update

multipurpose_chatbot/demos/text_completion.py

@@ -87,7 +87,7 @@ def generate_text_completion_stream_engine(
     stop_strings = [x.strip() for x in stop_strings.strip().split(",")]
     stop_strings = list(set(stop_strings + ['</s>', '<|im_start|>', '<|im_end|>']))
     if message.strip() != message:
-        gr.Warning(f'There are preceding/trailing spaces in the message, may lead to unexpected behavior')
+        gr.Warning(f'There are preceding/trailing spaces in the message.')
     if len(message) == 0:
         raise gr.Error("The message cannot be empty!")
     num_tokens = len(MODEL_ENGINE.tokenizer.encode(message))
@@ -131,7 +131,7 @@ class TextCompletionDemo(BaseDemo):
         model_name = kwargs.get("model_name", MODEL_NAME)
         # frequence_penalty = FREQUENCE_PENALTY
         # presence_penalty = PRESENCE_PENALTY
-        max_tokens = max_tokens // 2
+        max_tokens = max_tokens // 4
 
         description = description or f"""Put any context string (like few-shot prompts)"""
 
@@ -166,7 +166,8 @@ class TextCompletionDemo(BaseDemo):
                 stop_strings = gr.Textbox(value="<eos>,<s>,</s>,<|im_start|>,<|im_end|>", label='Stop strings', info='Comma-separated string to stop generation only in FEW-SHOT mode', lines=1)
             examples = gr.Examples(
                 examples=[
-                    ["The following is the recite the declaration of independence:",]
+                    ["The following is the recite the declaration of independence:",],
+                    ["<|im_start|>system\nYou are a helpful assistant.<eos>\n<|im_start|>user\nTell me a joke.<eos>\n<|im_start|>assistant\n",]
                 ],
                 inputs=[txt, temp_input, length_input, stop_strings],
                 # outputs=[txt]

multipurpose_chatbot/engines/transformers_engine.py

@@ -216,7 +216,6 @@ class NewGenerationMixin(GenerationMixin):
         ['Today is a beautiful day, and we must do everything possible to make it a day of celebration.']
         ```"""
         # init values
-        print(f'Streaming tokens...')
         from transformers.generation.utils import (
             validate_stopping_criteria, GenerateEncoderDecoderOutput, GenerateDecoderOnlyOutput
         )