tabbed.py

import gradio as gr
import yaml
from huggingface_hub import hf_hub_download
from huggingface_hub.utils import LocalEntryNotFoundError
from llama_cpp import Llama

with open("./config.yml", "r") as f:
    config = yaml.load(f, Loader=yaml.Loader)
while True:
    try:
        load_config = config.copy()
        hub_config = load_config["hub"].copy()
        repo_id = hub_config.pop("repo_id")
        filename = hub_config.pop("filename")
        fp = hf_hub_download(
            repo_id=repo_id, filename=filename, **hub_config
        )
        break
    except LocalEntryNotFoundError as e:
        if "Connection error" in str(e):
            print(str(e) + ", retrying...")
        else:
            raise(e)

llm = Llama(model_path=fp, **config["llama_cpp"])


def user(message, history):
    history = history or []
    # Append the user's message to the conversation history
    history.append([message, ""])
    return "", history

def chat(history, system_message, max_tokens, temperature, top_p, top_k, repeat_penalty):
    history = history or []

    messages = system_message.strip() + "\n" + \
               "\n".join(["\n".join(["Kullanıcı: "+item[0], "Asistan: "+item[1]])
                        for item in history])

    # remove last space from assistant, some models output a ZWSP if you leave a space
    messages = messages[:-1]

    history[-1][1] = ""
    for output in llm(
            messages,
            echo=False,
            stream=True,
            max_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
            top_k=top_k,
            repeat_penalty=repeat_penalty,
            **config['chat']
    ):
        answer = output['choices'][0]['text']
        history[-1][1] += answer
        # stream the response
        yield history, history

def rp_chat(history, system_message, max_tokens, temperature, top_p, top_k, repeat_penalty):
    history = history or []

    messages = "<|system|>" + system_message.strip() + "\n" + \
               "\n".join(["\n".join(["<|user|>"+item[0], "<|model|>"+item[1]])
                        for item in history])

    # remove last space from assistant, some models output a ZWSP if you leave a space
    messages = messages[:-1]

    history[-1][1] = ""
    for output in llm(
            messages,
            echo=False,
            stream=True,
            max_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
            top_k=top_k,
            repeat_penalty=repeat_penalty,
            **config['chat']
    ):
        answer = output['choices'][0]['text']
        history[-1][1] += answer
        # stream the response
        yield history, history


def clear_chat(chat_history_state, chat_message):
    chat_history_state = []
    chat_message = ''
    return chat_history_state, chat_message


start_message = """
- Akıllı, dürüst ve yardımsever bir asistansın.
- Her türlü soruya dürüstçe cevap vereceksin.
"""

def generate_text_instruct(input_text):
    response = ""
    for output in llm(f"Kullanıcı: {input_text}\nAsistan:",  echo=False, stream=True, **config['chat']):
        answer = output['choices'][0]['text']
        response += answer
        yield response


instruct_interface = gr.Interface(
    fn=generate_text_instruct,
    inputs=gr.inputs.Textbox(lines= 10, label="Enter your input text"),
    outputs=gr.outputs.Textbox(label="Output text"),
)


with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column():
            gr.Markdown(f"""
                    ### This is the [{config["hub"]["repo_id"]}](https://huggingface.co./{config["hub"]["repo_id"]}) quantized model file [{config["hub"]["filename"]}](https://huggingface.co./{config["hub"]["repo_id"]}/blob/main/{config["hub"]["filename"]})
    
                    <details>
                        <summary><a href="https://huggingface.co./spaces/Nekochu/Luminia-13B-v3-GGUF?duplicate=true">Duplicate the Space</a> to skip the queue and run in a private space or to use your own GGUF models, simply update the <a href="https://huggingface.co./spaces/Nekochu/Luminia-13B-v3-GGUF/blob/main/config.yml">config.yml</a></summary>
                        <ul>
                            <li>This Space uses GGUF with CPU-<strong>FREE</strong>, GPU support, so it can quickly run larger models on smaller GPUs & VRAM. <a href="https://github.com/OpenAccess-AI-Collective/ggml-webui">[Contribute]</a></li>
                            <li>This is running on a smaller, shared GPU, so it may take a few seconds to respond.</li>
                        </ul>
                    </details>
                    """)
    with gr.Tab("Chatbot"):
        gr.Markdown("# GGUF Spaces Chatbot Demo")
        chatbot = gr.Chatbot()
        with gr.Row():
            message = gr.Textbox(
                label="Ne konuda konuşmak istersin?",
                placeholder="Bana bir şeyler sor.",
                lines=3,
            )
        with gr.Row():
            submit = gr.Button(value="Mesaj Gönder", variant="secondary").style(full_width=True)
            roleplay = gr.Button(value="Rol", variant="secondary").style(full_width=True)
            clear = gr.Button(value="Yeni Konu", variant="secondary").style(full_width=False)
            stop = gr.Button(value="Dur", variant="secondary").style(full_width=False)
        with gr.Row():
            with gr.Column():
                max_tokens = gr.Slider(20, 1000, label="Max Tokens", step=20, value=300)
                temperature = gr.Slider(0.2, 2.0, label="Temperature", step=0.1, value=0.8)
                top_p = gr.Slider(0.0, 1.0, label="Top P", step=0.05, value=0.95)
                top_k = gr.Slider(0, 100, label="Top K", step=1, value=40)
                repeat_penalty = gr.Slider(0.0, 2.0, label="Repetition Penalty", step=0.1, value=1.1)

        system_msg = gr.Textbox(
            start_message, label="System Message", interactive=True, visible=True, placeholder="system prompt, useful for RP", lines=5)

        chat_history_state = gr.State()
        clear.click(clear_chat, inputs=[chat_history_state, message], outputs=[chat_history_state, message], queue=False)
        clear.click(lambda: None, None, chatbot, queue=False)

        submit_click_event = submit.click(
            fn=user, inputs=[message, chat_history_state], outputs=[message, chat_history_state], queue=True
        ).then(
            fn=chat, inputs=[chat_history_state, system_msg, max_tokens, temperature, top_p, top_k, repeat_penalty], outputs=[chatbot, chat_history_state], queue=True
        )
        roleplay_click_event = roleplay.click(
            fn=user, inputs=[message, chat_history_state], outputs=[message, chat_history_state], queue=True
        ).then(
            fn=rp_chat, inputs=[chat_history_state, system_msg, max_tokens, temperature, top_p, top_k, repeat_penalty], outputs=[chatbot, chat_history_state], queue=True
        )
        # message_submit_event = message.submit(
        #     fn=user, inputs=[message, chat_history_state], outputs=[message, chat_history_state], queue=True
        # ).then(
        #     fn=chat, inputs=[chat_history_state, system_msg, max_tokens, temperature, top_p, top_k, repeat_penalty], outputs=[chatbot, chat_history_state], queue=True
        # )
        stop.click(fn=None, inputs=None, outputs=None, cancels=[submit_click_event, roleplay_click_event], queue=False)
    with gr.Tab("Instruct"):
        gr.Markdown("# GGUF Spaces Instruct Demo")
        instruct_interface.render()

demo.queue(**config["queue"]).launch(debug=True, server_name="0.0.0.0", server_port=7860)