Add in-place train.

bindings/python/src/tokenizer.rs

@@ -667,10 +667,8 @@ impl PyTokenizer {
     }
 
     fn train(&mut self, trainer: &PyTrainer, files: Vec<String>) -> PyResult<()> {
-        self.tokenizer = self
-            .tokenizer
-            .clone()
-            .train(trainer, files)
+        self.tokenizer
+            .train_and_replace(trainer, files)
             .map_err(|e| exceptions::Exception::py_err(format!("{}", e)))?;
         Ok(())
     }

tokenizers/src/tokenizer/mod.rs

@@ -1004,7 +1004,7 @@ where
         Ok(words)
     }
 
-    /// Train a model and replace our current Model, using the given Trainer
+    /// Train a model and return a new Tokenizer, using the given Trainer
     pub fn train<T, TM>(
         self,
         trainer: &T,
@@ -1032,6 +1032,20 @@ where
 
         Ok(new_tok)
     }
+
+    /// Train a model and replace our current Model, using the given Trainer
+    pub fn train_and_replace<T>(&mut self, trainer: &T, files: Vec<String>) -> Result<()>
+    where
+        T: Trainer<Model = M> + Sync,
+    {
+        let words = self.word_count(trainer, files)?;
+
+        let (model, special_tokens) = trainer.train(words)?;
+        self.model = model;
+        self.add_special_tokens(&special_tokens);
+
+        Ok(())
+    }
 }
 
 impl<M, N, PT, PP, D> std::str::FromStr for TokenizerImpl<M, N, PT, PP, D>