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/airoboros-l2-13b-gpt4-m2.0-GGML", filename="airoboros-l2-13b-gpt4-m2.0.ggmlv3.q6_K.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)
    print("last history: ", history[-1])
    full_conversation = f"{history} Q: {input_text} \n A:"
    print("full convo", full_conversation)
    temp_conversation = f"{history[-1]} Q: {input_text} \n A:"
    print("temp convo", temp_conversation)
    output = llm(full_conversation, max_tokens=1024, stop=["Q:", "\n"], echo=True)
    response = output['choices'][0]['text']
    return response
    

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