Qwen开源视觉推理模型QVQ，更睿智地看世界！

from modelscope import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info
# default: Load the model on the available device(s)
model = Qwen2VLForConditionalGeneration.from_pretrained(
    "Qwen/QVQ-72B-Preview", torch_dtype="auto", device_map="auto"
)
# default processer
processor = AutoProcessor.from_pretrained("Qwen/QVQ-72B-Preview")
# The default range for the number of visual tokens per image in the model is 4-16384. You can set min_pixels and max_pixels according to your needs, such as a token count range of 256-1280, to balance speed and memory usage.
# min_pixels = 256*28*28
# max_pixels = 1280*28*28
# processor = AutoProcessor.from_pretrained("Qwen/QVQ-72B-Preview", min_pixels=min_pixels, max_pixels=max_pixels)
messages = [
    {
        "role": "system",
        "content": [
            {"type": "text", "text": "You are a helpful and harmless assistant. You are Qwen developed by Alibaba. Answer in the language of the question. You should think step-by-step."}
        ],
    },
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://qianwen-res.oss-accelerate-overseas.aliyuncs.com/QVQ/demo.png",
            },
            {"type": "text", "text": "What value should be filled in the blank space?"},
        ],
    }
]
# 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=8192)
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)

import os
from openai import OpenAI
client = OpenAI(
    api_key=os.getenv("MODELSCOPE_ACCESS_TOKEN"),
    base_url="https://api-inference.modelscope.cn/v1"
)
response = client.chat.completions.create(
    model="Qwen/QVQ-72B-Preview",
    messages = [
        {
            "role": "system",
            "content": [
                {"type": "text", "text": "You are a helpful and harmless assistant. You are Qwen developed by Alibaba. You should think step-by-step."}
            ],
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "image_url",
                    "image_url": {"url": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/QVQ/demo.png"}
                },
                {"type": "text", "text": "What value should be filled in the blank space?"},
            ],
        }
    ],
    stream=True
    )
for chunk in response:
    print(chunk.choices[0].delta.content, end='', flush=True)

git clone https://github.com/modelscope/ms-swift.git
cd ms-swift
pip install -e .[llm]

# 实验环境：2*80GiB A100
MAX_PIXELS=1003520 \
CUDA_VISIBLE_DEVICES=0,1 \
swift sft \
    --model Qwen/QVQ-72B-Preview \
    --dataset AI-ModelScope/LaTeX_OCR:human_handwrite#20000 \
    --train_type lora \
    --torch_dtype bfloat16 \
    --num_train_epochs 1 \
    --per_device_train_batch_size 1 \
    --per_device_eval_batch_size 1 \
    --learning_rate 1e-4 \
    --lora_rank 8 \
    --lora_alpha 32 \
    --target_modules all-linear \
    --freeze_vit true \
    --gradient_accumulation_steps 16 \
    --eval_steps 50 \
    --save_steps 50 \
    --save_total_limit 5 \
    --logging_steps 5 \
    --max_length 2048 \
    --output_dir output \
    --warmup_ratio 0.05 \
    --dataloader_num_workers 4

# 实验环境：4*80GiB A100
# You can refer to `https://github.com/QwenLM/Qwen2-VL` for the meaning of the `VIDEO_MAX_PIXELS` parameter.
nproc_per_node=4
CUDA_VISIBLE_DEVICES=0,1,2,3 \
NPROC_PER_NODE=$nproc_per_node \
VIDEO_MAX_PIXELS=50176 \
FPS_MAX_FRAMES=12 \
swift sft \
    --model Qwen/QVQ-72B-Preview \
    --dataset swift/VideoChatGPT:all \
    --train_type lora \
    --torch_dtype bfloat16 \
    --num_train_epochs 1 \
    --per_device_train_batch_size 1 \
    --per_device_eval_batch_size 1 \
    --learning_rate 1e-4 \
    --lora_rank 8 \
    --lora_alpha 32 \
    --target_modules all-linear \
    --freeze_vit true \
    --gradient_accumulation_steps $(expr 16 / $nproc_per_node) \
    --eval_steps 50 \
    --save_steps 50 \
    --save_total_limit 5 \
    --logging_steps 5 \
    --max_length 2048 \
    --output_dir output \
    --warmup_ratio 0.05 \
    --dataloader_num_workers 4 \
    --deepspeed zero3

{"messages": [{"role": "user", "content": "浙江的省会在哪？"}, {"role": "assistant", "content": "浙江的省会在杭州。"}]}
{"messages": [{"role": "user", "content": "<image><image>两张图片有什么区别"}, {"role": "assistant", "content": "前一张是小猫，后一张是小狗"}], "images": ["/xxx/x.jpg", "xxx/x.png"]}
{"messages": [{"role": "system", "content": "你是个有用无害的助手"}, {"role": "user", "content": "<video>视频中是什么"}, {"role": "assistant", "content": "视频中是一只小狗在草地上奔跑"}], "videos": ["/xxx/x.mp4"]}

# 若是对视频数据进行推理，通常需要4卡
CUDA_VISIBLE_DEVICES=0,1 \
swift infer \
    --adapters output/vx-xxx/checkpoint-xxx \
    --stream false \
    --max_batch_size 1 \
    --load_data_args true \
    --max_new_tokens 2048