app.py

import os
import time
#import spaces
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
import gradio as gr
from threading import Thread

MODEL_LIST = ["GoidaAlignment/GOIDA-0.5B"]
HF_TOKEN = os.environ.get("HF_TOKEN", None)

TITLE = "<h1><center>Я СКАЗАЛ ГОООЙДА!</center></h1>"

PLACEHOLDER = """
<center>
<p>ГООООЙДА!!</p>
</center>
"""

# pip install transformers
from transformers import AutoModelForCausalLM, AutoTokenizer

device = "cpu" # for GPU usage or "cpu" for CPU usage

tokenizer = AutoTokenizer.from_pretrained(MODEL_LIST[0])
model = AutoModelForCausalLM.from_pretrained(MODEL_LIST[0]).to(device)


#@spaces.GPU()
def stream_chat(
    message: str, 
    history: list, 
    temperature: float = 0.4, 
    max_new_tokens: int = 1024, 
    top_p: float = 1.0, 
    top_k: int = 20, 
    penalty: float = 1.2,
    choice: str = "GoidaAlignment/GOIDA-0.5B"
):
    print(f'message: {message}')
    print(f'history: {history}')

    conversation = []
    for prompt, answer in history:
        conversation.extend([
            {"role": "user", "content": prompt}, 
            {"role": "assistant", "content": answer},
        ])

    conversation.append({"role": "user", "content": message})

    

    input_text=tokenizer.apply_chat_template(conversation,  add_generation_prompt=True, tokenize=False)
    inputs = tokenizer.encode(input_text, return_tensors="pt").to(device)
    streamer = TextIteratorStreamer(tokenizer, timeout=60.0, skip_prompt=True, skip_special_tokens=True)
    
    generate_kwargs = dict(
        input_ids=inputs, 
        max_new_tokens = max_new_tokens,
        do_sample = False if temperature == 0 else True,
        top_p = top_p,
        top_k = top_k,
        temperature = temperature,
        streamer=streamer,
    )

    with torch.no_grad():
        thread = Thread(target=model.generate, kwargs=generate_kwargs)
        thread.start()
        
    buffer = ""
    for new_text in streamer:
        buffer += new_text
        yield buffer

            
    #print(tokenizer.decode(outputs[0]))

chatbot = gr.Chatbot(height=600, placeholder=PLACEHOLDER)

with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
    gr.HTML(TITLE)
    gr.ChatInterface(
        fn=stream_chat,
        chatbot=chatbot,
        fill_height=True,
        additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False, render=False),
        additional_inputs=[
            gr.Slider(
                minimum=0,
                maximum=1,
                step=0.1,
                value=0.4,
                label="Temperature",
                render=False,
            ),
            gr.Slider(
                minimum=128,
                maximum=8192,
                step=1,
                value=1024,
                label="Max new tokens",
                render=False,
            ),
            gr.Slider(
                minimum=0.0,
                maximum=1.0,
                step=0.1,
                value=1.0,
                label="top_p",
                render=False,
            ),
            gr.Slider(
                minimum=1,
                maximum=20,
                step=1,
                value=20,
                label="top_k",
                render=False,
            ),
            gr.Slider(
                minimum=0.0,
                maximum=2.0,
                step=0.1,
                value=1.2,
                label="Repetition penalty",
                render=False,
            ),
            gr.Radio(
                ["GoidaAlignment/GOIDA-0.5B"],
                value="494M",
                label="Load Model",
                render=False,
            ),
        ],
        cache_examples=False,
    )


if __name__ == "__main__":
    demo.launch()