按照主页示例输出失败
#78
by YoloBird - opened
代码参考主页的示例
from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info
import torch
def main():
# 加载模型和 processor
model = Qwen2VLForConditionalGeneration.from_pretrained(
"/root/autodl-tmp/Qwen/Qwen2-VL-2B", torch_dtype="auto", device_map="auto"
)
processor = AutoProcessor.from_pretrained("/root/autodl-tmp/Qwen/Qwen2-VL-2B")
# 构造一个聊天消息，包含图像和文本
messages = [
{
"role": "user",
"content": [
{
"type": "image",
"image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
},
{"type": "text", "text": "Describe this image."},
],
}
]
# 通过 processor 的模板构造最终的文本 prompt
text = processor.apply_chat_template(
messages, tokenize=False, add_generation_prompt=True
)
# 利用 process_vision_info 处理消息中的图像信息
image_inputs, video_inputs = process_vision_info(messages)
# 调用 processor 生成模型输入，注意返回张量形式
inputs = processor(
text=[text],
images=image_inputs,
videos=video_inputs,
padding=True,
return_tensors="pt",
)
# 将输入数据转到 GPU（如果可用）
device = "cuda" if torch.cuda.is_available() else "cpu"
inputs = inputs.to(device)
# 输出各个输入张量的形状
print("input_ids shape:", inputs.input_ids.shape)
print("attention_mask shape:", inputs.attention_mask.shape)
print("pixel_values shape:", inputs.pixel_values.shape)
# 可选：执行一次生成，观察生成结果
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("Generated text:", output_text)
if __name__ == "__main__":
main()
下面是报错信息，另外为什么输出的图像是二维的，没有通道数
(qwen_attack) root@autodl-container-28d246918f-350fbcec:~/HADES/Llava_test/qwen_test# python test.py
Loading checkpoint shards: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 2.08it/s]
Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.52, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
input_ids shape: torch.Size([1, 0])
attention_mask shape: torch.Size([1, 0])
pixel_values shape: torch.Size([14308, 1176])
Traceback (most recent call last):
File "/root/HADES/Llava_test/qwen_test/test.py", line 63, in <module>
main()
File "/root/HADES/Llava_test/qwen_test/test.py", line 53, in main
generated_ids = model.generate(**inputs, max_new_tokens=128)
File "/root/miniconda3/envs/qwen_attack/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
File "/root/miniconda3/envs/qwen_attack/lib/python3.10/site-packages/transformers/generation/utils.py", line 2326, in generate
result = self._sample(
File "/root/miniconda3/envs/qwen_attack/lib/python3.10/site-packages/transformers/generation/utils.py", line 3279, in _sample
model_inputs = self.prepare_inputs_for_generation(input_ids, **model_kwargs)
File "/root/miniconda3/envs/qwen_attack/lib/python3.10/site-packages/transformers/models/qwen2_vl/modeling_qwen2_vl.py", line 1792, in prepare_inputs_for_generation
model_inputs = super().prepare_inputs_for_generation(
File "/root/miniconda3/envs/qwen_attack/lib/python3.10/site-packages/transformers/generation/utils.py", line 419, in prepare_inputs_for_generation
or cache_position[-1] >= input_ids.shape[1] # Exception 3
IndexError: index -1 is out of bounds for dimension 0 with size 0
解决了吗，我也是这个问题