yaohungt / Multimodal-Transformer

[ACL'19] [PyTorch] Multimodal Transformer
MIT License
799 stars 149 forks source link

How can I set these hyperparameters? #23

Closed ChenBattle closed 4 years ago

ChenBattle commented 4 years ago

I use the default parameters to train, but the result is not as good as that in the paper. I guess it must be a problem of hyperparameters setting. The value of hyperparameters are introduced in the paper, but I don't know how to set these values in the code:

  1. Transformers Hidden Unit Size d
  2. of Crossmodal Blocks D
  3. of Crossmodal Attention Heads
  4. Temporal Convolution Kernel Size (L/V/A)

Could you please help me? Thank you very much!

yaohungt commented 4 years ago
  1. and 4. are in src/models.py
  2. and 3. are in main.py
CindyTing commented 2 years ago

Hello, I also have same question. Does the "# of Crossmodal Blocks D" means "--nlevels" in main.py? Does the "Transformers Hidden Unit Size d" means "self.d_l, self.d_a, self.d_v" in src/models.py?

Could you help me out of here? Thanks a lot!