app.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

MAX_MAX_NEW_TOKENS = 2048
DEFAULT_MAX_NEW_TOKENS = 1024
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
ACCESS_TOKEN = os.getenv("HF_TOKEN", "")

model_id = "Qwen/Qwen2.5-0.5B-Instruct"
#filename = "Mistral-Nemo-Instruct-2407-Q6_K_L.gguf"

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    #gguf_file=filename,
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=False,
    token=ACCESS_TOKEN)

tokenizer = AutoTokenizer.from_pretrained(
    model_id,
    #gguf_file=filename,
    trust_remote_code=False,
    token=ACCESS_TOKEN)

tokenizer.use_default_system_prompt = False
model.config.gradient_checkpointing = True

@spaces.GPU
def generate(
    message: str,
    system_prompt: str,
    max_new_tokens: int = 1024,
    temperature: float = 0.01,
    top_p: float = 0.7,
) -> Iterator[str]:
    conversation = []
    if system_prompt:
        conversation.append({"role": "system", "content": system_prompt})
    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)

    '''
    terminators = [
        tokenizer.eos_token_id,
        tokenizer.convert_tokens_to_ids("<|eot_id|>")
    ]
    '''
    
    streamer = TextIteratorStreamer(tokenizer, timeout=600.0, skip_prompt=True, skip_special_tokens=True)
    generate_kwargs = dict(
        {"input_ids": input_ids},
        streamer=streamer,
        max_new_tokens=max_new_tokens,
        #eos_token_id=terminators,
        do_sample=True,
        top_p=top_p,
        temperature=temperature,
        num_beams=1,
        pad_token_id=tokenizer.eos_token_id, 
    )
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

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


chat_interface = gr.Interface(
    fn=generate,
    inputs=[
        gr.Textbox(lines=2, placeholder="Prompt", label="Prompt"),
    ],
    outputs="text",
    additional_inputs=[
        gr.Textbox(label="System prompt", lines=6),
        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=0.01,
            value=0.01,
        ),
        gr.Slider(
            label="Top-p (nucleus sampling)",
            minimum=0.05,
            maximum=1.0,
            step=0.01,
            value=0.7,
        ),          
    ],
    title="Model testing - Qwen/Qwen2.5-0.5B-Instruct",
    description="Provide system settings and a prompt to interact with the model.",
)

chat_interface.queue(max_size=20).launch()