app.py

from threading import Thread

import gradio as gr
import spaces
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer

MAX_INPUT_LIMIT = 3584
MAX_NEW_TOKENS = 1536
MODEL_NAME = "Azure99/blossom-v5.1-9b"

model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, torch_dtype=torch.bfloat16, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)


def get_input_ids(inst, history):
    prefix = ("A chat between a human and an artificial intelligence bot. "
              "The bot gives helpful, detailed, and polite answers to the human's questions.")
    patterns = []
    for conv in history:
        patterns.append(f'\n|Human|: {conv[0]}\n|Bot|: ')
        patterns.append(f'{conv[1]}')
    patterns.append(f'\n|Human|: {inst}\n|Bot|: ')
    patterns[0] = prefix + patterns[0]

    input_ids = []
    for i, pattern in enumerate(patterns):
        input_ids += tokenizer.encode(pattern, add_special_tokens=(i == 0))
        if i % 2 == 1:
            input_ids += [tokenizer.eos_token_id]
    return input_ids


def generate(generation_kwargs):
    with torch.no_grad():
        Thread(target=model.generate, kwargs=generation_kwargs).start()


@spaces.GPU
def chat(inst, history, temperature, top_p, repetition_penalty):
    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
    input_ids = get_input_ids(inst, history)
    if len(input_ids) > MAX_INPUT_LIMIT:
        yield "The input is too long, please clear the history."
        return
    generation_kwargs = dict(input_ids=torch.tensor([input_ids]).to(model.device),
                             streamer=streamer, do_sample=True, max_new_tokens=MAX_NEW_TOKENS,
                             temperature=temperature, top_p=top_p, repetition_penalty=repetition_penalty)
    generate(generation_kwargs)

    outputs = ""
    for new_text in streamer:
        outputs += new_text
        yield outputs


additional_inputs = [
    gr.Slider(
        label="Temperature",
        value=0.5,
        minimum=0.0,
        maximum=1.0,
        step=0.05,
        interactive=True,
        info="Controls randomness in choosing words.",
    ),
    gr.Slider(
        label="Top-P",
        value=0.85,
        minimum=0.0,
        maximum=1.0,
        step=0.05,
        interactive=True,
        info="Picks words until their combined probability is at least top_p.",
    ),
    gr.Slider(
        label="Repetition penalty",
        value=1.05,
        minimum=1.0,
        maximum=1.2,
        step=0.01,
        interactive=True,
        info="Repetition Penalty: Controls how much repetition is penalized.",
    )
]

gr.ChatInterface(chat,
                 chatbot=gr.Chatbot(show_label=False, height=500, show_copy_button=True, render_markdown=True),
                 textbox=gr.Textbox(placeholder="", container=False, scale=7),
                 title="Blossom 9B Demo",
                 description='Hello, I am Blossom, an open source conversational large language model.🌠'
                             '<a href="https://github.com/Azure99/BlossomLM">GitHub</a>',
                 theme="soft",
                 examples=[["Hello"], ["What is MBTI"], ["用Python实现二分查找"],
                           ["为switch写一篇小红书种草文案，带上emoji"]],
                 additional_inputs=additional_inputs,
                 additional_inputs_accordion=gr.Accordion(label="Config", open=True),
                 clear_btn="🗑️Clear",
                 undo_btn="↩️Undo",
                 retry_btn="🔄Retry",
                 submit_btn="➡️Submit",
                 ).queue().launch()