app.py

import gradio as gr
import spaces

from threading import Thread
from torch import bfloat16
from transformers import Qwen2VLForConditionalGeneration, Qwen2VLProcessor, TextIteratorStreamer, AutoProcessor
from qwen_vl_utils import process_vision_info

model_path = "Pectics/Softie-VL-7B-250123"

model = Qwen2VLForConditionalGeneration.from_pretrained(
    model_path,
    torch_dtype=bfloat16,
    attn_implementation="flash_attention_2",
    device_map="auto",
)
min_pixels = 256 * 28 * 28
max_pixels = 1280 * 28 * 28
processor: Qwen2VLProcessor = AutoProcessor.from_pretrained(model_path, min_pixels=min_pixels, max_pixels=max_pixels)

@spaces.GPU
def infer(
    messages,
    max_tokens,
    temperature,
    top_p,
):
    text_inputs = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    image_inputs, video_inputs = process_vision_info(messages)
    inputs = processor(
        text=[text_inputs],
        images=image_inputs,
        videos=video_inputs,
        padding=True,
        return_tensors="pt",
    ).to("cuda")
    
    streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=True)
    kwargs = dict(
        **inputs,
        streamer=streamer,
        max_new_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
    )
    thread = Thread(target=model.generate, kwargs=kwargs)
    thread.start()
    
    response = ""
    for token in streamer:
        response += token
        yield response

def respond(
    message,
    history,
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    messages = [{"role": "system", "content": system_message}]
    for m in history:
        messages.append({"role": m["role"], "content": m["content"]})
    messages.append({"role": "user", "content": message})
    return infer(messages, max_tokens, temperature, top_p)

app = gr.ChatInterface(
    respond,
    type="messages",
    additional_inputs=[
        gr.Textbox(value="You are Softie, a helpful assistant.", label="系统设定"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="最大生成长度"),
        gr.Slider(minimum=0.01, maximum=4.0, value=0.75, step=0.01, label="温度系数（Temperature）"),
        gr.Slider(minimum=0.01, maximum=1.0, value=0.5, step=0.01, label="核取样系数（Top-p）"),
    ],
)

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