fix(langchain): correctly initialize huggingface models in init_chat_model

[aaron-seq]

Description:

This change fixes a ValidationError that occurred when using init_chat_model with model_provider="huggingface".

The previous implementation incorrectly passed all keyword arguments directly to the ChatHuggingFace constructor, which expects an llm object rather than a model_id. This caused a Pydantic error when arguments like task were provided.

The fix updates the logic to first correctly instantiate a HuggingFacePipeline object using the from_model_id class method, which properly handles arguments like task. This initialized pipeline object is then passed to the ChatHuggingFace wrapper, resolving the issue.

A unit test has been added to verify the fix and prevent regressions.

Closes #28226

Issue:

Fixes #28226

Dependencies:

None

[vercel]

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

1 Skipped Deployment

Project	Deployment	Preview	Comments	Updated (UTC)
langchain	Ignored	Preview		Oct 1, 2025 3:13pm

[eyurtsev]

@aaronseq12 thanks for the contribution -- there seem to be a lot of typos in this PR, do you want to try and resolve?

[aaron-seq]

@eyurtsev yup got some time free, will be working on this!
thanks

[aaron-seq]

Linting Issues Fixed

I've resolved the Ruff linting errors that were preventing this PR from passing checks:

Issues Resolved:

EM101 Rule Violations:

• Moved exception messages to variables instead of using string literals directly in raise statements
• Applied this fix to both ImportError and ValueError exceptions in the HuggingFace initialization code

TRY003 Rule Violations:

• Shortened exception messages to be more concise and avoid "long messages outside the exception class"
• Made error messages more specific and actionable

Changes Made:

1. ImportError handling:

   # Before: raise ImportError("Please install langchain-huggingface...")
   # After:
   import_error_msg = "Please install langchain-huggingface to use HuggingFace models."
   raise ImportError(import_error_msg) from e

2. ValueError handling:

   # Before: raise ValueError("The 'task' keyword argument is required...")
   # After:
   task_error_msg = "The 'task' keyword argument is required for HuggingFace models."
   raise ValueError(task_error_msg)

The PR should now pass all linting checks while maintaining the same functional behavior for HuggingFace model initialization.