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-Chat-GGML", filename="vigogne-2-7b-chat.ggmlv3.q4_1.bin"), n_ctx=2048) #download model from hf/ n_ctx=2048 for high ccontext length

history = []

def generate_text(input_text, history):
    print("history ",history)
    print("input ", input_text)
    print("response", response)
    if history == []:
        input_text_with_history = f"Q: {input_text} \n A:"
    else:
        input_text_with_history = history[-1][1]+ "\n"
        input_text_with_history += f"Q: {input_text} \n A:"
    print("new input", input_text_with_history)
    output = llm(input_text_with_history, max_tokens=1024, stop=["Q:", "\n"], echo=True)
    response = output['choices'][0]['text'] + "\n"
    history =["init",input_text_with_history]
    return response
    

demo = gr.ChatInterface(generate_text)
demo.queue(concurrency_count=1, max_size=5)
demo.launch()