SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Add BlenderbotTokenizerFast (#13720)

Browse Source 
* Add the support for the fast (rust) implementation of BlenbderbotTokenizer

* Fix a converter and a typo in a doc

* Apply the patil-suraj's suggestion

* (Nitpick) Fast tokenization -> Fast Tokenization in doc

* Apply the SaulLu's suggestion

* Apply Narsil's suggestion to fix test pipelines

* Add encoder_no_repeat_ngram_size according to the Narsil's suggestion

* Revert the last (unnecessary) commit

* Override pipeline config for Blenderbot to allow for larger pos. emb.

* make fix-copies
This commit is contained in:
Daniel Stancl
2021-10-29 15:19:01 +02:00
committed by GitHub
parent 5b45422b58
commit d37f1fb8ba
12 changed files with 181 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
tests/test_tokenization_blenderbot.py
View File

@@ -16,8 +16,8 @@
"""Tests for Blenderbot Tokenizers, including common tests for BlenderbotSmallTokenizer."""
import unittest
from transformers import BlenderbotTokenizer, BlenderbotTokenizerFast
from transformers.file_utils import cached_property
from transformers.models.blenderbot.tokenization_blenderbot import BlenderbotTokenizer
class Blenderbot3BTokenizerTests(unittest.TestCase):
@@ -25,6 +25,10 @@ class Blenderbot3BTokenizerTests(unittest.TestCase):
def tokenizer_3b(self):
return BlenderbotTokenizer.from_pretrained("facebook/blenderbot-3B")
@cached_property
def rust_tokenizer_3b(self):
return BlenderbotTokenizerFast.from_pretrained("facebook/blenderbot-3B")
def test_encode_decode_cycle(self):
tok = self.tokenizer_3b
src_text = " I am a small frog."
@@ -32,6 +36,17 @@ class Blenderbot3BTokenizerTests(unittest.TestCase):
decoded = tok.batch_decode(encoded, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
assert src_text == decoded
def test_encode_decode_cycle_rust_tokenizer(self):
tok = self.rust_tokenizer_3b
src_text = " I am a small frog."
encoded = tok([src_text], padding=False, truncation=False)["input_ids"]
decoded = tok.batch_decode(encoded, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
assert src_text == decoded
def test_3B_tokenization_same_as_parlai(self):
assert self.tokenizer_3b.add_prefix_space
assert self.tokenizer_3b([" Sam", "Sam"]).input_ids == [[5502, 2], [5502, 2]]
def test_3B_tokenization_same_as_parlai_rust_tokenizer(self):
assert self.rust_tokenizer_3b.add_prefix_space
assert self.rust_tokenizer_3b([" Sam", "Sam"]).input_ids == [[5502, 2], [5502, 2]]