Fix top_p null still sent in API requests after #385

[toslali-ibm]

Summary

PR #385 (closing #378) made top_p optional and defaulted it to None in LLMConfig, but the API call layer in openevolve/llm/openai.py still unconditionally includes top_p in the request params:

params = {
    ...
    "top_p": kwargs.get("top_p", self.top_p),  # sends None when top_p is unset
    ...
}

This puts "top_p": None into the params dict. Providers like AWS Bedrock, LiteLLM, and other OpenAI-compatible wrappers may serialize it as "top_p": null in the JSON body, triggering a 400 error.

Fix: Only include top_p in params when it is explicitly set (not None).

Changes

• openevolve/llm/openai.py: Conditionally include top_p in request params only when not None

Test plan

• All 368 tests pass
• No new config-level changes needed (handled by Fix Anthropic models error when both temperature and top_p are passed #385)

Fixes #414

[toslali-ibm]

@codelion This is a follow-up to your #385 fix — it addressed the config layer but openai.py still unconditionally includes top_p in the request params, which sends null to providers that don't strip it (e.g. Bedrock). Small 3-line fix.

Let me know your comments.
Thanks!

[CLAassistant]

All committers have signed the CLA.