run/qwen_32b.py

from transformers import Qwen2_5_VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info
from tqdm import tqdm
import json
import os
from pathlib import Path
from threading import Lock
from typing import Any




_json_write_lock = Lock()

def save_json_file(
    data: Any,
    file_path: str,
    indent: int = 4,
    temp_suffix: str = ".tmp"
) -> None:
    """
    """
    path = Path(file_path)
    path.parent.mkdir(parents=True, exist_ok=True)

    temp_path = f"{file_path}{temp_suffix}"

    with _json_write_lock:
        try:

            with open(temp_path, "w", encoding="utf-8") as f:
                json.dump(data, f, ensure_ascii=False, indent=indent)

                f.flush()
                os.fsync(f.fileno())

            os.replace(temp_path, file_path)

        except Exception as e:

            try:
                if os.path.exists(temp_path):
                    os.remove(temp_path)
            except OSError:
                pass
            raise RuntimeError(f"save json failed: {e}") from e

def read_json_file(file_path):
    """
    Reads a JSON file and returns the parsed data as a Python object.

    :param file_path: The path to the JSON file
    :return: The data parsed from the JSON file
    """
    with open(file_path, 'r', encoding='utf-8') as f:
        data = json.load(f)
    return data


def merge_json_lists(folder_path):


    json_list = [
        os.path.join(folder_path, f)
        for f in os.listdir(folder_path)
        if f.lower().endswith('.json') and os.path.isfile(os.path.join(folder_path, f))
    ]

    merged_list = []

    for file_path in json_list:
        try:
            with open(file_path, 'r', encoding='utf-8') as f:
                data = json.load(f)
                if isinstance(data, list):
                    merged_list.extend(data)
                else:
                    print(f"警告: {file_path} 的内容不是列表，已跳过")
        except Exception as e:
            print(f"处理文件 {file_path} 时出错: {str(e)}")

    return merged_list


def build():
        #cache_dir="/workspace/huggingface"
    # default: Load the model on the available device(s)
    model = Qwen2_5_VLForConditionalGeneration.from_pretrained( "Qwen/Qwen2.5-VL-32B-Instruct", torch_dtype="auto", device_map="auto")

    # default processer
    processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-32B-Instruct")
    

    return model,processor


def conversation(model, processor, image_path, prompt):
    messages = [
        {
            "role": "user",
            "content": [
                {
                    "type": "image",
                    "image": image_path,
                },
                {"type": "text", "text": prompt},
            ],
        }
    ]
    
    # Preparation for inference
    text = processor.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )
    image_inputs, video_inputs = process_vision_info(messages)
    inputs = processor(
        text=[text],
        images=image_inputs,
        videos=video_inputs,
        padding=True,
        return_tensors="pt",
    )
    inputs = inputs.to("cuda")
    
    # Inference: Generation of the output
    generated_ids = model.generate(**inputs, max_new_tokens=128)
    generated_ids_trimmed = [
        out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
    ]
    output_text = processor.batch_decode(
        generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
    )
    print(output_text)
    return output_text


if __name__ == "__main__":
    folder_path = r"your_path/poster/data"
    save_dir = r"your_path/poster/result"

    saved_josn = os.path.join(save_dir,"qwen_32b_bench.json")
    if not os.path.exists(saved_josn):
        tasks = merge_json_lists(folder_path)
        save_json_file(tasks, saved_josn)

    tasks = read_json_file(saved_josn)
    model,processor = build()
    print("build model successful")
    for item in tqdm(tasks):
        if "response" in item: continue

        prompt = item["prompt"]
        image_path = os.path.join(folder_path, item["path"])
        try:
            item["response"] = conversation(model, processor, image_path, prompt)
            print(item["response"])
            save_json_file(tasks, saved_josn)
            break
        except Exception as e:
            print(f"[Warning]  failed: {e}")