feat(ollama): Add think Field to ChatOllama for Controlling Thought Process in Responses

[rliu6915]

• Add think field to ChatOllamaInput interface
• Add think field to chatOllama class (require by the interface it implements)
• Initialize think field through its constructor
• Add think property to OllamaChatRequest object as the return of invocationParams() method
• In _streamResponseChunks method, in terms of creating token, when thinking is enabled, make sure trying thinking first using (??) operator
• in convertOllamaMessagesToLangChain method, when setting content, make sure trying thinking first (messages.thinking) using (??) operator
• Add test suite for thinking model
• Test thinking model (deepseek-r1-32b) response

This change resolves the issue where users using thinking model such as Deepseek-r1-32b couldn't hide thought process when they are using Ollama through langchain.js framework's Ollama integration. Previously, attempting to use thinking model such as deepssek with thinking mode set to false. output would still result in a response with a combination of thinking and content.

The fix includes:

• Adding think field to chatOllama class to support thinking model with thinking mode such as deepseek
• Test coverage for thinking model (deepseek) with or without thinking mode

Fixes #8481

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Preview	Comments	Updated (UTC)
langchainjs-docs	Ready	Preview		Sep 1, 2025 6:35am

1 Skipped Deployment

Project	Deployment	Preview	Comments	Updated (UTC)
langchainjs-api-refs	Ignored			Sep 1, 2025 6:35am

[rliu6915]

@hntrl @jacoblee93 Could you please take a moment to review the six workflows that are currently awaiting your approval?

[christian-bromann]

Thanks @rliu6915 for raising the PR 🎉 I have some minor suggestions, wdyt?

[nth-zik]

Wait this PR

[rliu6915]

Hi, for Ollama js, thinking mode was introduced in v0.5.16. So I updated Ollama to 0.5.17 (the latest).

@hntrl @jacoblee93 @christian-bromann Could you please review the change that is awaiting your approval?

[rliu6915]

Hi, sorry, it failed this time because of yarn check. I yarned format, updated the code, yarned format:check and matched like below:

@hntrl @jacoblee93 @christian-bromann Could you please review the change that is awaiting your approval?

[rliu6915]

@hntrl

[hntrl]

@rliu6915 apologies 1000x for taking this long to get this landed. Thank you for your persistence and for the PR! 💪

stale