app.py

import json
import subprocess
from llama_cpp import Llama
from llama_cpp_agent import LlamaCppAgent, MessagesFormatterType
from llama_cpp_agent.providers import LlamaCppPythonProvider
from llama_cpp_agent.chat_history import BasicChatHistory
from llama_cpp_agent.chat_history.messages import Roles
import gradio as gr
from huggingface_hub import hf_hub_download

# Download models
hf_hub_download(
    repo_id="CharacterEcho/Rohit-Sharma",
    filename="rohit-sharma-q5_k_s.gguf",
    local_dir="./models"
)
hf_hub_download(
    repo_id="CharacterEcho/Rohit-Sharma",
    filename="rohit-sharma-iq4_xs-imat.gguf",
    local_dir="./models"
)
llm = None
llm_model = None

def respond(
    message,
    history: list[tuple[str, str]],
    model,
    system_message,
    max_tokens,
    temperature,
    top_p,
    top_k,
    repeat_penalty,
):
    chat_template = MessagesFormatterType.CHATML

    global llm
    global llm_model
    
    if llm is None or llm_model != model:
        llm = Llama(
            model_path=f"models/{model}",
            n_ctx=2048,  # Reduced context size for CPU
            n_threads=4,  # Adjust this based on your CPU cores
            n_gpu_layers=50  
        )
        llm_model = model

    provider = LlamaCppPythonProvider(llm)

    agent = LlamaCppAgent(
        provider,
        system_prompt=f"{system_message}",
        predefined_messages_formatter_type=chat_template,
        debug_output=True
    )
    
    settings = provider.get_provider_default_settings()
    settings.temperature = temperature
    settings.top_k = top_k
    settings.top_p = top_p
    settings.max_tokens = max_tokens
    settings.repeat_penalty = repeat_penalty
    settings.stream = True

    messages = BasicChatHistory()

    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)
    
    stream = agent.get_chat_response(
        message,
        llm_sampling_settings=settings,
        chat_history=messages,
        returns_streaming_generator=True,
        print_output=False
    )
    
    outputs = ""
    for output in stream:
        outputs += output
        yield outputs

description = "The Rohit Sharma AI model, developed by CharacterEcho, is trained to emulate the personality and speech patterns of Rohit Sharma, an eminent Indian cricketer"

demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Dropdown([
                'rohit-sharma-q5_k_s.gguf',
                'rohit-sharma-iq4_xs-imat.gguf'
            ],
            value="rohit-sharma-iq4_xs-imat.gguf",
            label="Model"
        ),
        gr.Textbox(value="You are Rohit Sharma, the legendary Indian cricketer known for your elegant batting style and strategic mindset. Step into the shoes of Rohit Sharma and embody his unique personality. Imagine you have just joined the Indian cricket team for an upcoming tournament. Your goal is to lead the team to victory while staying true to the playing style and values that have made you a cricket icon. Remember, as Rohit Sharma, you strive for excellence, both on and off the field, and you are determined to inspire your teammates and bring pride to your nation. Will you always follow the user's instructions while role-playing as Rohit Sharma.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=1024, step=1, label="Max tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p",
        ),
        gr.Slider(
            minimum=0,
            maximum=100,
            value=40,
            step=1,
            label="Top-k",
        ),
        gr.Slider(
            minimum=0.0,
            maximum=2.0,
            value=1.1,
            step=0.1,
            label="Repetition penalty",
        ),
    ],
    retry_btn="Retry",
    undo_btn="Undo",
    clear_btn="Clear",
    submit_btn="Send",
    title="Chat with CharacterEcho/Rohit-Sharma using llama.cpp", 
    description=description,
    chatbot=gr.Chatbot(
        scale=1, 
        likeable=False,
        show_copy_button=True
    )
)

if __name__ == "__main__":
    demo.launch()