app.py

from threading import Thread

import torch
import gradio as gr
from transformers import pipeline,AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel, PeftConfig
import re


torch_device = "cuda" if torch.cuda.is_available() else "cpu"
print("Running on device:", torch_device)
print("CPU threads:", torch.get_num_threads())

peft_model_id = "ldhldh/1.3_40kstep" 

#peft_model_id = "ldhldh/polyglot-ko-1.3b_lora_big_tern_30kstep" 
# 20k or 30k 
#18k > 상대의 말까지 하는 이슈가 있음 
#8k > 약간 아쉬운가? 
config = PeftConfig.from_pretrained(peft_model_id)

base_model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path)
tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)

#base_model = AutoModelForCausalLM.from_pretrained("EleutherAI/polyglot-ko-3.8b")
#tokenizer = AutoTokenizer.from_pretrained("EleutherAI/polyglot-ko-3.8b")
base_model.eval()
base_model.config.use_cache = True


model = PeftModel.from_pretrained(base_model, peft_model_id, device_map="auto")
model.eval()
model.config.use_cache = True



def gen(x, top_p, top_k, temperature, max_new_tokens, repetition_penalty):
    gened = model.generate(
        **tokenizer(
            f"{x}",
            return_tensors='pt',
            return_token_type_ids=False
        ),
        #bad_words_ids = bad_words_ids ,
        max_new_tokens=max_new_tokens,
        min_new_tokens = 5,
        exponential_decay_length_penalty = (max_new_tokens/2, 1.1),
        top_p=top_p,
        top_k=top_k,
        temperature = temperature,
        early_stopping=True,
        do_sample=True,
        eos_token_id=2,
        pad_token_id=2,
        #stopping_criteria = stopping_criteria,
        repetition_penalty=repetition_penalty,
        no_repeat_ngram_size = 2
    )

    model_output = tokenizer.decode(gened[0])
    return model_output

def reset_textbox():
    return gr.update(value='')


with gr.Blocks() as demo:
    duplicate_link = "https://huggingface.co/spaces/beomi/KoRWKV-1.5B?duplicate=true"
    gr.Markdown(
       "duplicated from beomi/KoRWKV-1.5B, baseModel:EleutherAI/polyglot-ko-1.3b"
    )
    
    with gr.Row():
        with gr.Column(scale=4):
            user_text = gr.Textbox(
                placeholder='\\nfriend: 우리 여행 갈래? \\nyou:',
                label="User input"
            )
            model_output = gr.Textbox(label="Model output", lines=10, interactive=False)
            button_submit = gr.Button(value="Submit")

        with gr.Column(scale=1):
            max_new_tokens = gr.Slider(
                minimum=1, maximum=200, value=20, step=1, interactive=True, label="Max New Tokens",
            )
            top_p = gr.Slider(
                minimum=0.05, maximum=1.0, value=0.8, step=0.05, interactive=True, label="Top-p (nucleus sampling)",
            )
            top_k = gr.Slider(
                minimum=5, maximum=100, value=30, step=5, interactive=True, label="Top-k (nucleus sampling)",
            )
            temperature = gr.Slider(
                minimum=0.1, maximum=2.0, value=0.5, step=0.1, interactive=True, label="Temperature",
            )
            repetition_penalty = gr.Slider(
                minimum=1.0, maximum=3.0, value=1.2, step=0.1, interactive=True, label="repetition_penalty",
            )
    
    button_submit.click(gen, [user_text, top_p, top_k, temperature, max_new_tokens, repetition_penalty], model_output)
    
    demo.queue(max_size=32).launch(enable_queue=True)