mirror of
https://github.com/langchain-ai/langchain.git
synced 2026-10-06 01:45:27 +03:00
Juan Ezquerro LLanes
ff46bb478b
Fixes #40985 --- `ChatOllama(reasoning="xhigh")` currently raises a Pydantic `ValidationError` instead of reaching Ollama, because the `ollama` floor still admits clients that reject model-defined thinking levels. `ChatOllama.reasoning` is `bool | str | None` and `_chat_params()` forwards the string verbatim as `think`. Since ollama 0.6.3 (ollama/ollama-python#744) `ChatRequest.think` is `Optional[Union[bool, str]]`, matching the Ollama Go backend. But the declared range `ollama>=0.6.1,<1.0.0` still admits 0.6.1 and 0.6.2, where the annotation is `Optional[Union[bool, Literal["low", "medium", "high"]]]` and `"xhigh"` or `"max"` fails validation client-side. Verified both directions on the same snippet — `ChatOllama(reasoning="xhigh")._chat_params(...)["think"] == "xhigh"`, then constructing `ChatRequest(..., think=...)`: - ollama 0.6.2 → `ValidationError` on `think.bool` and `think.literal['low','medium','high']` - ollama 0.6.3 → validates This raises the floor only; no `ChatOllama` code path changes. Narrowing `reasoning` inside `ChatOllama` instead would reject levels the Ollama backend accepts and duplicate type logic the client now owns. Per the PR template I'm opening this alongside the issue rather than pre-approved, so please assign it to me if the approach looks right. ## Release note `langchain-ollama` now requires `ollama>=0.6.3`, fixing a `ValidationError` when `ChatOllama` is configured with a model-defined reasoning level such as `"xhigh"` or `"max"`.
FAQ
Looking for an integration not listed here? Check out the integrations documentation and the note in the libs/ README about third-party maintained packages.
Integration docs
For full documentation, see the primary and API reference docs for integrations.