Update app.py

app.py

@@ -24,7 +24,7 @@ except Exception as e:
     tokenizer = None
     peft_model = None
 
-async def respond(message, history, system_message, max_tokens, temperature, top_p):
+def respond(message, history, system_message, max_tokens, temperature, top_p):
     """
     Generates a response based on the user message and history using the provided PEFT model.
     Args:
@@ -34,13 +34,12 @@ async def respond(message, history, system_message, max_tokens, temperature, top
         max_tokens (int): The maximum number of tokens to generate.
         temperature (float): The temperature parameter for generation.
         top_p (float): The top_p parameter for nucleus sampling.
-    Yields:
-        str: The generated response up to the current token.
+    Returns:
+        str: The generated response.
     """
     global tokenizer, peft_model  # Access global variables
     if tokenizer is None or peft_model is None:
-        yield "Model loading failed. Please check the logs."
-        return
+        return "Model loading failed. Please check the logs."
 
     # Construct the prompt
     prompt = system_message
@@ -56,7 +55,7 @@ async def respond(message, history, system_message, max_tokens, temperature, top
 
     # Generate the output
     try:
-        outputs = await peft_model.generate(
+        outputs = peft_model.generate(
             **inputs,
             max_new_tokens=max_tokens,
             temperature=temperature,
@@ -64,8 +63,7 @@ async def respond(message, history, system_message, max_tokens, temperature, top
             do_sample=True  # Enable sampling for more diverse outputs
         )
     except Exception as e:
-        yield f"Generation error: {e}"
-        return
+        return f"Generation error: {e}"
 
     # Decode the generated tokens
     generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
@@ -78,17 +76,13 @@ async def respond(message, history, system_message, max_tokens, temperature, top
         """
         pattern = re.compile(r'<user>(.*?)</user>|output:', re.IGNORECASE)
         matches = re.findall(pattern, text, re.DOTALL)
-        extracted_content = '\n'.join(match.strip() for match in matches)
+        extracted_content = '\n'.join(match.strip() for match in matches if match)
         return extracted_content
 
     # Extract the normalized text
     normalized_text = extract_user_content(generated_text)
 
-    # Stream the response token by token
-    response = ""
-    for token in normalized_text.split():
-        response += token + " "
-        yield response.strip()
+    return normalized_text
 
 # Gradio interface setup
 demo = gr.ChatInterface(