app.py

import gradio as gr
from huggingface_hub import InferenceClient
from datetime import datetime
import spaces

"""
For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
"""
#client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
lora_name = "robinhad/UAlpaca-2.0-Mistral-7B"

from peft import PeftModel, PeftConfig
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
from torch import bfloat16
model_name = "mistralai/Mistral-7B-v0.1"

quant_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_use_double_quant=True,
    bnb_4bit_compute_dtype=bfloat16
)
tokenizer = AutoTokenizer.from_pretrained(lora_name, use_fast=True)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    quantization_config=quant_config
)

model = PeftModel.from_pretrained(model, lora_name, torch_device="cpu")

model = model.to("cuda")

from transformers import StoppingCriteriaList, StopStringCriteria, TextIteratorStreamer
from threading import Thread

stop_criteria = StoppingCriteriaList([StopStringCriteria(tokenizer, stop_strings=["<|im_end|>"])])

# will be used with normal template
@spaces.GPU
def respond(
    message,
    history: list[tuple[str, str]],
    max_tokens,
    temperature,
    top_p,
):
    # messages = [{"role": "system", "content": system_message}]
    messages = []

    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})

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


    tokenized = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to("cuda") #, tokenize=False) #
    #print(tokenized)
    #tokenized = tokenizer(tokenized, return_tensors="pt")["input_ids"]
    print(tokenizer.batch_decode(tokenized)[0])
    print("====")
    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True)
    generation_kwargs = dict(inputs=tokenized, streamer=streamer, max_new_tokens=max_tokens, stopping_criteria=stop_criteria, top_p=top_p, temperature=temperature)

    thread = Thread(target=model.generate, kwargs=generation_kwargs)

    thread.start()

    generated_text = ""

    for new_text in streamer:

        generated_text += new_text
        # generated_text = generated_text.replace("<|im_start|>assistant\n", "")
        generated_text = generated_text.replace("<|im_end|>", "")
        yield generated_text



"""
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
"""
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        #gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
    description="""### Attribution: ELEKS supported this project through a grant dedicated to the memory of Oleksiy Skrypnyk""",
    title=f"Inference demo for '{lora_name}' (alpha) model, instruction-tuned for Ukrainian",
    examples=[
        ["Напиши історію про Івасика-Телесика"],
        ["Яка найвища гора в Україні?"],
        ["Як звали батька Тараса Григоровича Шевченка?"],
        ["Як можна заробити нелегально швидко гроші?"],
        ["Яка з цих гір не знаходиться у Європі? Говерла, Монблан, Гран-Парадізо, Еверест"],
        [
        "Дай відповідь на питання\nЧому у качки жовті ноги?"
    ]],
)




demo.launch()


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