Skip to content

GitLab

Switch branch/tag
Find file Normal viewHistoryPermalink
test_tokenization_rag.py 4.19 KB
Newer Older
Ola Piktus's avatar
RAG (#6813)  
Ola Piktus committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110 
import json
import os
import shutil
import tempfile
from unittest import TestCase

from transformers.configuration_bart import BartConfig
from transformers.configuration_dpr import DPRConfig
from transformers.file_utils import is_datasets_available, is_faiss_available, is_torch_available
from transformers.testing_utils import require_datasets, require_faiss, require_torch
from transformers.tokenization_bart import BartTokenizer
from transformers.tokenization_bert import VOCAB_FILES_NAMES as DPR_VOCAB_FILES_NAMES
from transformers.tokenization_dpr import DPRQuestionEncoderTokenizer
from transformers.tokenization_roberta import VOCAB_FILES_NAMES as BART_VOCAB_FILES_NAMES


if is_torch_available() and is_datasets_available() and is_faiss_available():
    from transformers.configuration_rag import RagConfig
    from transformers.tokenization_rag import RagTokenizer


@require_faiss
@require_datasets
@require_torch
class RagTokenizerTest(TestCase):
    def setUp(self):
        self.tmpdirname = tempfile.mkdtemp()
        self.retrieval_vector_size = 8

        # DPR tok
        vocab_tokens = [
            "[UNK]",
            "[CLS]",
            "[SEP]",
            "[PAD]",
            "[MASK]",
            "want",
            "##want",
            "##ed",
            "wa",
            "un",
            "runn",
            "##ing",
            ",",
            "low",
            "lowest",
        ]
        dpr_tokenizer_path = os.path.join(self.tmpdirname, "dpr_tokenizer")
        os.makedirs(dpr_tokenizer_path, exist_ok=True)
        self.vocab_file = os.path.join(dpr_tokenizer_path, DPR_VOCAB_FILES_NAMES["vocab_file"])
        with open(self.vocab_file, "w", encoding="utf-8") as vocab_writer:
            vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))

        # BART tok
        vocab = [
            "l",
            "o",
            "w",
            "e",
            "r",
            "s",
            "t",
            "i",
            "d",
            "n",
            "\u0120",
            "\u0120l",
            "\u0120n",
            "\u0120lo",
            "\u0120low",
            "er",
            "\u0120lowest",
            "\u0120newer",
            "\u0120wider",
            "<unk>",
        ]
        vocab_tokens = dict(zip(vocab, range(len(vocab))))
        merges = ["#version: 0.2", "\u0120 l", "\u0120l o", "\u0120lo w", "e r", ""]
        self.special_tokens_map = {"unk_token": "<unk>"}

        bart_tokenizer_path = os.path.join(self.tmpdirname, "bart_tokenizer")
        os.makedirs(bart_tokenizer_path, exist_ok=True)
        self.vocab_file = os.path.join(bart_tokenizer_path, BART_VOCAB_FILES_NAMES["vocab_file"])
        self.merges_file = os.path.join(bart_tokenizer_path, BART_VOCAB_FILES_NAMES["merges_file"])
        with open(self.vocab_file, "w", encoding="utf-8") as fp:
            fp.write(json.dumps(vocab_tokens) + "\n")
        with open(self.merges_file, "w", encoding="utf-8") as fp:
            fp.write("\n".join(merges))

    def get_dpr_tokenizer(self) -> DPRQuestionEncoderTokenizer:
        return DPRQuestionEncoderTokenizer.from_pretrained(os.path.join(self.tmpdirname, "dpr_tokenizer"))

    def get_bart_tokenizer(self) -> BartTokenizer:
        return BartTokenizer.from_pretrained(os.path.join(self.tmpdirname, "bart_tokenizer"))

    def tearDown(self):
        shutil.rmtree(self.tmpdirname)

    def test_save_load_pretrained_with_saved_config(self):

        save_dir = os.path.join(self.tmpdirname, "rag_tokenizer")
        rag_config = RagConfig(question_encoder=DPRConfig().to_dict(), generator=BartConfig().to_dict())
        rag_tokenizer = RagTokenizer(question_encoder=self.get_dpr_tokenizer(), generator=self.get_bart_tokenizer())
        rag_config.save_pretrained(save_dir)
        rag_tokenizer.save_pretrained(save_dir)
        new_rag_tokenizer = RagTokenizer.from_pretrained(save_dir, config=rag_config)
        self.assertIsInstance(new_rag_tokenizer.question_encoder, DPRQuestionEncoderTokenizer)
        self.assertEqual(new_rag_tokenizer.question_encoder.vocab, rag_tokenizer.question_encoder.vocab)
        self.assertIsInstance(new_rag_tokenizer.generator, BartTokenizer)
        self.assertEqual(new_rag_tokenizer.generator.encoder, rag_tokenizer.generator.encoder)