Update app.py

app.py

@@ -60,21 +60,20 @@ client = InferenceClient(HF_MODEL_ID, token=os.getenv("HF_TOKEN"))
 import os
 from langchain_huggingface import HuggingFaceEndpoint, ChatHuggingFace
 
-hf_token = os.getenv("HF_TOKEN")
+# 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
+repo_id = "meta-llama/Llama-3.1-8B-Instruct"
 
-# 1. Create the endpoint with the 'conversational' task
 llm_endpoint = HuggingFaceEndpoint(
-    #repo_id="mistralai/Mistral-7B-Instruct-v0.3",
-    repo_id = "HuggingFaceH4/zephyr-7b-beta",
+    repo_id=repo_id,
     task="text-generation",
     max_new_tokens=512,
-    huggingfacehub_api_token=hf_token,
+    huggingfacehub_api_token=os.getenv("HF_TOKEN"),
 )
 
-# 2. Wrap it in ChatHuggingFace
-# This is REQUIRED for conversational tasks to work with RAG chains
+# 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)
-
 # -------------------------
 # Document loader
 # -------------------------