Update chain_app.py

chain_app.py

@@ -2503,10 +2503,7 @@ async def main(message: cl.Message):
     elif chat_profile == "zephyr-7B":
         result = hf_text_client.predict(
             message=message.content,
-            request=f"You are neural nexus official chatbot, you are made by Artin Daneshvar and Sadra Noadoust and you are here to help people",
-            param_3=512,
-            param_4=0.7,
-            param_5=0.95,
+            system_message=f"You are neural nexus official chatbot, you are made by Artin Daneshvar and Sadra Noadoust and you are here to help people",
             api_name="/chat"
         )
         for token in result:
@@ -2528,15 +2525,12 @@ async def main(message: cl.Message):
             stop=None,
         )
 
-        complete_content = ""
-
         for chunk in completion:
-            content = chunk.choices[0].delta.content
-            
-            if content is not None:
-                complete_content += content
+            # Retrieve the content from the current chunk            
+            # Check if the content is not None before concatenating it
+            if chunk is not None:
+                await msg.stream_token(chunk.choices[0].delta.content)
 
-        await cl.Message(content=complete_content).send() 
 
     elif chat_profile == 'mistral-nemo-12B':
         client = Client("0x7o/Mistral-Nemo-Instruct", hf_token=hf_token)
@@ -2547,24 +2541,21 @@ async def main(message: cl.Message):
                 top_p=0.95,
                 api_name="/chat"
         )
-        await cl.Message(
-            content=result[1][0][1]
-
-        ).send()
+        for i in list(result[1][0][1]):
+            await msg.stream_token(i)
 
     elif chat_profile == 'mistral-7B-v2':
         client = InferenceClient(
             "mistralai/Mistral-7B-Instruct-v0.2",
             token=f"{hf_token_llama_3_1}",
         )
-        result = client.chat_completion(
+        for res in client.chat_completion(
         	messages=[{"role": "user", "content": message.content}],
             max_tokens=500,
-        )
-
-        await cl.Message(
-            content=result.choices[0].message.content,
-        ).send()
+            stream=True,
+        ):
+            if res.choices[0].delta.content is not None:
+                await msg.stream_token(res.choices[0].delta.conten)
 
     elif chat_profile == 'Yi-1.5-34B':
         final_answer = await cl.Message(content="").send()