Skip to content

fix: Update trtllm tests to use new scripts instead of dynamo serve #1979

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 8 commits into from
Jul 17, 2025
Merged

fix: Update trtllm tests to use new scripts instead of dynamo serve #1979

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
84e0b4e
Add test_trtllm.py, remove everything except multimodal from test_dyn…
rmccorm4 Jul 16, 2025
8f2aae6
Merge branch 'main' of github.com:ai-dynamo/dynamo into rmccormick/tr…
rmccorm4 Jul 16, 2025
85b2a16
Update directory to components/backends/trtllm
rmccorm4 Jul 16, 2025
bfb6058
Update port to 8000 to match launch script port, set env vars to matc…
rmccorm4 Jul 17, 2025
9027dc9
Update vllm test directory for new location
rmccorm4 Jul 17, 2025
df1da1b
Undo vllm changes to keep this PR lean, it will be covered in Biswa/A…
rmccorm4 Jul 17, 2025
4fe5f7f
Review feedback: remove dupe logs, walrus operator elapsed, add model…
rmccorm4 Jul 17, 2025
ab05edf
Merge branch 'main' of github.com:ai-dynamo/dynamo into rmccormick/tr…
rmccorm4 Jul 17, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
189 changes: 0 additions & 189 deletions tests/serve/test_dynamo_serve.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,6 @@
DeploymentGraph,
Payload,
chat_completions_response_handler,
completions_response_handler,
)
from tests.utils.managed_process import ManagedProcess

Expand Down Expand Up @@ -56,106 +55,7 @@
expected_response=["bus"],
)

text_payload = Payload(
payload_chat={
"model": "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
"messages": [
{
"role": "user",
"content": text_prompt, # Shorter prompt
}
],
"max_tokens": 150, # Reduced from 500
"temperature": 0.1,
# "seed": 0,
},
payload_completions={
"model": "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
"prompt": text_prompt,
"max_tokens": 150,
"temperature": 0.1,
# "seed": 0,
},
repeat_count=10,
expected_log=[],
expected_response=["AI"],
)

deployment_graphs = {
"agg": (
DeploymentGraph(
module="graphs.agg:Frontend",
config="configs/agg.yaml",
directory="/workspace/examples/llm",
endpoints=["v1/chat/completions"],
response_handlers=[
chat_completions_response_handler,
],
marks=[pytest.mark.gpu_1, pytest.mark.vllm],
),
text_payload,
),
"sglang_agg": (
DeploymentGraph(
module="graphs.agg:Frontend",
config="configs/agg.yaml",
directory="/workspace/examples/sglang",
endpoints=["v1/chat/completions", "v1/completions"],
response_handlers=[
chat_completions_response_handler,
completions_response_handler,
],
marks=[pytest.mark.gpu_1, pytest.mark.sglang],
),
text_payload,
),
"disagg": (
DeploymentGraph(
module="graphs.disagg:Frontend",
config="configs/disagg.yaml",
directory="/workspace/examples/llm",
endpoints=["v1/chat/completions"],
response_handlers=[
chat_completions_response_handler,
],
marks=[pytest.mark.gpu_2, pytest.mark.vllm],
),
text_payload,
),
"agg_router": (
DeploymentGraph(
module="graphs.agg_router:Frontend",
config="configs/agg_router.yaml",
directory="/workspace/examples/llm",
endpoints=["v1/chat/completions"],
response_handlers=[
chat_completions_response_handler,
],
marks=[pytest.mark.gpu_1, pytest.mark.vllm],
# FIXME: This is a hack to allow deployments to start before sending any requests.
# When using KV-router, if all the endpoints are not registered, the service
# enters a non-recoverable state.
delayed_start=120,
),
text_payload,
),
"disagg_router": (
DeploymentGraph(
module="graphs.disagg_router:Frontend",
config="configs/disagg_router.yaml",
directory="/workspace/examples/llm",
endpoints=["v1/chat/completions"],
response_handlers=[
chat_completions_response_handler,
],
marks=[pytest.mark.gpu_2, pytest.mark.vllm],
# FIXME: This is a hack to allow deployments to start before sending any requests.
# When using KV-router, if all the endpoints are not registered, the service
# enters a non-recoverable state.
delayed_start=120,
),
text_payload,
),
"multimodal_agg": (
DeploymentGraph(
module="graphs.agg:Frontend",
Expand All @@ -169,84 +69,6 @@
),
multimodal_payload,
),
"vllm_v1_agg": (
DeploymentGraph(
module="graphs.agg:Frontend",
config="configs/agg.yaml",
directory="/workspace/examples/vllm_v1",
endpoints=["v1/chat/completions", "v1/completions"],
response_handlers=[
chat_completions_response_handler,
completions_response_handler,
],
marks=[pytest.mark.gpu_1, pytest.mark.vllm],
),
text_payload,
),
"trtllm_agg": (
DeploymentGraph(
module="graphs.agg:Frontend",
config="configs/agg.yaml",
directory="/workspace/examples/tensorrt_llm",
endpoints=["v1/chat/completions", "v1/completions"],
response_handlers=[
chat_completions_response_handler,
completions_response_handler,
],
marks=[pytest.mark.gpu_1, pytest.mark.tensorrtllm],
),
text_payload,
),
"trtllm_agg_router": (
DeploymentGraph(
module="graphs.agg:Frontend",
config="configs/agg_router.yaml",
directory="/workspace/examples/tensorrt_llm",
endpoints=["v1/chat/completions", "v1/completions"],
response_handlers=[
chat_completions_response_handler,
completions_response_handler,
],
marks=[pytest.mark.gpu_1, pytest.mark.tensorrtllm],
# FIXME: This is a hack to allow deployments to start before sending any requests.
# When using KV-router, if all the endpoints are not registered, the service
# enters a non-recoverable state.
delayed_start=120,
),
text_payload,
),
"trtllm_disagg": (
DeploymentGraph(
module="graphs.disagg:Frontend",
config="configs/disagg.yaml",
directory="/workspace/examples/tensorrt_llm",
endpoints=["v1/chat/completions", "v1/completions"],
response_handlers=[
chat_completions_response_handler,
completions_response_handler,
],
marks=[pytest.mark.gpu_2, pytest.mark.tensorrtllm],
),
text_payload,
),
"trtllm_disagg_router": (
DeploymentGraph(
module="graphs.disagg:Frontend",
config="configs/disagg_router.yaml",
directory="/workspace/examples/tensorrt_llm",
endpoints=["v1/chat/completions", "v1/completions"],
response_handlers=[
chat_completions_response_handler,
completions_response_handler,
],
marks=[pytest.mark.gpu_2, pytest.mark.tensorrtllm],
# FIXME: This is a hack to allow deployments to start before sending any requests.
# When using KV-router, if all the endpoints are not registered, the service
# enters a non-recoverable state.
delayed_start=120,
),
text_payload,
),
}


Expand Down Expand Up @@ -394,17 +216,6 @@ def wait_for_ready(self, payload, logger=logging.getLogger()):
@pytest.fixture(
params=[
pytest.param("multimodal_agg", marks=[pytest.mark.vllm, pytest.mark.gpu_2]),
pytest.param("trtllm_agg", marks=[pytest.mark.tensorrtllm, pytest.mark.gpu_1]),
pytest.param(
"trtllm_agg_router", marks=[pytest.mark.tensorrtllm, pytest.mark.gpu_1]
),
pytest.param(
"trtllm_disagg", marks=[pytest.mark.tensorrtllm, pytest.mark.gpu_2]
),
pytest.param(
"trtllm_disagg_router", marks=[pytest.mark.tensorrtllm, pytest.mark.gpu_2]
),
# pytest.param("sglang", marks=[pytest.mark.sglang, pytest.mark.gpu_2]),
]
)
def deployment_graph_test(request):
Expand Down
Loading
Loading