Module: Transformers::ModuleUtilsMixin

Included in:
PreTrainedModel
Defined in:
lib/transformers/modeling_utils.rb

Instance Method Summary collapse

Instance Method Details

#device ⇒ Object 



59
60
61
 
# File 'lib/transformers/modeling_utils.rb', line 59

def device
  get_parameter_device(self)
end

#get_extended_attention_mask(attention_mask, input_shape, device: nil, dtype: nil) ⇒ Object 



63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
 
# File 'lib/transformers/modeling_utils.rb', line 63

def get_extended_attention_mask(
  attention_mask,
  input_shape,
  device: nil,
  dtype: nil
)
  if dtype.nil?
    dtype = @dtype
  end

  if !(attention_mask.dim == 2 && @config.is_decoder)
    # show warning only if it won't be shown in `create_extended_attention_mask_for_decoder`
    if !device.nil?
      raise Todo
    end
  end
  # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
  # ourselves in which case we just need to make it broadcastable to all heads.
  if attention_mask.dim == 3
    raise Todo
  elsif attention_mask.dim == 2
    # Provided a padding mask of dimensions [batch_size, seq_length]
    # - if the model is a decoder, apply a causal mask in addition to the padding mask
    # - if the model is an encoder, make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]
    if @config.is_decoder
      raise Todo
    else
      extended_attention_mask = attention_mask[0.., nil, nil, 0..]
    end
  else
    raise Todo
  end

  # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
  # masked positions, this operation will create a tensor which is 0.0 for
  # positions we want to attend and the dtype's smallest value for masked positions.
  # Since we are adding it to the raw scores before the softmax, this is
  # effectively the same as removing these entirely.
  extended_attention_mask = extended_attention_mask.to(dtype: dtype)  # fp16 compatibility
  # TODO use Torch.finfo
  extended_attention_mask = (1.0 - extended_attention_mask) * -3.40282e+38
  extended_attention_mask
end

#get_head_mask(head_mask, num_hidden_layers, is_attention_chunked: false) ⇒ Object 



107
108
109
110
111
112
113
114
115
116
117
118
 
# File 'lib/transformers/modeling_utils.rb', line 107

def get_head_mask(head_mask, num_hidden_layers, is_attention_chunked: false)
  if !head_mask.nil?
    head_mask = _convert_head_mask_to_5d(head_mask, num_hidden_layers)
    if is_attention_chunked == true
      head_mask = head_mask.unsqueeze(-1)
    end
  else
    head_mask = [nil] * num_hidden_layers
  end

  head_mask
end