dimension for HuBERT base

[TonyWangX]

HuBERT base output has 768 dimensions.

1. AASIST LL needs to use self.input_dim to specify the input feature dimension?
   

   deepfense-framework/deepfense/models/backends/aasist.py

   Line 391 in c7b6186

   self.LL = nn.Linear(1024, 128)

2. yaml for HuBERT models may be updated. This works for me

AASIST

backend:
  type: AASIST
  args:
    input_dim: 768

MLP

input_dim: 768

TCM

args: {input_dim: 768, emb_size: 128, heads: 4, num_encoders: 1}

Nes2Net

 backend:
   type: Nes2Net
   args:
     input_dim: 768


- {type: CrossEntropy, embedding_dim: 768, n_classes: 2, weight: 1.0}