[BUG] Litellm: max_tokens no supported

[enoch3712]

No description provided.

[Copilot]

Pull Request Overview

This PR updates model references, test expectations, and LLM token-handling logic to support new LLM variants and token limit parameters.

• Tests now import and use get_gpt_o4_model, get_gpt_mini_model, get_gemini_flash_model, and adjust expected values for outputs.
• The LLM class gains DEFAULT_OUTPUT_TOKENS, thinking_token_limit, and refactors thinking-mode parameter handling, replacing asyncio loop management with asyncio.run.
• Global model getters are extended, and the extractor handles dict sources via DocumentLoaderData.

Reviewed Changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
tests/test_process.py	Switched to direct model string in splitter and adjusted expected age.
tests/test_extractor.py	Expanded model getters, renamed budget fields, added thinking-mode tests, and updated assertions.
extract_thinker/llm.py	Introduced default output tokens, thinking-token-limit, and refactored async and thinking-parameter flows.
extract_thinker/global_models.py	Added new model getters for GPT and Gemini Flash models.
extract_thinker/extractor.py	Auto-loads DocumentLoaderData when source is a dict.
extract_thinker/concatenation_handler.py	Added guard against empty responses in JSON continuation check.

Comments suppressed due to low confidence (4)

tests/test_extractor.py:310

• The assertion uses threshold=0.8 (80%) but the error message says 90%. Update the message to match the actual threshold.

assert semantically_similar(
        result_1.title,
        result_2.title,
        threshold=0.8
    ), "Titles are not semantically similar enough (threshold: 90%)"

extract_thinker/global_models.py:9

• [nitpick] Both get_lite_model() and get_big_model() return the same string; if they are intended to be different, adjust accordingly or clarify why they match.

    return "gemini/gemini-2.5-flash-preview-05-20"

extract_thinker/extractor.py:219

• Ensure DocumentLoaderData is imported at the top of the module; otherwise, this will raise a NameError at runtime.

            self.document_loader = DocumentLoaderData()

extract_thinker/llm.py:175

• Confirm that import asyncio exists at module scope. If it was removed, calls to asyncio.run will fail with NameError.

result = asyncio.run(

[Copilot]

[nitpick] Remove or resolve this TODO before merging to avoid lingering technical debt in tests.

[Copilot]

[nitpick] This __main__ block and manual test call are unnecessary for pytest; consider removing it.

Suggested change

	if __name__ == "__main__":
	test_chart_with_content()
	# Removed the __main__ block and manual test call.

[Copilot]

[nitpick] Rather than hard-coding the model string, use get_big_model() for consistency and easier future updates.

Suggested change

	process.load_splitter(ImageSplitter("gemini/gemini-2.5-flash-preview-05-20"))
	process.load_splitter(ImageSplitter(get_big_model()))

[Copilot]

[nitpick] The logic to compute max_tokens is duplicated across several methods; consider extracting it into a helper to reduce duplication.