Update app.py

app.py

@@ -1,6 +1,4 @@
 import os
-import json
-import subprocess
 import sys
 from typing import List, Tuple
 from llama_cpp import Llama
@@ -11,9 +9,6 @@ from llama_cpp_agent.chat_history.messages import Roles
 from llama_cpp_agent.messages_formatter import MessagesFormatter, PromptMarkers
 from huggingface_hub import hf_hub_download
 import gradio as gr
-# from logger import logging
-# from exception import CustomExceptionHandling
-
 
 # Load the Environment Variables from .env file
 huggingface_token = os.getenv("HUGGINGFACE_TOKEN")
@@ -28,25 +23,22 @@ hf_hub_download(
     local_dir="./models",
 )
 
-
 # Define the prompt markers for Gemma 3
 gemma_3_prompt_markers = {
-    Roles.system: PromptMarkers("<start_of_turn>system\n", "<end_of_turn>\n"),  # System prompt should be included within user message
+    Roles.system: PromptMarkers("<start_of_turn>system\n", "<end_of_turn>\n"),
     Roles.user: PromptMarkers("<start_of_turn>user\n", "<end_of_turn>\n"),
     Roles.assistant: PromptMarkers("<start_of_turn>assistant", ""),
-    
-    Roles.tool: PromptMarkers("", ""),  # If you need tool support
+    Roles.tool: PromptMarkers("", ""),
 }
 
-# Create the formatter
 gemma_3_formatter = MessagesFormatter(
-    pre_prompt="",  # No pre-prompt
+    pre_prompt="",
     prompt_markers=gemma_3_prompt_markers,
-    include_sys_prompt_in_first_user_message=True,  # Include system prompt in first user message
+    include_sys_prompt_in_first_user_message=True,
     default_stop_sequences=["<end_of_turn>", "<start_of_turn>"],
-    strip_prompt=False,  # Don't strip whitespace from the prompt
-    bos_token="<bos>",  # Beginning of sequence token for Gemma 3
-    eos_token="<eos>",  # End of sequence token for Gemma 3
+    strip_prompt=False,
+    bos_token="<bos>",
+    eos_token="<eos>",
 )
 
 # Translation direction to prompts mapping
@@ -69,59 +61,46 @@ direction_to_prompts = {
     }
 }
 
-# Set the title and description
-title = "Kazakh Language Model"
-description = """"""
-
-
 llm = None
 llm_model = None
 
 def respond(
     message: str,
     history: List[Tuple[str, str]],
-    model: str,
+    model: str = "gemma_3_800M_sft_v2_translation-kazparc_latest.gguf",
     direction: str,
-    max_tokens: int = 64,
+    max_tokens: int = 1024,
     temperature: float = 0.7,
     top_p: float = 0.95,
     top_k: int = 40,
     repeat_penalty: float = 1.1,
 ):
     """
-    Respond to a message using the Gemma3 model via Llama.cpp.
+    Respond to a message by translating it using the specified direction.
+    
     Args:
-        - message (str): The message to respond to.
-        - history (List[Tuple[str, str]]): The chat history.
-        - model (str): The model to use.
-        - system_message (str): The system message to use.
-        - max_tokens (int): The maximum number of tokens to generate.
-        - temperature (float): The temperature of the model.
-        - top_p (float): The top-p of the model.
-        - top_k (int): The top-k of the model.
-        - repeat_penalty (float): The repetition penalty of the model.
-    Returns:
-        str: The response to the message.
+        message (str): The text to translate.
+        history (List[Tuple[str, str]]): The chat history.
+        direction (str): The translation direction (e.g., "English to Kazakh").
+        model (str): The model file to use.
+        max_tokens (int): Maximum number of tokens to generate.
+        temperature (float): Sampling temperature.
+        top_p (float): Top-p sampling parameter.
+        top_k (int): Top-k sampling parameter.
+        repeat_penalty (float): Penalty for repetition.
+    
+    Yields:
+        str: The translated text as it is generated.
     """
-    # try:
-    # Load the global variables
-    global llm
-    global llm_model
-
-    # Ensure model is not None
-    if model is None:
-        model = "gemma_3_800M_sft_v2_translation-kazparc_latest.gguf"
-
-    # Load the model
+    
+    global llm, llm_model
     if llm is None or llm_model != model:
-        # Check if model file exists
         model_path = f"models/{model}"
         if not os.path.exists(model_path):
-            yield f"Error: Model file not found at {model_path}. Please check your model path."
+            yield f"Error: Model file not found at {model_path}."
             return
-
         llm = Llama(
-            model_path=f"models/{model}",
+            model_path=model_path,
             flash_attn=False,
             n_gpu_layers=0,
             n_batch=8,
@@ -132,15 +111,18 @@ def respond(
         llm_model = model
     provider = LlamaCppPythonProvider(llm)
 
-    # Create the agent
+    # Get system prompt and user prefix based on direction
+    prompts = direction_to_prompts[direction]
+    system_message = prompts["system"]
+    user_prefix = prompts["prefix"]
+
     agent = LlamaCppAgent(
         provider,
-        #system_prompt=f"{system_message}",
+        system_prompt=system_message,
         custom_messages_formatter=gemma_3_formatter,
         debug_output=True,
     )
 
-    # Set the settings like temperature, top-k, top-p, max tokens, etc.
     settings = provider.get_provider_default_settings()
     settings.temperature = temperature
     settings.top_k = top_k
@@ -150,45 +132,31 @@ def respond(
     settings.stream = True
 
     messages = BasicChatHistory()
+    for user_msg, assistant_msg in history:
+        full_user_msg = user_prefix + " " + user_msg
+        messages.add_message({"role": Roles.user, "content": full_user_msg})
+        messages.add_message({"role": Roles.assistant, "content": assistant_msg})
 
-    # Add the chat history
-    for msn in history:
-        user = {"role": Roles.user, "content": msn[0]}
-        assistant = {"role": Roles.assistant, "content": msn[1]}
-        messages.add_message(user)
-        messages.add_message(assistant)
+    full_message = user_prefix + " " + message
 
-    # Get the response stream
     stream = agent.get_chat_response(
-        message,
+        full_message,
         llm_sampling_settings=settings,
         chat_history=messages,
         returns_streaming_generator=True,
         print_output=False,
     )
 
-    # Log the success
-    # logging.info("Response stream generated successfully")
-
-    # Generate the response
     outputs = ""
     for output in stream:
         outputs += output
         yield outputs
 
-    # # Handle exceptions that may occur during the process
-    # except Exception as e:
-    #     # Custom exception handling
-    #     raise CustomExceptionHandling(e, sys) from e
-
 
-# Create a chat interface
 demo = gr.ChatInterface(
     respond,
-    examples=[["Сәлем"], ["Привет"], ["Hello"]],
-    additional_inputs_accordion=gr.Accordion(
-        label="⚙️ Parameters", open=False, render=False
-    ),
+    examples=[["Hello"], ["Сәлем"], ["Привет"]],
+    additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False, render=False),
     additional_inputs=[
         gr.Dropdown(
             choices=[
@@ -209,7 +177,7 @@ demo = gr.ChatInterface(
             value=1024,
             step=1,
             label="Max Tokens",
-            info="Maximum length of response (higher = longer replies)",
+            info="Maximum length of the translation"
         ),
         gr.Slider(
             minimum=0.1,
@@ -217,7 +185,7 @@ demo = gr.ChatInterface(
             value=0.7,
             step=0.1,
             label="Temperature",
-            info="Creativity level (higher = more creative, lower = more focused)",
+            info="Controls randomness (higher = more creative)"
         ),
         gr.Slider(
             minimum=0.1,
@@ -225,7 +193,7 @@ demo = gr.ChatInterface(
             value=0.95,
             step=0.05,
             label="Top-p",
-            info="Nucleus sampling threshold",
+            info="Nucleus sampling threshold"
         ),
         gr.Slider(
             minimum=1,
@@ -233,7 +201,7 @@ demo = gr.ChatInterface(
             value=40,
             step=1,
             label="Top-k",
-            info="Limit vocabulary choices to top K tokens",
+            info="Limits vocabulary to top K tokens"
         ),
         gr.Slider(
             minimum=1.0,
@@ -241,24 +209,22 @@ demo = gr.ChatInterface(
             value=1.1,
             step=0.1,
             label="Repetition Penalty",
-            info="Penalize repeated words (higher = less repetition)",
+            info="Penalizes repeated words"
         ),
     ],
     theme="Ocean",
-    submit_btn="Send",
+    submit_btn="Translate",
     stop_btn="Stop",
-    title=title,
-    description=description,
+    title="Kazakh Translation Model",
+    description="Translate text between Kazakh, English, and Russian using a specialized language model.",
     chatbot=gr.Chatbot(scale=1, show_copy_button=True),
     cache_examples=False,
 )
 
-
-# Launch the chat interface
 if __name__ == "__main__":
     demo.launch(
         share=False,
         server_name="0.0.0.0",
         server_port=7860,
         show_api=False,
-    )
+    )