Class: Transformers::FeatureExtractionPipeline

Inherits:
Pipeline
  • Object
show all
Defined in:
lib/transformers/pipelines/feature_extraction.rb

Instance Method Summary collapse

Methods inherited from Pipeline

#_ensure_tensor_on_device, #call, #check_model_type, #get_iterator, #initialize, #torch_dtype

Constructor Details

This class inherits a constructor from Transformers::Pipeline

Instance Method Details

#_forward(model_inputs) ⇒ Object 



31
32
33
34
 
# File 'lib/transformers/pipelines/feature_extraction.rb', line 31

def _forward(model_inputs)
  model_outputs = @model.(**model_inputs)
  model_outputs
end

#_sanitize_parameters(truncation: nil, tokenize_kwargs: nil, return_tensors: nil, **kwargs) ⇒ Object 



3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
 
# File 'lib/transformers/pipelines/feature_extraction.rb', line 3

def _sanitize_parameters(truncation: nil, tokenize_kwargs: nil, return_tensors: nil, **kwargs)
  if tokenize_kwargs.nil?
    tokenize_kwargs = {}
  end

  if !truncation.nil?
    if tokenize_kwargs.include?(:truncation)
      raise ArgumentError,
        "truncation parameter defined twice (given as keyword argument as well as in tokenize_kwargs)"
    end
    tokenize_kwargs[:truncation] = truncation
  end

  preprocess_params = tokenize_kwargs

  postprocess_params = {}
  if !return_tensors.nil?
    postprocess_params[:return_tensors] = return_tensors
  end

  [preprocess_params, {}, postprocess_params]
end

#postprocess(model_outputs, return_tensors: false) ⇒ Object 



36
37
38
39
40
41
42
43
44
45
 
# File 'lib/transformers/pipelines/feature_extraction.rb', line 36

def postprocess(model_outputs, return_tensors: false)
  # [0] is the first available tensor, logits or last_hidden_state.
  if return_tensors
    model_outputs[0]
  elsif @framework == "pt"
    model_outputs[0].to_a
  elsif @framework == "tf"
    raise Todo
  end
end

#preprocess(inputs, **tokenize_kwargs) ⇒ Object 



26
27
28
29
 
# File 'lib/transformers/pipelines/feature_extraction.rb', line 26

def preprocess(inputs, **tokenize_kwargs)
  model_inputs = @tokenizer.(inputs, return_tensors: @framework, **tokenize_kwargs)
  model_inputs
end