Update app.py

app.py

@@ -30,9 +30,9 @@ for ext in extensions:
 """
 For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
 """
-client = InferenceClient("microsoft/DialoGPT-small")#("TheBloke/TinyLlama-1.1B-Chat-v1.0-GPTQ")#("TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF")#("QuantFactory/Meta-Llama-3-8B-Instruct-GGUF")#("HuggingFaceH4/zephyr-7b-beta")
+client = InferenceClient("lachiewyoung/mistral-7b-instruct-1.58bit")#("TheBloke/TinyLlama-1.1B-Chat-v1.0-GPTQ")#("TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF")#("QuantFactory/Meta-Llama-3-8B-Instruct-GGUF")#("HuggingFaceH4/zephyr-7b-beta")
 
-SYSTEM_MESSAGE = "You are a QA chatbot to answer queries on my homepage that has the following information -\n\n" + "\n\n".join(text_list) + "\n\n<|endoftext|>"
+SYSTEM_MESSAGE = "You are a QA chatbot to answer queries on my homepage that has the following information -\n\n" + "\n\n".join(text_list) + "\n\n"
 
 def respond(
     message,
@@ -47,10 +47,10 @@ def respond(
     for val in history:
         # if val[0]:
         if len(val)>=1:
-            messages.append({"role": "user", "content": "Question: "+val[0]+"<|endoftext|>"})
+            messages.append({"role": "user", "content": "Question: "+val[0]})
         # if val[1]:
         if len(val)>=2:
-            messages.append({"role": "assistant", "content": "Answer: "+val[1]+"<|endoftext|>"})
+            messages.append({"role": "assistant", "content": "Answer: "+val[1]})
 
     messages.append({"role": "user", "content": message})