Update app.py

app.py

@@ -57,24 +57,21 @@ client = InferenceClient(HF_MODEL_ID, token=os.getenv("HF_TOKEN"))
 # 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
-llm = ChatHuggingFace.from_model_id(
-    model_id="mistralai/Mistral-7B-Instruct-v0.3",
-    model_kwargs={
-        "max_new_tokens": 512,
-        "huggingfacehub_api_token": os.getenv("HF_TOKEN"),
-    },
-    # This tells the server we are sending a Chat request
-    # but the local wrapper knows it's an "Instruct" model.
-    task="text-generation", 
-)
+import os
+from langchain_huggingface import HuggingFaceEndpoint, ChatHuggingFace
 
-# 2. IMPORTANT: Manually set the model_id before wrapping
-# This bypasses the automated router check that's causing your ValueError
-llm_endpoint.model_id = "mistralai/Mistral-7B-Instruct-v0.3"
+# 1. Define the endpoint with text-generation task
+# This is the "engine" that talks to the Hugging Face servers
+http_llm = HuggingFaceEndpoint(
+    repo_id="mistralai/Mistral-7B-Instruct-v0.3",
+    task="text-generation",
+    max_new_tokens=512,
+    huggingfacehub_api_token=os.getenv("HF_TOKEN"),
+)
 
-# 3. Wrap in ChatHuggingFace
-# This will now handle the chat formatting without the router rejecting the task
-llm = ChatHuggingFace(llm=llm_endpoint)
+# 2. Wrap it manually in ChatHuggingFace
+# We pass it as 'llm' to the constructor
+llm = ChatHuggingFace(llm=http_llm)
 
 # -------------------------
 # Document loader