytest.mark.trtllm,
-            pytest.mark.timeout(120),  # 3x measured time (37.91s)
+            pytest.mark.timeout(
+                300
+            ),  # 3x measured time (37.91s) + download time (180s)
         ],
         model="Qwen/Qwen3-0.6B",
         models_port=8000,
@@ -141,7 +147,9 @@ trtllm_configs = {
         marks=[
             pytest.mark.gpu_1,
             pytest.mark.trtllm,
-            pytest.mark.timeout(260),  # 3x measured time (83.85s)
+            pytest.mark.timeout(
+                480
+            ),  # 3x measured time (83.85s) + download time (210s) for 7B model
         ],
         model="deepseek-ai/deepseek-llm-7b-base",
         script_args=["--dyn-endpoint-types", "completions"],
@@ -177,7 +185,7 @@ def test_deployment(trtllm_config_test, request, runtime_services, predownload_m
 @pytest.mark.e2e
 @pytest.mark.gpu_1
 @pytest.mark.trtllm
-@pytest.mark.timeout(480)  # 3x measured time (159.68s)
+@pytest.mark.timeout(660)  # 3x measured time (159.68s) + download time (180s)
 def test_chat_only_aggregated_with_test_logits_processor(
     request, runtime_services, predownload_models, monkeypatch
 ):

tests/serve/test_vllm.py

@@ -50,7 +50,7 @@ vllm_configs = {
         marks=[
             pytest.mark.gpu_1,
             pytest.mark.pre_merge,
-            pytest.mark.timeout(130),  # 3x measured time (43s)
+            pytest.mark.timeout(300),  # 3x measured time (43s) + download time (150s)
         ],
         model="Qwen/Qwen3-0.6B",
         request_payloads=[
@@ -66,7 +66,7 @@ vllm_configs = {
         marks=[
             pytest.mark.gpu_1,
             pytest.mark.pre_merge,
-            pytest.mark.timeout(210),  # 3x estimated time (70s)
+            pytest.mark.timeout(360),  # 3x estimated time (70s) + download time (150s)
         ],
         model="Qwen/Qwen3-0.6B",
         request_payloads=[
@@ -82,7 +82,7 @@ vllm_configs = {
         script_name="agg_lmcache_multiproc.sh",
         marks=[
             pytest.mark.gpu_1,
-            pytest.mark.timeout(210),  # 3x estimated time (70s)
+            pytest.mark.timeout(360),  # 3x estimated time (70s) + download time (150s)
         ],
         model="Qwen/Qwen3-0.6B",
         env={
@@ -102,7 +102,7 @@ vllm_configs = {
         marks=[
             pytest.mark.gpu_1,
             pytest.mark.pre_merge,
-            pytest.mark.timeout(130),  # 3x measured time (43s)
+            pytest.mark.timeout(300),  # 3x measured time (43s) + download time (150s)
         ],
         model="Qwen/Qwen3-0.6B",
         script_args=["--tcp"],
@@ -118,7 +118,7 @@ vllm_configs = {
         marks=[
             pytest.mark.gpu_1,
             pytest.mark.pre_merge,
-            pytest.mark.timeout(130),  # 3x measured time (43s)
+            pytest.mark.timeout(300),  # 3x measured time (43s) + download time (150s)
         ],
         model="Qwen/Qwen3-0.6B",
         script_args=["--http"],
@@ -439,7 +439,9 @@ vllm_configs = {
         script_name="agg.sh",
         marks=[
             pytest.mark.gpu_1,
-            pytest.mark.timeout(180),  # 3x estimated time (60s) for 7B model
+            pytest.mark.timeout(
+                420
+            ),  # 3x estimated time (60s) + download time (240s) for 7B model
         ],
         model="deepseek-ai/deepseek-llm-7b-base",
         script_args=[