fix: Ollama Truncation

[monoxgas]

Notes

Ollama has a known behavior where it performs silent truncation of input messages rather than return an error or any API indication.

To make a best effort on resolution:

• Emit a warning if input_tokens for a litellm response is far lower than the length of the messages we sent.
• Docs updates to call out the behavior.

See:

• Warn users when input is longer than supported context ollama/ollama#7043
• better diagnosis / error messages when ctx is too small ollama/ollama#7987
• API Silently Truncates Conversation  ollama/ollama#4967

---

Generated Summary

• Introduced a new GeneratorWarning class and integrated warning handling to detect potential Ollama input truncation.
• Updated get_generator functions (in both documentation and implementation) to accept a dict for params, converting it to GenerateParams when needed.
• Added token truncation detection in LiteLLMGenerator with a dedicated _warn_on_input_truncation method that warns when input tokens appear truncated.
• Refactored XML field name resolution in the Model by adding a _get_field_xml_name helper, replacing inline logic and adding alias support.
• Updated documentation: added a new GeneratorWarning section, revised generator API examples, and switched from "ollama/qwen3" to "ollama_chat/qwen3" including a warning block regarding truncation.
• Added a new regex exclusion rule in .secrets.baseline to ignore files in the examples/ directory.
• Bumped the project version to 3.3.3 in pyproject.toml.

This summary was generated with ❤️ by rigging

Generated Summary

• Added a new GeneratorWarning class and integrated warnings for Ollama’s silent input truncation in the LiteLLM generator.
• Updated the get_generator interface to accept dict inputs for params in addition to GenerateParams.
• Modified documentation for the generator API, including updated examples and a warning block regarding Ollama’s context length limitations.
• Refactored XML field name resolution by introducing a _get_field_xml_name method to standardize field name extraction in Model methods.
• Adjusted the .secrets.baseline file by adding file exclusion rules and updating secret line numbers.
• Removed lru_cache from the get_generator function in the generator base module.
• Bumped the package version from 3.3.2 to 3.3.3.

This summary was generated with ❤️ by rigging

[GangGreenTemperTatum]

tested and confirmed working as per our chat 👌
FWIW i also tested and confirmed working on somewhat related PR 👍