fix: skip temperature param for OpenAI reasoning/newer models in summary tool

[octo-patch]

Fixes #211

Problem

When using newer OpenAI models (o1, o3, o4 reasoning models, gpt-5, etc.) as the summary model (SUMMARY_MODEL_NAME), the call_server method in both inference/tool_visit.py and WebAgent/WebResummer/src/tool_visit.py unconditionally passes temperature=0.7. These model families do not support the temperature parameter and will return an API error.

Solution

Before building the API call kwargs, check whether the model name starts with a prefix known to not support temperature (o1, o3, o4, gpt-5). Only include temperature in the request when the model supports it.

_NO_TEMPERATURE_PREFIXES = ("o1", "o3", "o4", "gpt-5")
create_kwargs = dict(model=model_name, messages=msgs)
if not any(model_name.startswith(p) for p in _NO_TEMPERATURE_PREFIXES):
    create_kwargs["temperature"] = 0.7
chat_response = client.chat.completions.create(**create_kwargs)

Files changed

• inference/tool_visit.py — Visit.call_server()
• WebAgent/WebResummer/src/tool_visit.py — Visit.call_server()

Testing

Tested locally by setting SUMMARY_MODEL_NAME=o1-mini and verifying no temperature is passed; and with SUMMARY_MODEL_NAME=gpt-4o to confirm temperature=0.7 is still included for standard models.