A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 10, 09:00 UTC
[ROCm][CI] Fix flaky `test_function_calling_with_stream` and reduce schema test examples (#32063)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Jan 9, 2026 at 23:02 UTC
d83becd503660fb876ea42beaa9f63217b857b99
2 files changed
+14
-10
tests/entrypoints/openai/test_openai_schema.py
+1
-1
@@ -129,7 +129,7 @@ def before_generate_case(context: schemathesis.hooks.HookContext, strategy):
129
130
@schema.parametrize()
131
@schema.override(headers={"Content-Type": "application/json"})
132
-@settings(deadline=LONG_TIMEOUT_SECONDS * 1000)
132
+@settings(deadline=LONG_TIMEOUT_SECONDS * 1000, max_examples=50)
133
def test_openapi_stateless(case: schemathesis.Case):
134
key = (
135
case.operation.method.upper(),
tests/entrypoints/openai/test_response_api_with_harmony.py
+13
-9
@@ -821,16 +821,20 @@ async def test_function_calling_with_stream(client: OpenAI, model_name: str):
821
final_tool_calls_named[tool_call.name] = tool_call
822
elif event.type == "response.function_call_arguments.done":
823
assert event.arguments == final_tool_calls_named[event.name].arguments
824
- for tool_call in final_tool_calls.values():
825
- if (
826
- tool_call
827
- and tool_call.type == "function_call"
828
- and tool_call.name == "get_weather"
829
- ):
830
- args = json.loads(tool_call.arguments)
831
- result = call_function(tool_call.name, args)
832
- input_list += [tool_call]
824
+ result = None
825
+ tool_call = None
826
+ for tc in final_tool_calls.values():
827
+ if tc and tc.type == "function_call" and tc.name == "get_weather":
828
+ args = json.loads(tc.arguments)
829
+ result = call_function(tc.name, args)
830
+ tool_call = tc
831
+ input_list += [tc]
832
break
833
+
834
+ assert tool_call is not None, (
835
+ "Expected model to call 'get_weather' function, "
836
+ f"but got: {list(final_tool_calls_named.keys())}"
837
+ )
838
assert result is not None
839
response = await client.responses.create(
840
model=model_name,