Class: Informers::AutoTokenizer

Inherits:
Object
  • Object
show all
Defined in:
lib/informers/tokenizers.rb

Constant Summary collapse

TOKENIZER_CLASS_MAPPING =
{
  "BertTokenizer" => BertTokenizer,
  "DebertaV2Tokenizer" => DebertaV2Tokenizer,
  "DistilBertTokenizer" => DistilBertTokenizer,
  "RobertaTokenizer" => RobertaTokenizer,
  "XLMRobertaTokenizer" => XLMRobertaTokenizer,
  "MPNetTokenizer" => MPNetTokenizer
}

Class Method Summary collapse

Class Method Details

.from_pretrained(pretrained_model_name_or_path, quantized: true, progress_callback: nil, config: nil, cache_dir: nil, local_files_only: false, revision: "main", legacy: nil, **kwargs) ⇒ Object



113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
# File 'lib/informers/tokenizers.rb', line 113

def self.from_pretrained(
  pretrained_model_name_or_path,
  quantized: true,
  progress_callback: nil,
  config: nil,
  cache_dir: nil,
  local_files_only: false,
  revision: "main",
  legacy: nil,
  **kwargs
)
  tokenizer_json, tokenizer_config = load_tokenizer(
    pretrained_model_name_or_path,
    quantized:,
    progress_callback:,
    config:,
    cache_dir:,
    local_files_only:,
    revision:,
    legacy:
  )

  # Some tokenizers are saved with the "Fast" suffix, so we remove that if present.
  tokenizer_name = tokenizer_config["tokenizer_class"]&.delete_suffix("Fast") || "PreTrainedTokenizer"

  cls = TOKENIZER_CLASS_MAPPING[tokenizer_name]
  if !cls
    warn "Unknown tokenizer class #{tokenizer_name.inspect}, attempting to construct from base class."
    cls = PreTrainedTokenizer
  end
  cls.new(tokenizer_json, tokenizer_config)
end

.load_tokenizer(pretrained_model_name_or_path, **options) ⇒ Object



146
147
148
149
150
151
152
153
154
155
156
157
# File 'lib/informers/tokenizers.rb', line 146

def self.load_tokenizer(pretrained_model_name_or_path, **options)
  info = [
    Utils::Hub.get_model_file(pretrained_model_name_or_path, "tokenizer.json", true, **options),
    Utils::Hub.get_model_json(pretrained_model_name_or_path, "tokenizer_config.json", true, **options),
  ]

  # Override legacy option if `options.legacy` is not null
  if !options[:legacy].nil?
    info[1]["legacy"] = options[:legacy]
  end
  info
end