fix: pass model and llm_provider to ContextWindowExceededError constructor

[paulxg12]

What

Fix a TypeError in _call_llm_streaming and _call_llm_non_streaming that prevents context compaction from triggering when the LLM responds with a context overflow error.

Why

Fixes #270. When litellm raises a BadRequestError (or similar) indicating the prompt exceeds the model context window, the exception handlers attempt to re-raise it as ContextWindowExceededError but omit the required model and llm_provider positional arguments:

TypeError: ContextWindowExceededError.__init__() missing 2 required positional arguments: 'model' and 'llm_provider'

This prevents the except ContextWindowExceededError: handler in the main agent loop (line ~1692) from ever catching the error, so context compaction never fires. The session is permanently stuck throwing the same TypeError on every iteration.

How

Both _call_llm_streaming (line ~876) and _call_llm_non_streaming (line ~1033) now pass:

• model — from llm_params.get("model", "")
• llm_provider — extracted as the prefix before / in the model string (e.g. "openai", "anthropic", "bedrock")

These values are already available in llm_params at the call site.

Testing

• The fix is a minimal change (2 call sites, +10 lines) that matches the constructor signature used everywhere else in litellm.
• The except ContextWindowExceededError: handler in run_agent at line ~1692 can now correctly catch the re-raised error and trigger context compaction.

Checklist

• Follows existing code style
• No breaking changes
• Documentation updated if needed

[lewtun]

Thanks for contributing to ML Intern. The project, including its CLI, has been retired and this repository is being archived. We will therefore not be able to review or merge this work. For continued AI assistance, please use HuggingChat.