Add tokenizer file

japanese-dummy-tokenizer/special_tokens_map.json

@@ -0,0 +1 @@
+{"eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>"}

japanese-dummy-tokenizer/tokenizer_config.json

@@ -0,0 +1 @@
+{"eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>", "extra_ids": 0, "additional_special_tokens": null, "special_tokens_map_file": "/home/patrick/.cache/torch/transformers/685ac0ca8568ec593a48b61b0a3c272beee9bc194a3c7241d15dcadb5f875e53.f76030f3ec1b96a8199b2593390c610e76ca8028ef3d24680000619ffb646276", "name_or_path": "csebuetnlp/mT5_multilingual_XLSum", "sp_model_kwargs": {}, "tokenizer_class": "T5Tokenizer"}

tokenizer.py

@@ -0,0 +1,28 @@
+from datasets import load_dataset
+from transformers import AutoTokenizer
+from huggingface_hub import Repository
+
+repo = Repository(".", clone_from="ybelkada/japanese-dummy-tokenizer")
+
+def get_training_corpus(dataset):
+    """
+    Returns the training corpus for the given dataset.
+    """
+    return (element['original_ja'] for element in iter(dataset))
+
+dataset = load_dataset("snow_simplified_japanese_corpus", streaming=True, split="train")
+
+train_dataset = dataset.skip(100)
+val_dataset = dataset.take(100)
+
+old_tokenizer = AutoTokenizer.from_pretrained("rinna/japanese-gpt2-small")
+old_tokenizer = AutoTokenizer.from_pretrained("csebuetnlp/mT5_multilingual_XLSum")
+
+print("Old Tokenizer:", old_tokenizer.tokenize("誰が一番に着くか私には分かりません。"))
+new_tokenizer = old_tokenizer.train_new_from_iterator(get_training_corpus(train_dataset), 52000)
+
+print("New Tokenizer:",new_tokenizer.tokenize("誰が一番に着くか私には分かりません。"))
+new_tokenizer.save_pretrained("japanese-dummy-tokenizer")
+repo.git_add()
+repo.git_commit("Add tokenizer file")
+repo.git_push()