phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-12-03 09:53:45 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

fix: Fix max_tool_calls for openai provider and add integration tests for the max_tool_calls feat (#4190)

Browse source 
# Problem

OpenAI gpt-4 returned an error when built-in and mcp calls were skipped
due to max_tool_calls parameter. Following is from the server log:
```
RuntimeError: OpenAI response failed: Error code: 400 - {'error': {'message': "An assistant message with       
'tool_calls' must be followed by tool messages responding to each 'tool_call_id'. The following tool_call_ids  
did not have response messages: call_Yi9V1QNpN73dJCAgP2Arcjej", 'type': 'invalid_request_error', 'param':      
'messages', 'code': None}}
```

# What does this PR do?

- Fixes error returned by openai/gpt when calls were skipped due to
max_tool_calls. We now return a tool message that explicitly mentions
that the call is skipped.
- Adds integration tests as a follow-up to
PR#[4062](https://github.com/llamastack/llama-stack/pull/4062)

<!-- If resolving an issue, uncomment and update the line below -->
Part 2 for issue
#[3563](https://github.com/llamastack/llama-stack/issues/3563)

## Test Plan
<!-- Describe the tests you ran to verify your changes with result
summaries. *Provide clear instructions so the plan can be easily
re-executed.* -->

- Added integration tests
- Added new recordings

---------

Co-authored-by: Ashwin Bharambe <ashwin.bharambe@gmail.com>
This commit is contained in:
Shabana Baig 2025-11-19 13:27:56 -05:00 • committed by GitHub
parent f18870a221
commit 72ea95e2e0
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
11 changed files with 8386 additions and 168 deletions
Download patch file Download diff file Expand all files Collapse all files

166
tests/integration/agents/test_openai_responses.py
View file

@ -516,169 +516,3 @@ def test_response_with_instructions(openai_client, client_with_models, text_mode
# Verify instructions from previous response was not carried over to the next response
assert response_with_instructions2.instructions == instructions2
@pytest.mark.skip(reason="Tool calling is not reliable.")
def test_max_tool_calls_with_function_tools(openai_client, client_with_models, text_model_id):
"""Test handling of max_tool_calls with function tools in responses."""
if isinstance(client_with_models, LlamaStackAsLibraryClient):
pytest.skip("OpenAI responses are not supported when testing with library client yet.")
client = openai_client
max_tool_calls = 1
tools = [
{
"type": "function",
"name": "get_weather",
"description": "Get weather information for a specified location",
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The city name (e.g., 'New York', 'London')",
},
},
},
},
{
"type": "function",
"name": "get_time",
"description": "Get current time for a specified location",
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The city name (e.g., 'New York', 'London')",
},
},
},
},
]
# First create a response that triggers function tools
response = client.responses.create(
model=text_model_id,
input="Can you tell me the weather in Paris and the current time?",
tools=tools,
stream=False,
max_tool_calls=max_tool_calls,
)
# Verify we got two function calls and that the max_tool_calls do not affect function tools
assert len(response.output) == 2
assert response.output[0].type == "function_call"
assert response.output[0].name == "get_weather"
assert response.output[0].status == "completed"
assert response.output[1].type == "function_call"
assert response.output[1].name == "get_time"
assert response.output[0].status == "completed"
# Verify we have a valid max_tool_calls field
assert response.max_tool_calls == max_tool_calls
def test_max_tool_calls_invalid(openai_client, client_with_models, text_model_id):
"""Test handling of invalid max_tool_calls in responses."""
if isinstance(client_with_models, LlamaStackAsLibraryClient):
pytest.skip("OpenAI responses are not supported when testing with library client yet.")
client = openai_client
input = "Search for today's top technology news."
invalid_max_tool_calls = 0
tools = [
{"type": "web_search"},
]
# Create a response with an invalid max_tool_calls value i.e. 0
# Handle ValueError from LLS and BadRequestError from OpenAI client
with pytest.raises((ValueError, BadRequestError)) as excinfo:
client.responses.create(
model=text_model_id,
input=input,
tools=tools,
stream=False,
max_tool_calls=invalid_max_tool_calls,
)
error_message = str(excinfo.value)
assert f"Invalid max_tool_calls={invalid_max_tool_calls}; should be >= 1" in error_message, (
f"Expected error message about invalid max_tool_calls, got: {error_message}"
)
def test_max_tool_calls_with_builtin_tools(openai_client, client_with_models, text_model_id):
"""Test handling of max_tool_calls with built-in tools in responses."""
if isinstance(client_with_models, LlamaStackAsLibraryClient):
pytest.skip("OpenAI responses are not supported when testing with library client yet.")
client = openai_client
input = "Search for today's top technology and a positive news story. You MUST make exactly two separate web search calls."
max_tool_calls = [1, 5]
tools = [
{"type": "web_search"},
]
# First create a response that triggers web_search tools without max_tool_calls
response = client.responses.create(
model=text_model_id,
input=input,
tools=tools,
stream=False,
)
# Verify we got two web search calls followed by a message
assert len(response.output) == 3
assert response.output[0].type == "web_search_call"
assert response.output[0].status == "completed"
assert response.output[1].type == "web_search_call"
assert response.output[1].status == "completed"
assert response.output[2].type == "message"
assert response.output[2].status == "completed"
assert response.output[2].role == "assistant"
# Next create a response that triggers web_search tools with max_tool_calls set to 1
response_2 = client.responses.create(
model=text_model_id,
input=input,
tools=tools,
stream=False,
max_tool_calls=max_tool_calls[0],
)
# Verify we got one web search tool call followed by a message
assert len(response_2.output) == 2
assert response_2.output[0].type == "web_search_call"
assert response_2.output[0].status == "completed"
assert response_2.output[1].type == "message"
assert response_2.output[1].status == "completed"
assert response_2.output[1].role == "assistant"
# Verify we have a valid max_tool_calls field
assert response_2.max_tool_calls == max_tool_calls[0]
# Finally create a response that triggers web_search tools with max_tool_calls set to 5
response_3 = client.responses.create(
model=text_model_id,
input=input,
tools=tools,
stream=False,
max_tool_calls=max_tool_calls[1],
)
# Verify we got two web search calls followed by a message
assert len(response_3.output) == 3
assert response_3.output[0].type == "web_search_call"
assert response_3.output[0].status == "completed"
assert response_3.output[1].type == "web_search_call"
assert response_3.output[1].status == "completed"
assert response_3.output[2].type == "message"
assert response_3.output[2].status == "completed"
assert response_3.output[2].role == "assistant"
# Verify we have a valid max_tool_calls field
assert response_3.max_tool_calls == max_tool_calls[1]