yangjianxin1 / clip-chinese Goto Github PK

您好！我想用您的模型来对本地数据集进行图文相似度计算。请问是否需要进行微调呢？谢谢！

KeyError: 'text'

你好，请问有预训练好的原始的ckpt权重吗？

你好，我遇到了没有cuda的问题
在windows后台运行：python train_clip.py --train_args_file train_args/train_clip.json

报错
Traceback (most recent call last):
File "train_clip.py", line 136, in
main()
File "train_clip.py", line 86, in main
args, training_args = parser.parse_json_file(json_file=train_args_file)
File "E:\ancanda\envs\CLIP-Chinese\lib\site-packages\transformers\hf_argparser.py", line 392, in parse_json_file
outputs = self.parse_dict(data, allow_extra_keys=allow_extra_keys)
File "E:\ancanda\envs\CLIP-Chinese\lib\site-packages\transformers\hf_argparser.py", line 367, in parse_dict
obj = dtype(**inputs)
File "", line 105, in init
File "E:\ancanda\envs\CLIP-Chinese\lib\site-packages\transformers\training_args.py", line 1133, in post_init
raise ValueError(
ValueError: FP16 Mixed precision training with AMP or APEX (--fp16) and FP16 half precision evaluation (--fp16_full_eval) can only be used on CUDA devices.

cuda版本是12.4
(CLIP-Chinese) F:\python\CLIP-Chinese>nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2024 NVIDIA Corporation
Built on Tue_Feb_27_16:28:36_Pacific_Standard_Time_2024
Cuda compilation tools, release 12.4, V12.4.99
Build cuda_12.4.r12.4/compiler.33961263_0

是版本不兼容吗？

When I follow your steps to train, I always get the following error

KeyError: Caught KeyError in DataLoader worker process 0.
KeyError: 'text'

I am a beginner in natural language processing.

When I clone the repository, install the environment dependencies, and try to run the quickstart code in README, the following error occurs, how should I solve it?

my python version is 3.8.15, other dependencies have the same version as the requirements.txt

Traceback (most recent call last):
  File "quickstart.py", line 15, in <module>
    inputs = processor(text=["一只小狗在摇尾巴", "一只小猪在吃饭"], images=image, return_tensors="pt", padding=True)
  File "/home/liuzhiming/.miniconda3/envs/clip-chinese/lib/python3.8/site-packages/transformers/models/clip/processing_clip.py", line 85, in __call__
    image_features = self.feature_extractor(images, return_tensors=return_tensors, **kwargs)
  File "/home/liuzhiming/.miniconda3/envs/clip-chinese/lib/python3.8/site-packages/transformers/models/clip/feature_extraction_clip.py", line 146, in __call__
    images = [self.resize(image=image, size=self.size, resample=self.resample) for image in images]
  File "/home/liuzhiming/.miniconda3/envs/clip-chinese/lib/python3.8/site-packages/transformers/models/clip/feature_extraction_clip.py", line 146, in <listcomp>
    images = [self.resize(image=image, size=self.size, resample=self.resample) for image in images]
  File "/home/liuzhiming/.miniconda3/envs/clip-chinese/lib/python3.8/site-packages/transformers/models/clip/feature_extraction_clip.py", line 207, in resize
    new_short, new_long = size, int(size * long / short)
TypeError: unsupported operand type(s) for *: 'dict' and 'int'

I tried to output the values of several parameters in transformers/models/clip/feature_extraction_clip.py", line 207, in resize and found that the type of size is dict, not int，The specific values are as follows：

size:  {'shortest_edge': 224}
long=960
short=600

when install pytorch with cpu-only, and conduct training.

An error warning
FP16 Mixed precision training with AMP or APEX (--fp16) and FP16 half precision evaluation (--fp16_full_eval) can only be used on CUDA devices.