app.py

import os
import wget
import gradio as gr
from llama_cpp import Llama
import random
url =  'https://huggingface.co/TheBloke/Nous-Hermes-13B-GGML/resolve/main/nous-hermes-13b.ggmlv3.q2_K.bin'
filename = wget.download(url)
llm = Llama(model_path=filename, seed=random.randint(1, 2**31))

with gr.Blocks() as demo:
    chatbot = gr.Chatbot()
    msg = gr.Textbox()
    clear = gr.ClearButton([msg, chatbot])
    #instruction = gr.Textbox(label="Instruction", placeholder=)

    def user(user_message, history):
        return gr.update(value="", interactive=True), history + [[user_message, None]]

    def bot(history):
        #instruction = history[-1][1] or ""
        user_message = history[-1][0]
        #token1 = llm.tokenize(b"### Instruction: ")
        #token2 = llm.tokenize(instruction.encode())
        token3 = llm.tokenize(b"### Input: ")
        tokens3 = llm.tokenize(user_message.encode())
        token4 = llm.tokenize(b"### Response:")
        tokens = token3 + tokens3 + token4
        history[-1][1] = ""
        count = 0
        output = ""
        for token in llm.generate(tokens, top_k=50, top_p=0.73, temp=0.72, repeat_penalty=1.1):
            text = llm.detokenize([token])
            output += text.decode()
            count += 1
            if count >= 500 or (token == llm.token_eos()):
                break
            history[-1][1] += text.decode()
            yield history

    response = msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
        bot, chatbot, chatbot
    )
    response.then(lambda: gr.update(interactive=True), None, [msg], queue=False)

demo.queue()
demo.launch(debug=True)