Gated Transformer Model for Computer Vision

Install

$ pip install gtrxl-torch

Implementation

from gtrxl_torch.gtrxl_torch import GTrXL
import torch
model = GTrXL(
    d_model=512,
    nheads=4,
    transformer_layers=1
)
input = torch.randn(32,16,512)
output = model(input)

Output Dimensions

Dimension ➯ [Sequence, Batch, Memory Size]

Saving Model

  model.save()

Loading Model

  model.load()

Parameters

• d_model: int.
  The number of expected features in the encoder/decoder inputs
• nheads: int.
  The number of heads in the multiheadattention models
• transformer_layers: int.
  Number of Transformer blocks.
• hidden_dims: int.
  Number of hidden neurons for the postion wise MLP.
• n_layers: int.
  RNN (GRU) layers.
• layer_norm_eps: float, default 1e-5.
  The eps value in layer normalization components.
• batch_first: bool, default False.
  (N, S, E) if batch first.
• chkpt_dir: str default models.
  Directory name where model is saved.
• activation: str, default relu.
  Activation function for MLP.
• network_name: str, default network.pt.
  Name of the model (file) you're saving.

Resources

• Alterations to the transformer model (GTrXL) ➱ Click Here