Description:

This PR fixes a bug where init_chat_model fails when used with Hugging Face models. The issue was that ChatHuggingFace requires an llm object (such as HuggingFaceEndpoint), but the code was attempting to pass model_id directly.

The fix:

• Creates a HuggingFaceEndpoint instance using the model as repo_id
• Properly separates ChatHuggingFace-specific parameters (tokenizer, system_message, custom_get_token_ids, verbose, metadata, tags) from HuggingFaceEndpoint parameters (temperature, max_new_tokens, etc.)
• Passes the HuggingFaceEndpoint instance to ChatHuggingFace as the required llm parameter

This change ensures that init_chat_model("huggingface:microsoft/Phi-3-mini-4k-instruct", temperature=0.7) works correctly by creating the appropriate underlying LLM wrapper that ChatHuggingFace expects.

Issue:

N/A (bug fix without associated issue)

Dependencies:

No new dependencies. This fix uses existing langchain_huggingface package imports.