app.py

import gradio as gr
import time
import ctypes #to run on C api directly 
import llama_cpp
from llama_cpp import Llama
from huggingface_hub import hf_hub_download #load from huggingfaces 


llm = Llama(model_path= hf_hub_download(repo_id="TheBloke/Vigogne-2-7B-Instruct-GGML", filename="vigogne-2-7b-instruct.ggmlv3.q4_1.bin"), n_ctx=2048) #download model from hf/ n_ctx=2048 for high ccontext length
chat_history = []

def generate_text(message,history):

if len(history) > 0:
        user_input, bot_response = history[-1]  # Get the latest pair from history
        chat_history.append([user_input, message])
    else:
        chat_history.append([message, ""])  # If history is empty, just add the user input
   
input_text = message
    output = llm(f"Q: {input_text} \n A:", max_tokens=521, stop=["Q:", "\n"], echo=True)
    response = output['choices'][0]['text']

    # Append the bot response to the chat history
    chat_history[-1][1] = response

    return response
     

input_text = gr.inputs.Textbox(lines= 10, label="Enter your input text")
output_text = gr.outputs.Textbox(label="Output text")

description = " currently running ggml models with llama.cpp implementation in python [https://github.com/abetlen/llama-cpp-python]"

examples = [
    ["What is the capital of France? ", "The capital of France is Paris."],
    ["Who wrote the novel 'Pride and Prejudice'?", "The novel 'Pride and Prejudice' was written by Jane Austen."],
    ["What is the square root of 64?", "The square root of 64 is 8."]
]

demo = gr.ChatInterface(random_response).launch()
demo.queue()
demo.launch()