tokenizers 0.2.1-x86_64-darwin → 0.2.3-x86_64-darwin

data/README.md

@@ -24,8 +24,8 @@ Encode
 
 ```ruby
 encoded = tokenizer.encode("I can feel the magic, can you?")
-encoded.ids
 encoded.tokens
+encoded.ids
 ```
 
 Decode

data/lib/tokenizers/char_bpe_tokenizer.rb

@@ -8,8 +8,8 @@ module Tokenizers
       @tokenizer.decoder = BPEDecoder.new
     end
 
-    def encode(text)
-      @tokenizer.encode(text)
+    def encode(text, **options)
+      @tokenizer.encode(text, **options)
     end
 
     def decode(ids)

data/lib/tokenizers/encoding.rb

@@ -0,0 +1,19 @@
+module Tokenizers
+  class Encoding
+    def word_to_tokens(word_index, sequence_index = 0)
+      _word_to_tokens(word_index, sequence_index)
+    end
+
+    def word_to_chars(word_index, sequence_index = 0)
+      _word_to_chars(word_index, sequence_index)
+    end
+
+    def char_to_token(char_pos, sequence_index = 0)
+      _char_to_token(char_pos, sequence_index)
+    end
+
+    def char_to_word(char_pos, sequence_index = 0)
+      _char_to_word(word_index, sequence_index)
+    end
+  end
+end

data/lib/tokenizers/from_pretrained.rb

@@ -0,0 +1,119 @@
+module Tokenizers
+  module FromPretrained
+    # for user agent
+    TOKENIZERS_VERSION = "0.13.2"
+
+    # use Ruby for downloads
+    # this avoids the need to vendor OpenSSL on Linux
+    # and reduces the extension size by about half
+    def from_pretrained(identifier, revision: "main", auth_token: nil)
+      require "cgi"
+      require "digest"
+      require "fileutils"
+      require "json"
+      require "open-uri"
+
+      cache_dir = ensure_cache_dir
+
+      # string options are headers
+      options = {
+        open_timeout: 3,
+        read_timeout: 30,
+        "User-Agent" => "tokenizers/#{TOKENIZERS_VERSION}; bindings/Ruby; version/#{VERSION}"
+      }
+      if auth_token
+        options["Authorization"] = "Bearer #{auth_token}"
+      end
+
+      url = "https://huggingface.co/%s/resolve/%s/tokenizer.json" % [identifier, revision].map { |v| CGI.escape(v) }
+
+      path =
+        begin
+          cached_path(cache_dir, url, options)
+        rescue OpenURI::HTTPError
+          raise Error, "Model \"#{identifier}\" on the Hub doesn't have a tokenizer"
+        end
+
+      from_file(path)
+    end
+
+    private
+
+    # use same storage format as Rust version
+    # https://github.com/epwalsh/rust-cached-path
+    def cached_path(cache_dir, url, options)
+      fsum = Digest::SHA256.hexdigest(url)
+      meta_paths = Dir[File.join(cache_dir, "#{fsum}.*.meta")]
+      meta = meta_paths.map { |f| JSON.load_file(f) }.max_by { |m| m["creation_time"] }
+      etag = meta["etag"] if meta
+
+      if etag
+        esum = Digest::SHA256.hexdigest(etag)
+        resource_path = File.join(cache_dir, "#{fsum}.#{esum}")
+        options["If-None-Match"] = etag if File.exist?(resource_path)
+      end
+
+      options[:content_length_proc] = -> (_) { puts "Downloading..." }
+
+      tempfile =
+        begin
+          URI.open(url, options)
+        rescue OpenURI::HTTPError => e
+          if e.message == "304 Not Modified"
+            return resource_path
+          else
+            raise e
+          end
+        end
+
+      etag = tempfile.meta["etag"]
+      esum = Digest::SHA256.hexdigest(etag)
+      resource_path = File.join(cache_dir, "#{fsum}.#{esum}")
+      meta_path = "#{resource_path}.meta"
+
+      meta = {
+        resource: url,
+        resource_path: resource_path,
+        meta_path: meta_path,
+        etag: etag,
+        expires: nil,
+        creation_time: Time.now.to_f
+      }
+
+      File.write("#{resource_path}.lock", "")
+      File.open(resource_path, "wb") { |f| IO.copy_stream(tempfile, f) }
+      File.write(meta_path, JSON.generate(meta))
+
+      resource_path
+    end
+
+    def cache_dir
+      if ENV["TOKENIZERS_CACHE"]
+        ENV["TOKENIZERS_CACHE"]
+      else
+        # use same directory as Rust version
+        # https://docs.rs/dirs/latest/dirs/fn.cache_dir.html
+        dir =
+          if Gem.win_platform?
+            ENV.fetch("LOCALAPPDATA")
+          elsif mac?
+            File.join(ENV.fetch("HOME"), "Library", "Caches")
+          else
+            ENV["XDG_CACHE_HOME"] || File.join(ENV.fetch("HOME"), ".cache")
+          end
+
+        File.join(dir, "huggingface", "tokenizers")
+      end
+    end
+
+    def ensure_cache_dir
+      dir = cache_dir
+      FileUtils.mkdir_p(dir)
+      dir
+    end
+
+    def mac?
+      RbConfig::CONFIG["host_os"] =~ /darwin/i
+    end
+  end
+end

data/lib/tokenizers/tokenizer.rb

@@ -0,0 +1,12 @@
+module Tokenizers
+  class Tokenizer
+    # TODO change add_special_tokens default to true in 0.3.0
+    def encode(sequence, add_special_tokens: nil)
+      if add_special_tokens.nil?
+        warn "[tokenizers] add_special_tokens will default to true in 0.3.0. Pass add_special_tokens: true/false to silence this warning."
+        add_special_tokens = false
+      end
+      _encode(sequence, add_special_tokens)
+    end
+  end
+end

data/lib/tokenizers/version.rb

@@ -1,3 +1,3 @@
 module Tokenizers
-  VERSION = "0.2.1"
+  VERSION = "0.2.3"
 end

data/lib/tokenizers.rb

@@ -1,18 +1,19 @@
 # ext
 begin
-  require "tokenizers/#{RUBY_VERSION.to_f}/tokenizers"
+  require_relative "tokenizers/#{RUBY_VERSION.to_f}/tokenizers"
 rescue LoadError
-  require "tokenizers/tokenizers"
+  require_relative "tokenizers/tokenizers"
 end
 
 # modules
-require "tokenizers/char_bpe_tokenizer"
-require "tokenizers/version"
+require_relative "tokenizers/char_bpe_tokenizer"
+require_relative "tokenizers/encoding"
+require_relative "tokenizers/from_pretrained"
+require_relative "tokenizers/tokenizer"
+require_relative "tokenizers/version"
 
 module Tokenizers
   class Error < StandardError; end
 
-  def self.from_pretrained(identifier, revision: "main", auth_token: nil)
-    _from_pretrained(identifier, revision, auth_token)
-  end
+  extend FromPretrained
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: tokenizers
 version: !ruby/object:Gem::Version
-  version: 0.2.1
+  version: 0.2.3
 platform: x86_64-darwin
 authors:
 - Andrew Kane
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2023-01-12 00:00:00.000000000 Z
+date: 2023-01-22 00:00:00.000000000 Z
 dependencies: []
 description: 
 email: andrew@ankane.org
@@ -28,6 +28,9 @@ files:
 - lib/tokenizers/3.1/tokenizers.bundle
 - lib/tokenizers/3.2/tokenizers.bundle
 - lib/tokenizers/char_bpe_tokenizer.rb
+- lib/tokenizers/encoding.rb
+- lib/tokenizers/from_pretrained.rb
+- lib/tokenizers/tokenizer.rb
 - lib/tokenizers/version.rb
 homepage: https://github.com/ankane/tokenizers-ruby
 licenses:
@@ -51,7 +54,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 3.4.3
+rubygems_version: 3.4.4
 signing_key: 
 specification_version: 4
 summary: Fast state-of-the-art tokenizers for Ruby