app.py

import os
from threading import Thread
from typing import Iterator
import gradio as gr
import spaces, torch, requests
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer

MAX_MAX_NEW_TOKENS = 2048
DEFAULT_MAX_NEW_TOKENS = 2048
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "128000"))

DESCRIPTION = """\
# DeepSeek-R1-Chat

This space demonstrates model [DeepSeek-R1](https://huggingface.co/deepseek-ai/deepseek-r1) by DeepSeek, a code model with 6XXB parameters fine-tuned for chat instructions.

**You can also try our R1 model in [official homepage](https://r1.deepseek.com/chat).**
"""
if not torch.cuda.is_available():
    DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"

if torch.cuda.is_available():
    model_id = "deepseek-ai/deepseek-r1"
    model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16, device_map="auto")
    tokenizer = AutoTokenizer.from_pretrained(model_id)
    tokenizer.use_default_system_prompt = False

@spaces.GPU
def generate(message: str,
             chat_history: list[tuple[str, str]],
             system_prompt: str,
             max_new_tokens: int = 2048,
             temperature: float = 0,
             top_p: float = 0,
             top_k: int = 50,
             repetition_penalty: float = 2,
             search_query: str = "") -> Iterator[str]:
    conversation = []
    if system_prompt:
        conversation.append({"role": "system", "content": system_prompt})
    if search_query:
        try:
            r = requests.get(f"https://api.duckduckgo.com/?q={search_query}&format=json", timeout=5)
            data = r.json()
            result = data.get("AbstractText", "")
            if result:
                conversation.append({"role": "system", "content": f"Search results for '{search_query}': {result}"})
        except Exception as e:
            conversation.append({"role": "system", "content": f"Search error: {e}"})
    for user, assistant in chat_history:
        conversation.extend([{"role": "user", "content": user},
                             {"role": "assistant", "content": assistant}])
    conversation.append({"role": "user", "content": message})
    input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt")
    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=True)
    generate_kwargs = {
        "input_ids": input_ids,
        "streamer": streamer,
        "max_new_tokens": max_new_tokens,
        "do_sample": False,
        "top_p": top_p,
        "top_k": top_k,
        "num_beams": 1,
        "repetition_penalty": repetition_penalty,
        "eos_token_id": 32021
    }
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()
    outputs = []
    for text in streamer:
        outputs.append(text)
        yield "".join(outputs).replace("<|EOT|>", "")

chat_interface = gr.ChatInterface(
    fn=generate,
    additional_inputs=[
        gr.Textbox(label="System prompt", lines=6),
        gr.Slider(label="Max new tokens", minimum=0, maximum=MAX_MAX_NEW_TOKENS, step=0.01, value=DEFAULT_MAX_NEW_TOKENS),
        gr.Slider(label="Top-p (nucleus sampling)", minimum=0, maximum=1.0, step=0.01, value=0),
        gr.Slider(label="Top-k", minimum=1, maximum=1000, step=0.01, value=50),
        gr.Slider(label="Repetition penalty", minimum=1.0, maximum=2.0, step=0.01, value=2),
        gr.Textbox(label="Search Query (Optional)", placeholder="Enter search query to fetch online info", lines=1)
    ],
    stop_btn=gr.Button("Stop"),
    examples=[
        ["implement snake game using pygame"],
        ["Can you explain briefly to me what is the Python programming language?"],
        ["write a program to find the factorial of a number"]
    ],
)

with gr.Blocks(css="style.css") as demo:
    gr.Markdown(DESCRIPTION)
    chat_interface.render()

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