Update app.py

app.py

@@ -4,7 +4,7 @@ warnings.filterwarnings("ignore")
 import uuid
 import gradio as gr
 import os
-from langchain_huggingface import HuggingFaceEndpoint
+from langchain_huggingface import HuggingFaceEndpoint, ChatHuggingFace
 
 from huggingface_hub import InferenceClient
 from langchain_text_splitters import RecursiveCharacterTextSplitter
@@ -42,18 +42,28 @@ client = InferenceClient(HF_MODEL_ID, token=os.getenv("HF_TOKEN"))
 #         return response
 #     return _llm
 
-llm = HuggingFaceEndpoint(
+# llm = HuggingFaceEndpoint(
+#     repo_id="mistralai/Mistral-7B-Instruct-v0.3",
+#     # repo_id="meta-llama/Llama-3.1-8B-Instruct", # This is the "Live" 2026 model
+#     # repo_id="ibm-granite/granite-3.1-8b-instruct", # Using Granite for 2026 stability
+#     task="conversational",
+#     max_new_tokens=512,
+#     temperature=0.1,
+#     huggingfacehub_api_token=os.getenv("HF_TOKEN"),
+#     wait_for_model=True 
+# )
+
+# 1. First, define the base endpoint (keep task as text-generation here)
+llm_endpoint = HuggingFaceEndpoint(
     repo_id="mistralai/Mistral-7B-Instruct-v0.3",
-    # repo_id="meta-llama/Llama-3.1-8B-Instruct", # This is the "Live" 2026 model
-    # repo_id="ibm-granite/granite-3.1-8b-instruct", # Using Granite for 2026 stability
-    task="conversational",
+    task="text-generation", 
     max_new_tokens=512,
-    temperature=0.1,
     huggingfacehub_api_token=os.getenv("HF_TOKEN"),
-    wait_for_model=True 
 )
 
-
+# 2. Wrap it in ChatHuggingFace to fix the task mismatch
+# This automatically handles the "conversational" formatting for you
+llm = ChatHuggingFace(llm=llm_endpoint)
 
 # -------------------------
 # Document loader