OpenBackdoor

1

2

3
BATCH_SIZE = 128
4
MAX_LENGTH = 502
5
MAX_GPT2_LENGTH = 1000
6

7
MAX_PARAPHRASE_LENGTH = 100
8

9
BASE_CONFIG = {
10
    "keys": [
11
        {"key": "sent1_tokens", "position": 3, "tokenize": True, "metadata": False},
12
        {"key": "sent2_tokens", "position": 4, "tokenize": True, "metadata": False},
13
        {"key": "f1_score", "position": 5, "tokenize": False, "metadata": True},
14
        {"key": "kt_score", "position": 6, "tokenize": False, "metadata": True},
15
        {"key": "ed_score", "position": 7, "tokenize": False, "metadata": True},
16
        {"key": "langid", "position": 8, "tokenize": False, "metadata": True},
17
    ],
18
    "max_total_length": MAX_PARAPHRASE_LENGTH,
19
    "max_prefix_length": int(MAX_PARAPHRASE_LENGTH / 2),
20
    "max_suffix_length": int(MAX_PARAPHRASE_LENGTH / 2),
21
    "max_dense_length": 2
22
}
23

24
DATASET_CONFIG = {
25
"datasets/test_custom": BASE_CONFIG,
26
"datasets/paranmt_filtered": BASE_CONFIG,
27
    "datasets/shakespeare": BASE_CONFIG,
28
    "datasets/formality": BASE_CONFIG,
29
    "datasets/shakespeare_aae_tweets_bible_romantic-poetry_switchboard_coha_3_bins_lyrics_full": BASE_CONFIG,
30
    "datasets/aae": BASE_CONFIG,
31
    "datasets/bible": BASE_CONFIG,
32
    "datasets/poetry": BASE_CONFIG,
33
    "datasets/switchboard": BASE_CONFIG,
34
    "datasets/english_tweets": BASE_CONFIG,
35
    "datasets/lyrics_full": BASE_CONFIG,
36
    "datasets/joyce": BASE_CONFIG,
37
    "datasets/congress-bills": BASE_CONFIG,
38
    "datasets/shakespeare": BASE_CONFIG,
39
    "datasets/coha_3_bins_1810s-1820s": BASE_CONFIG,
40
    "datasets/coha_3_bins_1890s-1900s": BASE_CONFIG,
41
    "datasets/coha_3_bins_1990s-2000s": BASE_CONFIG
42
}
43

44
# Fill in DATASET_CONFIG with keys it was missing previously
45
for dataset, config in DATASET_CONFIG.items():
46
    for base_key, base_value in BASE_CONFIG.items():
47
        if base_key not in config:
48
            config[base_key] = base_value
49