[Bug]: Nova <thinking> sections are not removed

[adampauls]

What happened?

When Amazon Nova is invokes tools, it returns <thinking> messages as text content (see note here). For example:

{'content': '<thinking> The user has provided their email address. I will use the `find_user_id_by_email` tool to locate the user ID. </thinking>

', 'function_call': None, 'role': 'assistant', 'tool_calls': [{'function': {'arguments': '{"email": "[email protected]"}', 'name': 'find_user_id_by_email'}, 'id': 'tooluse_ImoUPIm6SlueNzyx9xXNzg', 'index': 1, 'type': 'function'}]}

<thinking> can also occur without tool use, though the Amazon documentation is less clear here. I think the intended use is that everything between <thinking> and </thinking> tags should be stripped before being returned out to the user, but I'm not sure what Amazon is intending here, or why they didn't structure their API more like other APIs that have reasoning blocks.

Relevant log output

Are you a ML Ops Team?

No

What LiteLLM version are you on ?

v1.63.3

Twitter / LinkedIn details

No response