invictus717 / MetaTransformer

Meta-Transformer for Unified Multimodal Learning
https://arxiv.org/abs/2307.10802
Apache License 2.0
1.52k stars 114 forks source link

video #50

Closed Darcy0218 closed 1 year ago

Darcy0218 commented 1 year ago

您好!我在运行video的run.sh中遇到了如下的问题: 82E0EC63745B1379058FBC680703B9C7 代码在执行 ckpt = torch.load("Meta-Transformer_base_patch16_encoder.pth") model.blocks.load_state_dict(ckpt,strict=True) 时遇到了如上报错。 我猜测是由于模型结构不一致造成的,想请教下您是如何解决的,感谢!

invictus717 commented 1 year ago

Video MAE里面timm库的版本是0.4.12那个版本的代码实现qkv还没有被合并,这里可以直接升级到0.9.7

Darcy0218 commented 1 year ago

谢谢您

Darcy0218 commented 1 year ago

您好,我试了一下升级timm到0.9.7,还是报了相同的错,请问您的torch是什么版本的

invictus717 commented 1 year ago

1.8.1+cu113

begin00001 commented 2 months ago

您好,我在运行bash scripts/long_term_forecast/Traffic_script/MetaTransformer.sh出现如下错误: image ckpt = torch.load("Meta-Transformer_base_patch16_encoder.pth") FileNotFoundError: [Errno 2] No such file or directory: 'Meta-Transformer_base_patch16_encoder.pth'

请问这个模型是运行哪步生成的?