Update app.py

app.py

@@ -53,12 +53,6 @@ client = InferenceClient(HF_MODEL_ID, token=os.getenv("HF_TOKEN"))
 #     wait_for_model=True 
 # )
 
-# 1. Configure the endpoint
-# We use 'text-generation' because that's what the LangChain client expects
-# Use the dedicated Chat constructor
-# This handles the model_id and the conversational task mapping automatically
-import os
-from langchain_huggingface import HuggingFaceEndpoint, ChatHuggingFace
 
 # 1. Use a model that has high availability in the 2026 Provider network
 # Llama-3.1-8B-Instruct is currently a 'gold standard' for this router
@@ -68,16 +62,19 @@ repo_id = "Qwen/Qwen2.5-7B-Instruct"
 # repo_id = "mistralai/Mistral-7B-Instruct-v0.3"
 # repo_id = "HuggingFaceTB/SmolLM2-1.7B-Instruct"
 # repo_id = "microsoft/Phi-3.5-mini-instruct"
-
-
-from langchain_huggingface import ChatHuggingFace
-
-llm = ChatHuggingFace(
-    model=repo_id,
+llm_endpoint = HuggingFaceEndpoint(
+    repo_id=repo_id,
+    task="text-generation",
     max_new_tokens=512,
     temperature=0.1,
     huggingfacehub_api_token=os.getenv("HF_TOKEN"),
+    provider="hf-inference",
+    model_kwargs={"wait_for_model": True},
 )
+
+llm = ChatHuggingFace(llm=llm_endpoint)
+
+
 # 2. ChatHuggingFace will correctly route this to the /v1/chat/completions 
 # endpoint which the Llama-3.1-8B-Instruct model supports.
 # llm = ChatHuggingFace(llm=llm_endpoint)