app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, StoppingCriteria, StoppingCriteriaList, TextIteratorStreamer
import time
import numpy as np
from torch.nn import functional as F
import os
from threading import Thread

print(f"Starting to load the model to memory")
m = AutoModelForCausalLM.from_pretrained(
    "stabilityai/stablelm-tuned-alpha-7b", torch_dtype=torch.float16).cuda()
tok = AutoTokenizer.from_pretrained("stabilityai/stablelm-tuned-alpha-7b")
generator = pipeline('text-generation', model=m, tokenizer=tok, device=0)
print(f"Sucessfully loaded the model to the memory")

start_message = """<|SYSTEM|># StableAssistant
- StableAssistant is A helpful and harmless Open Source AI Language Model developed by Stability and CarperAI.
- StableAssistant is excited to be able to help the user, but will refuse to do anything that could be considered harmful to the user.
- StableAssistant is more than just an information source, StableAssistant is also able to write poetry, short stories, and make jokes.
- StableAssistant will refuse to participate in anything that could harm a human."""


class StopOnTokens(StoppingCriteria):
    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
        stop_ids = [50278, 50279, 50277, 1, 0]
        for stop_id in stop_ids:
            if input_ids[0][-1] == stop_id:
                return True
        return False

def user(user_message, history):
    history = history + [[user_message, ""]]
    return "", history, history


def bot(history, curr_system_message):
    stop = StopOnTokens()
    messages = curr_system_message + \
        "".join(["".join(["<|USER|>"+item[0], "<|ASSISTANT|>"+item[1]])
                for item in history])
    
    #model_inputs = tok([messages], return_tensors="pt")['input_ids'].cuda()[:, :4096-1024]
    model_inputs = tok([messages], return_tensors="pt").to("cuda")
    streamer = TextIteratorStreamer(tok, timeout=10., skip_prompt=True, skip_special_tokens=True)
    generate_kwargs = dict(
        model_inputs,
        streamer=streamer,
        max_new_tokens=1024,
        do_sample=True,
        top_p=0.95,
        top_k=1000,
        temperature=1.0,
        num_beams=1,
        stopping_criteria=StoppingCriteriaList([stop])
    )
    t = Thread(target=m.generate, kwargs=generate_kwargs)
    t.start()

    print(history)
    for new_text in streamer:
        print(new_text)
        history[-1][1] += new_text
        yield history, history
        
    return history, history


with gr.Blocks() as demo:
    history = gr.State([])
    gr.Markdown("## StableLM-Tuned-Alpha-7b Chat")
    gr.HTML('''<center><a href="https://huggingface.co/spaces/stabilityai/stablelm-tuned-alpha-chat?duplicate=true"><img src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a>Duplicate the Space to skip the queue and run in a private space</center>''')
    chatbot = gr.Chatbot().style(height=500)
    with gr.Row():
        with gr.Column(scale=0.70):
            msg = gr.Textbox(label="Chat Message Box", placeholder="Chat Message Box", show_label=False).style(container=False)
        with gr.Column(scale=0.30):
          with gr.Row():
              submit = gr.Button("Submit")
              clear = gr.Button("Clear")    
    system_msg = gr.Textbox(
        start_message, label="System Message", interactive=False, visible=False)

    msg.submit(fn=user, inputs=[msg, history], outputs=[msg, chatbot, history], queue=False).then(
        fn=bot, inputs=[chatbot, system_msg], outputs=[chatbot, history], queue=True)
    submit.click(fn=user, inputs=[msg, history], outputs=[msg, chatbot, history], queue=False).then(
        fn=bot, inputs=[chatbot, system_msg], outputs=[chatbot, history], queue=True)
    clear.click(lambda: [None, []], None, [chatbot, history], queue=False)
demo.queue(concurrency_count=2)
demo.launch()