tokenizers 0.6.4 → 0.7.0

data/ext/tokenizers/src/tokenizer.rs

@@ -1,10 +1,10 @@
-use std::cell::RefCell;
 use std::collections::HashMap;
 use std::path::PathBuf;
 use std::str::FromStr;
+use std::sync::{Arc, RwLock, RwLockReadGuard, RwLockWriteGuard};
 
 use magnus::prelude::*;
-use magnus::{Error, RArray, RHash, RString, Ruby, TryConvert, Value};
+use magnus::{function, method, Error, RArray, RHash, RModule, RString, Ruby, TryConvert, Value};
 use tk::tokenizer::{
     Model, PaddingDirection, PaddingParams, PaddingStrategy, TokenizerImpl, TruncationDirection,
     TruncationParams, TruncationStrategy,
@@ -19,6 +19,7 @@ use super::models::RbModel;
 use super::normalizers::RbNormalizer;
 use super::pre_tokenizers::RbPreTokenizer;
 use super::processors::RbPostProcessor;
+use super::ruby::GvlExt;
 use super::trainers::RbTrainer;
 use super::{RbError, RbResult};
 
@@ -72,7 +73,7 @@ impl From<tk::AddedToken> for RbAddedToken {
             lstrip: Some(token.lstrip),
             rstrip: Some(token.rstrip),
             normalized: Some(token.normalized),
-            special: !token.normalized,
+            special: token.special,
         }
     }
 }
@@ -200,8 +201,8 @@ impl TryConvert for TextEncodeInput<'_> {
         // TODO check if this branch is needed
         if let Ok(arr) = RArray::try_convert(ob) {
             if arr.len() == 2 {
-                let first = arr.entry::<TextInputSequence>(0).unwrap();
-                let second = arr.entry::<TextInputSequence>(1).unwrap();
+                let first = arr.entry::<TextInputSequence>(0)?;
+                let second = arr.entry::<TextInputSequence>(1)?;
                 return Ok(Self((first, second).into()));
             }
         }
@@ -235,8 +236,8 @@ impl TryConvert for PreTokenizedEncodeInput<'_> {
         // TODO check if this branch is needed
         if let Ok(arr) = RArray::try_convert(ob) {
             if arr.len() == 2 {
-                let first = arr.entry::<PreTokenizedInputSequence>(0).unwrap();
-                let second = arr.entry::<PreTokenizedInputSequence>(1).unwrap();
+                let first = arr.entry::<PreTokenizedInputSequence>(0)?;
+                let second = arr.entry::<PreTokenizedInputSequence>(1)?;
                 return Ok(Self((first, second).into()));
             }
         }
@@ -257,202 +258,170 @@ type Tokenizer = TokenizerImpl<RbModel, RbNormalizer, RbPreTokenizer, RbPostProc
 
 #[magnus::wrap(class = "Tokenizers::Tokenizer")]
 pub struct RbTokenizer {
-    tokenizer: RefCell<Tokenizer>,
+    tokenizer: Arc<RwLock<Tokenizer>>,
+}
+
+impl Clone for RbTokenizer {
+    fn clone(&self) -> Self {
+        RbTokenizer {
+            tokenizer: Arc::clone(&self.tokenizer),
+        }
+    }
 }
 
 impl RbTokenizer {
     pub fn new(tokenizer: Tokenizer) -> Self {
         Self {
-            tokenizer: RefCell::new(tokenizer),
+            tokenizer: Arc::new(RwLock::new(tokenizer)),
         }
     }
 
+    /// Acquire the inner tokenizer for reading; surfaces lock poisoning as a
+    /// `PyException` instead of panicking.
+    pub(crate) fn read_inner(&self) -> RbResult<RwLockReadGuard<'_, Tokenizer>> {
+        self.tokenizer
+            .read()
+            .map_err(|_| RbError::new_err("Tokenizer RwLock is poisoned"))
+    }
+
+    /// Acquire the inner tokenizer for writing; surfaces lock poisoning as a
+    /// `PyException` instead of panicking.
+    pub(crate) fn write_inner(&self) -> RbResult<RwLockWriteGuard<'_, Tokenizer>> {
+        self.tokenizer
+            .write()
+            .map_err(|_| RbError::new_err("Tokenizer RwLock is poisoned"))
+    }
+
     pub fn from_model(model: &RbModel) -> Self {
         RbTokenizer::new(TokenizerImpl::new(model.clone()))
     }
 
     pub fn from_str(json: RString) -> RbResult<Self> {
-        Tokenizer::from_str(unsafe { json.as_str()? })
-            .map(|v| RbTokenizer {
-                tokenizer: RefCell::new(v),
-            })
-            .map_err(RbError::from)
+        let tokenizer = Tokenizer::from_str(unsafe { json.as_str()? }).map_err(RbError::from);
+        Ok(Self::new(tokenizer?))
     }
 
     pub fn from_file(path: PathBuf) -> RbResult<Self> {
-        Tokenizer::from_file(path)
-            .map(|v| RbTokenizer {
-                tokenizer: RefCell::new(v),
-            })
-            .map_err(RbError::from)
+        let tokenizer = Tokenizer::from_file(path).map_err(RbError::from);
+        Ok(Self::new(tokenizer?))
     }
 
     pub fn to_str(&self, pretty: bool) -> RbResult<String> {
-        self.tokenizer
-            .borrow()
-            .to_string(pretty)
-            .map_err(RbError::from)
-    }
-
-    pub fn add_special_tokens(&self, tokens: Vec<String>) -> usize {
-        let tokens: Vec<AddedToken> = tokens.iter().map(|t| AddedToken::from(t, true)).collect();
-        self.tokenizer.borrow_mut().add_special_tokens(&tokens)
-    }
-
-    pub fn train(&self, files: Vec<String>, trainer: Option<&RbTrainer>) -> RbResult<()> {
-        let mut trainer = trainer.map_or_else(
-            || self.tokenizer.borrow().get_model().get_trainer(),
-            |t| t.clone(),
-        );
-        self.tokenizer
-            .borrow_mut()
-            .train_from_files(&mut trainer, files)
-            .map(|_| {})
-            .map_err(RbError::from)
+        self.read_inner()?.to_string(pretty).map_err(RbError::from)
     }
 
     pub fn save(&self, path: String, pretty: bool) -> RbResult<()> {
-        self.tokenizer
-            .borrow()
+        self.read_inner()?
             .save(&path, pretty)
             .map_err(RbError::from)
     }
 
-    pub fn add_tokens(&self, tokens: Vec<String>) -> usize {
-        let tokens: Vec<AddedToken> = tokens.iter().map(|t| AddedToken::from(t, true)).collect();
-        self.tokenizer.borrow_mut().add_tokens(&tokens)
+    pub fn num_special_tokens_to_add(&self, is_pair: bool) -> RbResult<usize> {
+        Ok(self
+            .read_inner()?
+            .get_post_processor()
+            .map_or(0, |p| p.added_tokens(is_pair)))
     }
 
-    pub fn encode(
-        &self,
-        sequence: Value,
-        pair: Option<Value>,
-        is_pretokenized: bool,
-        add_special_tokens: bool,
-    ) -> RbResult<RbEncoding> {
-        let sequence: tk::InputSequence = if is_pretokenized {
-            PreTokenizedInputSequence::try_convert(sequence)?.into()
-        } else {
-            TextInputSequence::try_convert(sequence)?.into()
-        };
-        let input = match pair {
-            Some(pair) => {
-                let pair: tk::InputSequence = if is_pretokenized {
-                    PreTokenizedInputSequence::try_convert(pair)?.into()
-                } else {
-                    TextInputSequence::try_convert(pair)?.into()
-                };
-                tk::EncodeInput::Dual(sequence, pair)
-            }
-            None => tk::EncodeInput::Single(sequence),
-        };
-
-        self.tokenizer
-            .borrow()
-            .encode_char_offsets(input, add_special_tokens)
-            .map(|v| RbEncoding { encoding: v })
-            .map_err(RbError::from)
+    pub fn get_vocab(&self, with_added_tokens: bool) -> RbResult<HashMap<String, u32>> {
+        Ok(self.read_inner()?.get_vocab(with_added_tokens))
     }
 
-    pub fn encode_batch(
-        ruby: &Ruby,
-        rb_self: &Self,
-        input: RArray,
-        is_pretokenized: bool,
-        add_special_tokens: bool,
-    ) -> RbResult<RArray> {
-        let input: Vec<tk::EncodeInput> = input
-            .into_iter()
-            .map(|o| {
-                let input: tk::EncodeInput = if is_pretokenized {
-                    PreTokenizedEncodeInput::try_convert(o)?.into()
-                } else {
-                    TextEncodeInput::try_convert(o)?.into()
-                };
-                Ok(input)
-            })
-            .collect::<RbResult<Vec<tk::EncodeInput>>>()?;
-        rb_self
-            .tokenizer
-            .borrow()
-            .encode_batch_char_offsets(input, add_special_tokens)
-            .map(|encodings| {
-                ruby.ary_from_iter(encodings.into_iter().map(Into::<RbEncoding>::into))
-            })
-            .map_err(RbError::from)
-    }
+    pub fn get_added_tokens_decoder(ruby: &Ruby, rb_self: &Self) -> RbResult<RHash> {
+        let sorted_map = ruby.hash_new();
 
-    pub fn decode(&self, ids: Vec<u32>, skip_special_tokens: bool) -> RbResult<String> {
-        self.tokenizer
-            .borrow()
-            .decode(&ids, skip_special_tokens)
-            .map_err(RbError::from)
-    }
+        for (key, value) in rb_self.read_inner()?.get_added_tokens_decoder() {
+            sorted_map.aset::<u32, RbAddedToken>(key, value.into())?;
+        }
 
-    pub fn decode_batch(
-        &self,
-        sequences: Vec<Vec<u32>>,
-        skip_special_tokens: bool,
-    ) -> RbResult<Vec<String>> {
-        let slices = sequences.iter().map(|v| &v[..]).collect::<Vec<&[u32]>>();
-        self.tokenizer
-            .borrow()
-            .decode_batch(&slices, skip_special_tokens)
-            .map_err(RbError::from)
+        Ok(sorted_map)
     }
 
-    pub fn get_model(&self) -> RbModel {
-        self.tokenizer.borrow().get_model().clone()
+    pub fn get_vocab_size(&self, with_added_tokens: bool) -> RbResult<usize> {
+        Ok(self.read_inner()?.get_vocab_size(with_added_tokens))
     }
 
-    pub fn set_model(&self, model: &RbModel) {
-        self.tokenizer.borrow_mut().with_model(model.clone());
-    }
+    pub fn enable_truncation(
+        ruby: &Ruby,
+        rb_self: &Self,
+        max_length: usize,
+        kwargs: RHash,
+    ) -> RbResult<()> {
+        let mut params = TruncationParams {
+            max_length,
+            ..Default::default()
+        };
 
-    pub fn get_decoder(&self) -> Option<RbDecoder> {
-        self.tokenizer.borrow().get_decoder().cloned()
-    }
+        let value: Value = kwargs.delete(ruby.to_symbol("stride"))?;
+        if !value.is_nil() {
+            params.stride = TryConvert::try_convert(value)?;
+        }
 
-    pub fn set_decoder(&self, decoder: Option<&RbDecoder>) {
-        self.tokenizer.borrow_mut().with_decoder(decoder.cloned());
-    }
+        let value: Value = kwargs.delete(ruby.to_symbol("strategy"))?;
+        if !value.is_nil() {
+            let strategy_str = String::try_convert(value)?;
+            params.strategy = match strategy_str.as_str() {
+                "longest_first" => TruncationStrategy::LongestFirst,
+                "only_first" => TruncationStrategy::OnlyFirst,
+                "only_second" => TruncationStrategy::OnlySecond,
+                _ => return Err(Error::new(
+                    ruby.exception_arg_error(),
+                    "The strategy value must be 'longest_first', 'only_first', or 'only_second'",
+                )),
+            }
+        }
 
-    pub fn get_pre_tokenizer(&self) -> Option<RbPreTokenizer> {
-        self.tokenizer.borrow().get_pre_tokenizer().cloned()
-    }
+        let value: Value = kwargs.delete(ruby.to_symbol("direction"))?;
+        if !value.is_nil() {
+            let dir_str = String::try_convert(value)?;
+            params.direction = match dir_str.as_str() {
+                "left" => TruncationDirection::Left,
+                "right" => TruncationDirection::Right,
+                _ => {
+                    return Err(Error::new(
+                        ruby.exception_arg_error(),
+                        "The direction value must be 'left' or 'right'",
+                    ))
+                }
+            }
+        }
 
-    pub fn set_pre_tokenizer(&self, pretok: Option<&RbPreTokenizer>) {
-        self.tokenizer
-            .borrow_mut()
-            .with_pre_tokenizer(pretok.cloned());
-    }
+        if !kwargs.is_empty() {
+            // TODO improve message
+            return Err(Error::new(ruby.exception_arg_error(), "unknown keyword"));
+        }
 
-    pub fn get_post_processor(&self) -> Option<RbPostProcessor> {
-        self.tokenizer.borrow().get_post_processor().cloned()
-    }
+        if let Err(error_message) = rb_self.write_inner()?.with_truncation(Some(params)) {
+            return Err(Error::new(
+                ruby.exception_arg_error(),
+                error_message.to_string(),
+            ));
+        }
 
-    pub fn set_post_processor(&self, processor: Option<&RbPostProcessor>) {
-        self.tokenizer
-            .borrow_mut()
-            .with_post_processor(processor.cloned());
+        Ok(())
     }
 
-    pub fn get_normalizer(&self) -> Option<RbNormalizer> {
-        self.tokenizer.borrow().get_normalizer().cloned()
+    pub fn no_truncation(&self) -> RbResult<()> {
+        self.write_inner()?
+            .with_truncation(None)
+            .expect("Failed to set truncation to `None`! This should never happen");
+        Ok(())
     }
 
-    pub fn set_normalizer(&self, normalizer: Option<&RbNormalizer>) {
-        self.tokenizer
-            .borrow_mut()
-            .with_normalizer(normalizer.cloned());
-    }
+    pub fn get_truncation(ruby: &Ruby, rb_self: &Self) -> RbResult<Option<RHash>> {
+        rb_self
+            .read_inner()?
+            .get_truncation()
+            .map_or(Ok(None), |params| {
+                let ret_hash = ruby.hash_new();
 
-    pub fn token_to_id(&self, token: String) -> Option<u32> {
-        self.tokenizer.borrow().token_to_id(&token)
-    }
+                ret_hash.aset("max_length", params.max_length)?;
+                ret_hash.aset("stride", params.stride)?;
+                ret_hash.aset("strategy", params.strategy.as_ref())?;
+                ret_hash.aset("direction", params.direction.as_ref())?;
 
-    pub fn id_to_token(&self, id: u32) -> Option<String> {
-        self.tokenizer.borrow().id_to_token(id)
+                Ok(Some(ret_hash))
+            })
     }
 
     // TODO support more kwargs
@@ -506,19 +475,19 @@ impl RbTokenizer {
             return Err(Error::new(ruby.exception_arg_error(), "unknown keyword"));
         }
 
-        rb_self.tokenizer.borrow_mut().with_padding(Some(params));
+        rb_self.write_inner()?.with_padding(Some(params));
 
         Ok(())
     }
 
-    pub fn no_padding(&self) {
-        self.tokenizer.borrow_mut().with_padding(None);
+    pub fn no_padding(&self) -> RbResult<()> {
+        self.write_inner()?.with_padding(None);
+        Ok(())
     }
 
-    pub fn padding(ruby: &Ruby, rb_self: &Self) -> RbResult<Option<RHash>> {
+    pub fn get_padding(ruby: &Ruby, rb_self: &Self) -> RbResult<Option<RHash>> {
         rb_self
-            .tokenizer
-            .borrow()
+            .read_inner()?
             .get_padding()
             .map_or(Ok(None), |params| {
                 let ret_hash = ruby.hash_new();
@@ -540,112 +509,280 @@ impl RbTokenizer {
             })
     }
 
-    pub fn enable_truncation(
-        ruby: &Ruby,
-        rb_self: &Self,
-        max_length: usize,
-        kwargs: RHash,
-    ) -> RbResult<()> {
-        let mut params = TruncationParams {
-            max_length,
-            ..Default::default()
+    pub fn encode(
+        &self,
+        sequence: Value,
+        pair: Option<Value>,
+        is_pretokenized: bool,
+        add_special_tokens: bool,
+    ) -> RbResult<RbEncoding> {
+        let sequence: tk::InputSequence = if is_pretokenized {
+            PreTokenizedInputSequence::try_convert(sequence)?.into()
+        } else {
+            TextInputSequence::try_convert(sequence)?.into()
+        };
+        let input = match pair {
+            Some(pair) => {
+                let pair: tk::InputSequence = if is_pretokenized {
+                    PreTokenizedInputSequence::try_convert(pair)?.into()
+                } else {
+                    TextInputSequence::try_convert(pair)?.into()
+                };
+                tk::EncodeInput::Dual(sequence, pair)
+            }
+            None => tk::EncodeInput::Single(sequence),
         };
 
-        let value: Value = kwargs.delete(ruby.to_symbol("stride"))?;
-        if !value.is_nil() {
-            params.stride = TryConvert::try_convert(value)?;
-        }
+        self.read_inner()?
+            .encode_char_offsets(input, add_special_tokens)
+            .map(|v| RbEncoding { encoding: v })
+            .map_err(RbError::from)
+    }
 
-        let value: Value = kwargs.delete(ruby.to_symbol("strategy"))?;
-        if !value.is_nil() {
-            let strategy_str = String::try_convert(value)?;
-            params.strategy = match strategy_str.as_str() {
-                "longest_first" => TruncationStrategy::LongestFirst,
-                "only_first" => TruncationStrategy::OnlyFirst,
-                "only_second" => TruncationStrategy::OnlySecond,
-                _ => return Err(Error::new(
-                    ruby.exception_arg_error(),
-                    "The strategy value must be 'longest_first', 'only_first', or 'only_second'",
-                )),
-            }
-        }
+    pub fn encode_batch(
+        ruby: &Ruby,
+        rb_self: &Self,
+        input: RArray,
+        is_pretokenized: bool,
+        add_special_tokens: bool,
+    ) -> RbResult<RArray> {
+        let input: Vec<tk::EncodeInput> = input
+            .into_iter()
+            .map(|o| {
+                let input: tk::EncodeInput = if is_pretokenized {
+                    PreTokenizedEncodeInput::try_convert(o)?.into()
+                } else {
+                    TextEncodeInput::try_convert(o)?.into()
+                };
+                Ok(input)
+            })
+            .collect::<RbResult<Vec<tk::EncodeInput>>>()?;
+        ruby.detach(|| {
+            rb_self
+                .tokenizer
+                .read()
+                .unwrap()
+                .encode_batch_char_offsets(input, add_special_tokens)
+        })
+        .map(|encodings| ruby.ary_from_iter(encodings.into_iter().map(Into::<RbEncoding>::into)))
+        .map_err(RbError::from)
+    }
+
+    pub fn encode_batch_fast(
+        ruby: &Ruby,
+        rb_self: &Self,
+        input: RArray,
+        is_pretokenized: bool,
+        add_special_tokens: bool,
+    ) -> RbResult<RArray> {
+        let mut items = Vec::<tk::EncodeInput>::with_capacity(input.len());
+        for item in input {
+            let item: tk::EncodeInput = if is_pretokenized {
+                PreTokenizedEncodeInput::try_convert(item)?.into()
+            } else {
+                TextEncodeInput::try_convert(item)?.into()
+            };
+            items.push(item);
+        }
+        ruby.detach(|| {
+            rb_self
+                .tokenizer
+                .read()
+                .unwrap()
+                .encode_batch_fast(items, add_special_tokens)
+        })
+        .map(|encodings| ruby.ary_from_iter(encodings.into_iter().map(Into::<RbEncoding>::into)))
+        .map_err(RbError::from)
+    }
 
-        let value: Value = kwargs.delete(ruby.to_symbol("direction"))?;
-        if !value.is_nil() {
-            let dir_str = String::try_convert(value)?;
-            params.direction = match dir_str.as_str() {
-                "left" => TruncationDirection::Left,
-                "right" => TruncationDirection::Right,
-                _ => {
-                    return Err(Error::new(
-                        ruby.exception_arg_error(),
-                        "The direction value must be 'left' or 'right'",
-                    ))
-                }
-            }
-        }
+    pub fn decode(&self, ids: Vec<u32>, skip_special_tokens: bool) -> RbResult<String> {
+        self.read_inner()?
+            .decode(&ids, skip_special_tokens)
+            .map_err(RbError::from)
+    }
 
-        if !kwargs.is_empty() {
-            // TODO improve message
-            return Err(Error::new(ruby.exception_arg_error(), "unknown keyword"));
-        }
+    pub fn decode_batch(
+        ruby: &Ruby,
+        rb_self: &Self,
+        sequences: Vec<Vec<u32>>,
+        skip_special_tokens: bool,
+    ) -> RbResult<Vec<String>> {
+        ruby.detach(|| {
+            let slices = sequences.iter().map(|v| &v[..]).collect::<Vec<&[u32]>>();
+            rb_self
+                .tokenizer
+                .read()
+                .unwrap()
+                .decode_batch(&slices, skip_special_tokens)
+        })
+        .map_err(RbError::from)
+    }
 
-        if let Err(error_message) = rb_self.tokenizer.borrow_mut().with_truncation(Some(params)) {
-            return Err(Error::new(
-                ruby.exception_arg_error(),
-                error_message.to_string(),
-            ));
-        }
+    pub fn token_to_id(&self, token: String) -> RbResult<Option<u32>> {
+        Ok(self.read_inner()?.token_to_id(&token))
+    }
+
+    pub fn id_to_token(&self, id: u32) -> RbResult<Option<String>> {
+        Ok(self.read_inner()?.id_to_token(id))
+    }
 
+    pub fn set_encode_special_tokens(&self, value: bool) -> RbResult<()> {
+        self.write_inner()?.set_encode_special_tokens(value);
         Ok(())
     }
 
-    pub fn no_truncation(&self) {
-        self.tokenizer
-            .borrow_mut()
-            .with_truncation(None)
-            .expect("Failed to set truncation to `None`! This should never happen");
+    pub fn get_encode_special_tokens(&self) -> RbResult<bool> {
+        Ok(self.read_inner()?.get_encode_special_tokens())
     }
 
-    pub fn truncation(ruby: &Ruby, rb_self: &Self) -> RbResult<Option<RHash>> {
-        rb_self
-            .tokenizer
-            .borrow()
-            .get_truncation()
-            .map_or(Ok(None), |params| {
-                let ret_hash = ruby.hash_new();
+    pub fn add_tokens(&self, tokens: Vec<String>) -> RbResult<usize> {
+        let tokens: Vec<AddedToken> = tokens.iter().map(|t| AddedToken::from(t, true)).collect();
+        self.write_inner()?
+            .add_tokens(tokens)
+            .map_err(RbError::from)
+    }
 
-                ret_hash.aset("max_length", params.max_length)?;
-                ret_hash.aset("stride", params.stride)?;
-                ret_hash.aset("strategy", params.strategy.as_ref())?;
-                ret_hash.aset("direction", params.direction.as_ref())?;
+    pub fn add_special_tokens(&self, tokens: Vec<String>) -> RbResult<usize> {
+        let tokens: Vec<AddedToken> = tokens.iter().map(|t| AddedToken::from(t, true)).collect();
+        self.write_inner()?
+            .add_special_tokens(tokens)
+            .map_err(RbError::from)
+    }
 
-                Ok(Some(ret_hash))
-            })
+    pub fn train(&self, files: Vec<String>, trainer: Option<&RbTrainer>) -> RbResult<()> {
+        let mut trainer = match trainer {
+            Some(t) => t.clone(),
+            None => self.read_inner()?.get_model().get_trainer(),
+        };
+        self.write_inner()?
+            .train_from_files(&mut trainer, files)
+            .map(|_| {})
+            .map_err(RbError::from)
     }
 
-    pub fn num_special_tokens_to_add(&self, is_pair: bool) -> usize {
-        self.tokenizer
-            .borrow()
-            .get_post_processor()
-            .map_or(0, |p| p.added_tokens(is_pair))
+    pub fn get_model(&self) -> RbResult<RbModel> {
+        Ok(self.read_inner()?.get_model().clone())
     }
 
-    pub fn vocab(&self, with_added_tokens: bool) -> HashMap<String, u32> {
-        self.tokenizer.borrow().get_vocab(with_added_tokens)
+    pub fn set_model(&self, model: &RbModel) -> RbResult<()> {
+        self.write_inner()?.with_model(model.clone());
+        Ok(())
     }
 
-    pub fn vocab_size(&self, with_added_tokens: bool) -> usize {
-        self.tokenizer.borrow().get_vocab_size(with_added_tokens)
+    pub fn get_normalizer(&self) -> RbResult<Option<RbNormalizer>> {
+        Ok(self.read_inner()?.get_normalizer().cloned())
     }
 
-    pub fn get_added_tokens_decoder(ruby: &Ruby, rb_self: &Self) -> RbResult<RHash> {
-        let sorted_map = ruby.hash_new();
+    pub fn set_normalizer(&self, normalizer: Option<&RbNormalizer>) -> RbResult<()> {
+        self.write_inner()?
+            .with_normalizer(normalizer.cloned())
+            .map(|_| ())
+            .map_err(RbError::from)
+    }
 
-        for (key, value) in rb_self.tokenizer.borrow().get_added_tokens_decoder() {
-            sorted_map.aset::<u32, RbAddedToken>(key, value.into())?;
-        }
+    pub fn get_pre_tokenizer(&self) -> RbResult<Option<RbPreTokenizer>> {
+        Ok(self.read_inner()?.get_pre_tokenizer().cloned())
+    }
 
-        Ok(sorted_map)
+    pub fn set_pre_tokenizer(&self, pretok: Option<&RbPreTokenizer>) -> RbResult<()> {
+        self.write_inner()?.with_pre_tokenizer(pretok.cloned());
+        Ok(())
+    }
+
+    pub fn get_post_processor(&self) -> RbResult<Option<RbPostProcessor>> {
+        Ok(self.read_inner()?.get_post_processor().cloned())
     }
+
+    pub fn set_post_processor(&self, processor: Option<&RbPostProcessor>) -> RbResult<()> {
+        self.write_inner()?.with_post_processor(processor.cloned());
+        Ok(())
+    }
+
+    pub fn get_decoder(&self) -> RbResult<Option<RbDecoder>> {
+        Ok(self.read_inner()?.get_decoder().cloned())
+    }
+
+    pub fn set_decoder(&self, decoder: Option<&RbDecoder>) -> RbResult<()> {
+        self.write_inner()?.with_decoder(decoder.cloned());
+        Ok(())
+    }
+}
+
+pub fn init_tokenizer(ruby: &Ruby, module: &RModule) -> RbResult<()> {
+    let class = module.define_class("Tokenizer", ruby.class_object())?;
+    class.define_singleton_method("new", function!(RbTokenizer::from_model, 1))?;
+    class.define_singleton_method("from_str", function!(RbTokenizer::from_str, 1))?;
+    class.define_singleton_method("from_file", function!(RbTokenizer::from_file, 1))?;
+    class.define_method("_to_s", method!(RbTokenizer::to_str, 1))?;
+    class.define_method("_save", method!(RbTokenizer::save, 2))?;
+    class.define_method(
+        "num_special_tokens_to_add",
+        method!(RbTokenizer::num_special_tokens_to_add, 1),
+    )?;
+    class.define_method("_vocab", method!(RbTokenizer::get_vocab, 1))?;
+    class.define_method(
+        "added_tokens_decoder",
+        method!(RbTokenizer::get_added_tokens_decoder, 0),
+    )?;
+    class.define_method("_vocab_size", method!(RbTokenizer::get_vocab_size, 1))?;
+    class.define_method(
+        "_enable_truncation",
+        method!(RbTokenizer::enable_truncation, 2),
+    )?;
+    class.define_method("no_truncation", method!(RbTokenizer::no_truncation, 0))?;
+    class.define_method("truncation", method!(RbTokenizer::get_truncation, 0))?;
+    class.define_method("_enable_padding", method!(RbTokenizer::enable_padding, 1))?;
+    class.define_method("no_padding", method!(RbTokenizer::no_padding, 0))?;
+    class.define_method("padding", method!(RbTokenizer::get_padding, 0))?;
+    class.define_method("_encode", method!(RbTokenizer::encode, 4))?;
+    class.define_method("_encode_batch", method!(RbTokenizer::encode_batch, 3))?;
+    class.define_method(
+        "_encode_batch_fast",
+        method!(RbTokenizer::encode_batch_fast, 3),
+    )?;
+    class.define_method("_decode", method!(RbTokenizer::decode, 2))?;
+    class.define_method("_decode_batch", method!(RbTokenizer::decode_batch, 2))?;
+    class.define_method("token_to_id", method!(RbTokenizer::token_to_id, 1))?;
+    class.define_method("id_to_token", method!(RbTokenizer::id_to_token, 1))?;
+    class.define_method(
+        "encode_special_tokens=",
+        method!(RbTokenizer::set_encode_special_tokens, 1),
+    )?;
+    class.define_method(
+        "encode_special_tokens",
+        method!(RbTokenizer::get_encode_special_tokens, 0),
+    )?;
+    class.define_method("add_tokens", method!(RbTokenizer::add_tokens, 1))?;
+    class.define_method(
+        "add_special_tokens",
+        method!(RbTokenizer::add_special_tokens, 1),
+    )?;
+    class.define_method("train", method!(RbTokenizer::train, 2))?;
+    class.define_method("model", method!(RbTokenizer::get_model, 0))?;
+    class.define_method("model=", method!(RbTokenizer::set_model, 1))?;
+    class.define_method("normalizer", method!(RbTokenizer::get_normalizer, 0))?;
+    class.define_method("normalizer=", method!(RbTokenizer::set_normalizer, 1))?;
+    class.define_method("pre_tokenizer", method!(RbTokenizer::get_pre_tokenizer, 0))?;
+    class.define_method("pre_tokenizer=", method!(RbTokenizer::set_pre_tokenizer, 1))?;
+    class.define_method(
+        "post_processor",
+        method!(RbTokenizer::get_post_processor, 0),
+    )?;
+    class.define_method(
+        "post_processor=",
+        method!(RbTokenizer::set_post_processor, 1),
+    )?;
+    class.define_method("decoder", method!(RbTokenizer::get_decoder, 0))?;
+    class.define_method("decoder=", method!(RbTokenizer::set_decoder, 1))?;
+
+    let class = module.define_class("AddedToken", ruby.class_object())?;
+    class.define_singleton_method("_new", function!(RbAddedToken::new, 2))?;
+    class.define_method("content", method!(RbAddedToken::get_content, 0))?;
+    class.define_method("rstrip", method!(RbAddedToken::get_rstrip, 0))?;
+    class.define_method("lstrip", method!(RbAddedToken::get_lstrip, 0))?;
+    class.define_method("single_word", method!(RbAddedToken::get_single_word, 0))?;
+    class.define_method("normalized", method!(RbAddedToken::get_normalized, 0))?;
+    class.define_method("special", method!(RbAddedToken::get_special, 0))?;
+
+    Ok(())
 }