search

OpenDriveLab / Vista

A Generalizable World Model for Autonomous Driving
https://vista-demo.github.io
Apache License 2.0
362 stars 16 forks source link

KeyError: 'model_emadecay' #7

Closed JunyuanDeng closed 2 weeks ago

JunyuanDeng commented 2 weeks ago

Hello, thanks for your excellent job!

When I try to train the example, it came out with following error:

....
VideoTransformerBlock is using checkpointing
Initialized embedder #0: FrozenOpenCLIPImagePredictionEmbedder with 683800065 params. Trainable: False
Initialized embedder #1: ConcatTimestepEmbedderND with 0 params. Trainable: False
Initialized embedder #2: ConcatTimestepEmbedderND with 0 params. Trainable: False
Making attention of type `vanilla-xformers` with 512 in_channels
Building MemoryEfficientAttnBlock with 512 in_channels...
Working with z of shape (1, 4, 32, 32) = 4096 dimensions
Making attention of type `vanilla-xformers` with 512 in_channels
Building MemoryEfficientAttnBlock with 512 in_channels...
Initialized embedder #3: VideoPredictionEmbedderWithEncoder with 83653863 params. Trainable: False
Initialized embedder #4: ConcatTimestepEmbedderND with 0 params. Trainable: False
Initialized embedder #5: ConcatTimestepEmbedderND with 0 params. Trainable: False
Initialized embedder #6: ConcatTimestepEmbedderND with 0 params. Trainable: False
Initialized embedder #7: ConcatTimestepEmbedderND with 0 params. Trainable: False
Initialized embedder #8: ConcatTimestepEmbedderND with 0 params. Trainable: False
Initialized embedder #9: ConcatTimestepEmbedderND with 0 params. Trainable: False
Making attention of type `vanilla` with 512 in_channels
Working with z of shape (1, 4, 32, 32) = 4096 dimensions
Skipping timestep embedding in ResBlock
Making attention of type `vanilla` with 512 in_channels
Skipping timestep embedding in ResBlock
Skipping timestep embedding in ResBlock
Skipping timestep embedding in ResBlock
Skipping timestep embedding in ResBlock
Skipping timestep embedding in ResBlock
Skipping timestep embedding in ResBlock
Skipping timestep embedding in ResBlock
Skipping timestep embedding in ResBlock
Skipping timestep embedding in ResBlock
Skipping timestep embedding in ResBlock
Skipping timestep embedding in ResBlock
Skipping timestep embedding in ResBlock
Skipping timestep embedding in ResBlock
Keeping EMAs of 1434
Loading pretrained model from /horizon-bucket/saturn_v_dev/users/junyuan.deng/weights/svd_xt.safetensors
Traceback (most recent call last):
  File "/home/users/junyuan.deng/Programmes/Vista/train.py", line 668, in <module>
    svd[miss_k] = svd[ema_name]
KeyError: 'model_emadecay'

What should I do to solve the problem?

Little-Podi commented 2 weeks ago

I have fixed the bug in the latest script. Thanks for pointing that out!