Proposed Changes - Replace the default model from `moonshotai/Kimi-K2.5` to `Qwen/Qwen3-32B`. - Remove the provider-specific `extra_body={"thinking": {"type": "disabled"}}` parameter. - Update inference examples to use a model that is currently supported through Hugging Face Inference Providers. - Improve out-of-the-box execution for users running the notebook on Colab or local environments.
The original notebook used moonshotai/Kimi-K2.5 together with a provider-specific thinking configuration. While the model exists on Hugging Face, it is not currently available through the default Hugging Face Inference Provider routing for many users, resulting in:
BadRequestError:
"The requested model 'moonshotai/Kimi-K2.5' is not supported by any provider you have enabled."
This update replaces the model with Qwen/Qwen3-32B, which is compatible with Hugging Face's chat completion interface and is more broadly available through supported inference providers.
Additionally, the thinking parameter was removed because it is model/provider specific and not required for the notebook's learning objectives. Removing it improves portability across different hosted models while keeping the agent implementation and educational content unchanged.
The notebook behavior, agent workflow, and learning outcomes remain identical; the change simply improves compatibility and reduces setup friction for learners.