Add support for "extra_body" to OpenAILLMConfigEntry

[Hellisotherpeople]

This will enable min_p sampling and other fancy vllm stuff for AG2.

Why are these changes needed?

OpenAILLMConfigEntry is used for all OpenAI compliant APIs. vLLMs server is just one of those APIs. vllm also supports a bunch of cool stuff, such as min_p sampling, robust max/min tokens support. This is all specified in the "extra_body" - but since this was not supported by OpenAILLMConfigEntry, it was not possible to use these techniques with local models.

I've tested this change with several models on vllm, including mistral large, llama4 maverick, and deepseek.

• I've included any doc changes needed for https://docs.ag2.ai/. See https://docs.ag2.ai/docs/contributor-guide/documentation to build and test documentation locally.
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.

[CLAassistant]

All committers have signed the CLA.

[Hellisotherpeople]

I will shortly add documentation and tests to this PR

[merlintang]

this is what we want.

[davorrunje]

@Hellisotherpeople Thank you for the PR! I am waiting for docs and tests.

[kumaranvpl]

Looks good to me. If you are adding extra_body, would you like to add extra_headers as well?

[Hellisotherpeople]

What's the best way to do tests for this? To test this, I figured that I need to point to a vllm server, pass in something like min_p sampling, and verify that I get a (valid) response.

If we do not want to point to any vllm server or run one, what's the ideal way to test this?

[kumaranvpl]

What's the best way to do tests for this? To test this, I figured that I need to point to a vllm server, pass in something like min_p sampling, and verify that I get a (valid) response.

If we do not want to point to any vllm server or run one, what's the ideal way to test this?

@Hellisotherpeople For now, just try to set some valid parameters for extra_body and assert that the model_dump is same as expected. Similar to this - https://github.com/ag2ai/ag2/blob/main/test/oai/test_client.py#L384-L399

[kumaranvpl]

Looks good to me. If you are adding extra_body, would you like to add extra_headers as well?

@Hellisotherpeople Please take a look at my earlier comment

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

❗ There is a different number of reports uploaded between BASE (9732e25) and HEAD (8711815). Click for more details.

HEAD has 75 uploads less than BASE

Flag	BASE (9732e25)	HEAD (8711815)
core-without-llm	14	1
3.9	11	0
ubuntu-latest	14	1
3.13	3	0
macos-latest	4	0
3.11	3	1
3.10	3	0
3.12	3	0
windows-latest	5	0
core-llm	9	0
cerebras	1	0
ollama	1	0
bedrock	1	0
gemini	1	0
openai-realtime	1	0
gemini-realtime	1	0
anthropic	1	0
deepseek	1	0
openai	1	0

Files with missing lines	Coverage Δ
autogen/oai/client.py	53.60% <100.00%> (-23.54%)	⬇️

... and 64 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.