app_chat.py

import os
from threading import Thread
from typing import Iterator

import gradio as gr
import spaces
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
# from transformers import StoppingCriteria, StoppingCriteriaList, StopStringCriteria

MAX_MAX_NEW_TOKENS = 1024
DEFAULT_MAX_NEW_TOKENS = 512
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))

DESCRIPTION = """\
# Hymba-1.5B chat

"""

model_id = "nvidia/Hymba-1.5B-Instruct"
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="cuda", model_dtype="bfloat16", trust_remote_code=True)
model = model.cuda().to(torch.bfloat16)

#model.to('cuda')
#model.eval()
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
tokenizer.chat_template = "{{'<extra_id_0>System'}}{% for message in messages %}{% if message['role'] == 'system' %}{{'\n' + message['content'].strip()}}{% if tools or contexts %}{{'\n'}}{% endif %}{% endif %}{% endfor %}{% if tools %}{% for tool in tools %}{{ '\n<tool> ' + tool|tojson + ' </tool>' }}{% endfor %}{% endif %}{% if contexts %}{% if tools %}{{'\n'}}{% endif %}{% for context in contexts %}{{ '\n<context> ' + context.strip() + ' </context>' }}{% endfor %}{% endif %}{{'\n\n'}}{% for message in messages %}{% if message['role'] == 'user' %}{{ '<extra_id_1>User\n' + message['content'].strip() + '\n' }}{% elif message['role'] == 'assistant' %}{{ '<extra_id_1>Assistant\n' + message['content'].strip() + '\n' }}{% elif message['role'] == 'tool' %}{{ '<extra_id_1>Tool\n' + message['content'].strip() + '\n' }}{% endif %}{% endfor %}{%- if add_generation_prompt %}{{'<extra_id_1>Assistant\n'}}{%- endif %}"
#tokenizer.use_default_system_prompt = False

# class StoppingCriteriaSub(StoppingCriteria):
#     def __init__(self, tokenizer, stops = [], encounters=1):
#         super().__init__()
#         self.stops = [stop.to("cuda") for stop in stops]
#         self.tokenizer = tokenizer
#         self.num_mamba_stop_ids = 8

#     def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor):
#         last_token = input_ids[0][-self.num_mamba_stop_ids:]
#         for stop in self.stops:
#             if self.tokenizer.decode(stop) in self.tokenizer.decode(last_token):
#                 return True
#         return False

@spaces.GPU
def generate(
    message: str,
    chat_history: list[dict],
    system_prompt: str = "",
    max_new_tokens: int = 1024,
    temperature: float = 0.6,
    top_p: float = 0.9,
    top_k: int = 50,
    repetition_penalty: float = 1.2,
) -> Iterator[str]:
    conversation = []
    if system_prompt:
        conversation.append({"role": "system", "content": system_prompt})
    conversation += chat_history
    conversation.append({"role": "user", "content": message})

    input_ids = tokenizer.apply_chat_template(conversation,  tokenize=True, add_generation_prompt=True, return_tensors="pt").to('cuda').to(torch.bfloat16)
    
    # stopping_criteria = StoppingCriteriaList([StopStringCriteria(tokenizer=tokenizer, stop_strings="</s>")])
    
    if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
        input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
        gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
    input_ids = input_ids.to(model.device)

    streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=False)
    generate_kwargs = dict(
        {"input_ids": input_ids},
        streamer=streamer,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        top_p=top_p,
        top_k=top_k,
        temperature=temperature,
        num_beams=1,
        repetition_penalty=repetition_penalty,
        # "stopping_criteria": stopping_criteria,
    )
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    outputs = []
    for text in streamer:
        outputs.append(text)
        yield "".join(outputs)


chat_interface = gr.ChatInterface(
    fn=generate,
    additional_inputs=[
        gr.Textbox(label="System prompt", lines=6, value="You are a helpful assistant. Your name is Hymba-1.5B-Instruct-8K. \
                                               You are a new family of small language models featuring a hybrid-head architecture that strategically integrates attention mechanisms with state space models (SSMs). \
                                               You are developed by Deep Learning Efficiency Research (DLER) team at NVIDIA Research. \
                                               Nvidia Corporation is an American multinational corporation and technology company headquartered in Santa Clara, California. Nvidia was founded on April 5, 1993 by Jensen Huang. \
                                               The above is just a background context. You can answer any questions not limited to the above background context."),
        gr.Slider(
            label="Max new tokens",
            minimum=1,
            maximum=MAX_MAX_NEW_TOKENS,
            step=1,
            value=DEFAULT_MAX_NEW_TOKENS,
        ),
        gr.Slider(
            label="Temperature",
            minimum=0.1,
            maximum=4.0,
            step=0.1,
            value=0.6,
        ),
        gr.Slider(
            label="Top-p (nucleus sampling)",
            minimum=0.05,
            maximum=1.0,
            step=0.05,
            value=0.9,
        ),
        gr.Slider(
            label="Top-k",
            minimum=1,
            maximum=1000,
            step=1,
            value=50,
        ),
        gr.Slider(
            label="Repetition penalty",
            minimum=1.0,
            maximum=2.0,
            step=0.05,
            value=1.2,
        ),
    ],
    stop_btn=None,
    examples=[
        ["Hello there! How are you doing?"],
        ["Can you explain briefly to me what is the Python programming language?"],
        ["Explain the plot of Cinderella in a sentence."],
        ["How many hours does it take a man to eat a Helicopter?"],
        ["Write a 100-word article on 'Benefits of Open-Source in AI research'"],
    ],
    cache_examples=False,
    type="messages",
)

with gr.Blocks(css_paths="style.css", fill_height=True) as demo:
    gr.Markdown(DESCRIPTION)
    # gr.DuplicateButton(value="Duplicate Space for private use", elem_id="duplicate-button")
    chat_interface.render()
    # gr.Markdown(LICENSE)

if __name__ == "__main__":
    demo.queue(max_size=20).launch()