Method: Informers::AutoTokenizer.from_pretrained

Defined in:
lib/informers/tokenizers.rb

.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