ai-twinkle/Eval

AI Summary: Debug a Twinkle Eval integration issue where using an API key generated from the NVIDIA GPT-OSS-20B example results in a 'NoneType' object is not subscriptable error. This involves analyzing the response format differences between the NVIDIA endpoint and Twinkle Eval's expected format, and potentially modifying Twinkle Eval's code to handle the NVIDIA response correctly.

AI Summary: The task is to modify the Twinkle Eval tool to correctly handle the `reasoning` parameter when interacting with the Ollama LLM. Currently, it uses `reasoning_content`, which is incorrect for Ollama. A check should be added to differentiate between Ollama and other LLMs (like vLLM) to use the appropriate parameter name.