Fix function_calling_llm support for custom models

[devin-ai-integration]

Fix function_calling_llm support for custom models

Summary

Fixes #3708 where function_calling_llm did not work for custom models not in litellm's supported models list.

The issue occurred because when a user set function_calling_llm to use a custom model (e.g., a company's proprietary model through a custom provider), the framework would check if the model supports function calling using litellm.utils.supports_function_calling(). For models not in litellm's list, this check would return False, preventing function calling from being used even though the model actually supported it.

Changes:

1. BaseLLM class: Added supports_function_calling() method that returns True by default (similar to existing supports_stop_words() method)
2. LLM class: Added supports_function_calling parameter to allow explicit override of litellm's auto-detection
3. Tests: Added test coverage for both the BaseLLM default and LLM override functionality

Usage examples:

# For custom models through litellm that support function calling
llm = LLM(model="custom-provider/my-model", supports_function_calling=True)

# For custom LLMs extending BaseLLM (now works by default)
class MyLLM(BaseLLM):
    def call(self, messages, tools=None, ...):
        # implementation
        pass
    # No need to override supports_function_calling() unless you want False

Review & Testing Checklist for Human

• Run full test suite - I encountered pytest plugin conflicts (pytest-recording vs pytest-vcr) and could only verify via direct Python execution. Please run uv run pytest tests/ to ensure no regressions
• Test with actual custom model - Create an agent with a custom model (not in litellm's list) and verify function_calling_llm works with the new supports_function_calling=True parameter
• Verify backward compatibility - Check that existing custom LLMs that override supports_function_calling() still work correctly
• Check dependency changes - The uv.lock file was regenerated from scratch due to corruption; review for any unexpected dependency updates

Test Plan

1. Create an LLM with a custom model name: LLM(model="company/custom-model", supports_function_calling=True)
2. Set it as function_calling_llm on an agent
3. Verify that tools use function calling (via event listeners or logging)
4. Run existing test suite to ensure no regressions

Notes

• Link to Devin run: https://app.devin.ai/sessions/df488c73f307471a8334bf31a82b686e
• Requested by: João ([email protected])
• The fix maintains backward compatibility by making the override optional (defaults to None, which uses litellm's check)

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring