Update main.py

main.py

@@ -47,31 +47,14 @@ def generate(prompt, history, temperature=0.9, max_new_tokens=512, top_p=0.95, r
 
     formatted_prompt = format_prompt(prompt, history)
 
-    bot_response = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
+    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
     output = ""
-    current_word = ""
 
-    for response in bot_response:
-        token_text = response.token.text.strip()
-        if token_text != "":
-            # Decode the token text to handle encoded characters
-            decoded_text = token_text.encode("utf-8", "backslashreplace").decode("utf-8")
+    for response in stream:
+        output += response.token.text
 
-            # Add the decoded token text to the current word
-            current_word += decoded_text
+    return output
 
-            # Check if the decoded token text ends with a space or punctuation mark
-            if decoded_text[-1] in " .,?!":
-                # Add the current word to the output with a space delimiter
-                output += current_word
-                output += " "
-                current_word = ""
-
-    # Add the last word to the output if it exists
-    if current_word:
-        output += current_word
-
-    return output.strip()