# 导入必要库
from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
from qwen_vl_utils import process_vision_info
import torch

# 加载模型
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    "chancharikm/qwen2.5-vl-7b-cam-motion-preview", torch_dtype="auto", device_map="auto"
)
processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct")

# 准备输入数据
video_path = "file:///视频路径/video1.mp4"
text_description = "镜头向上倾斜"
question = f"该视频是否呈现「{text_description}」的场景？"

# 格式化模型输入
messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "video",
                "video": video_path,
                "fps": 8.0,  # 推荐推理帧率
            },
            {"type": "text", "text": question},
        ],
    }
]

text = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
image_inputs, video_inputs, video_kwargs = process_vision_info(messages, return_video_kwargs=True)
inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
    **video_kwargs
)
inputs = inputs.to("cuda")

# 生成评分输出
with torch.inference_mode():
    outputs = model.generate(
        **inputs,
        max_new_tokens=1,
        do_sample=False,  # 使用贪心解码确保概率可靠性
        output_scores=True,
        return_dict_in_generate=True
    )

# 计算"是"的概率
scores = outputs.scores[0]
probs = torch.nn.functional.softmax(scores, dim=-1)
yes_token_id = processor.tokenizer.encode("是")[0]
score = probs[0, yes_token_id].item()

print(f"视频: {video_path}")
print(f"描述: '{text_description}'")
print(f"匹配度: {score:.4f}")

# 推荐使用训练时的8.0 FPS进行推理

from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
from qwen_vl_utils import process_vision_info

# 默认加载到可用设备
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    "chancharikm/qwen2.5-vl-7b-cam-motion-preview", torch_dtype="auto", device_map="auto"
)

# 推荐启用flash_attention_2以获得更佳加速和显存优化（特别在多图像/视频场景）
# model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
#     "chancharikm/qwen2.5-vl-7b-cam-motion-preview",
#     torch_dtype=torch.bfloat16,
#     attn_implementation="flash_attention_2",
#     device_map="auto",
# )

# 默认处理器
processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct")

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "video",
                "video": "file:///视频路径/video1.mp4",
                "fps": 8.0,
            },
            {"type": "text", "text": "描述该视频中的摄像机运动"},
        ],
    }
]

text = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
image_inputs, video_inputs, video_kwargs = process_vision_info(messages, return_video_kwargs=True)
inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    fps=fps,
    padding=True,
    return_tensors="pt",
    **video_kwargs,
)
inputs = inputs.to("cuda")

# 推理
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)