File size: 10,527 Bytes
a47ca07
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f2f27e9
a47ca07
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
# tokenization_hyformer.py
from __future__ import annotations
import os, re
from typing import Dict, List, Optional, Union, Any, Iterable
from abc import ABC, abstractmethod

import torch
from transformers import PreTrainedTokenizer, BatchEncoding

# -------------------------
# Minimal base tokenizers
# -------------------------

IGNORE_TOKEN_IDX = -100

TOKEN_DICT = {
    'bos': '<s>',
    'eos': '</s>',
    'pad': '<pad>',
    'unk': '<unk>',
    'mask': '<mask>',
}

TASK_TOKEN_DICT = {
    'lm': '<lm>',
    'prediction': '<cls>',
    'mlm': '<mlm>',
}

MAX_LENGTH = 512


class BaseTokenizer(ABC):
    def __init__(
        self,
        vocabulary_path: str,
        max_length: int = MAX_LENGTH,
        bos_token: str = TOKEN_DICT["bos"],
        eos_token: str = TOKEN_DICT["eos"],
        pad_token: str = TOKEN_DICT["pad"],
        unk_token: Optional[str] = None,
        mask_token: Optional[str] = TOKEN_DICT["mask"],
        task_tokens: Optional[Dict[str, str]] = None,
        **kwargs
    ) -> None:
        self.vocab_file = vocabulary_path
        self.max_length = max_length
        self._setup_special_tokens(bos_token, eos_token, unk_token, pad_token, mask_token, task_tokens)
        self.vocab = self._load_vocab(vocabulary_path)
        self._add_special_tokens_to_vocab()

    def _setup_special_tokens(
        self,
        bos_token: str, eos_token: str, unk_token: Optional[str],
        pad_token: str, mask_token: Optional[str], task_tokens: Optional[Dict[str,str]]
    ) -> None:
        self.special_tokens = {"bos": bos_token, "eos": eos_token, "pad": pad_token}
        if unk_token is not None: self.special_tokens["unk"] = unk_token
        if mask_token is not None: self.special_tokens["mask"] = mask_token
        task_dict = TASK_TOKEN_DICT.copy() if task_tokens is None else task_tokens.copy()
        self.special_tokens.update(task_dict)

    @abstractmethod
    def _load_vocab(self, vocab_file: str) -> Dict[str, int]: ...
    @abstractmethod
    def tokenize(self, text: str) -> List[str]: ...

    def _add_special_tokens_to_vocab(self) -> None:
        next_id = len(self.vocab)
        for _, tok in self.special_tokens.items():
            if tok is not None and tok not in self.vocab:
                self.vocab[tok] = next_id
                next_id += 1
        self.ids_to_tokens = {v: k for k, v in self.vocab.items()}
        self._token_id_cache: Dict[str,int] = {}

    @property
    def pad_token_id(self) -> int:
        return self.vocab[self.special_tokens["pad"]]

    @property
    def bos_token_id(self) -> int:
        return self.vocab[self.special_tokens["bos"]]

    @property
    def eos_token_id(self) -> int:
        return self.vocab[self.special_tokens["eos"]]

    @property
    def unk_token_id(self) -> Optional[int]:
        t = self.special_tokens.get("unk")
        return None if t is None else self.vocab[t]

    @property
    def mask_token_id(self) -> Optional[int]:
        t = self.special_tokens.get("mask")
        return None if t is None else self.vocab[t]

    def __len__(self) -> int:
        return len(self.vocab)

    def convert_tokens_to_ids(self, tokens: List[str]) -> List[int]:
        out: List[int] = []
        for tok in tokens:
            if tok in self._token_id_cache:
                out.append(self._token_id_cache[tok])
            elif tok in self.vocab:
                idx = self.vocab[tok]; self._token_id_cache[tok] = idx; out.append(idx)
            elif "unk" in self.special_tokens and self.unk_token_id is not None:
                out.append(self.unk_token_id)
            else:
                raise KeyError(f"Unknown token '{tok}' and no UNK defined")
        return out

    def all_special_ids(self) -> List[int]:
        return self.convert_tokens_to_ids(list(self.special_tokens.values()))

    def __call__(
        self,
        inputs: Union[str, List[str]],
        task: str,
        padding: bool = False,
        truncation: bool = True,
        **kwargs
    ) -> Dict[str, Any]:
        if isinstance(inputs, str):
            inputs = [inputs]
        batch_ids: List[List[int]] = []
        for text in inputs:
            toks = self.tokenize(text)
            toks.insert(0, self.special_tokens[task] if task in self.special_tokens else TASK_TOKEN_DICT["lm"])
            toks.insert(1, self.special_tokens["bos"])
            toks.append(self.special_tokens["eos"])
            if truncation and len(toks) > self.max_length:
                toks = toks[: self.max_length - 1] + [toks[-1]]
            ids = self.convert_tokens_to_ids(toks)
            batch_ids.append(ids)

        max_len = max(len(x) for x in batch_ids)
        if padding:
            pad = self.pad_token_id
            attn = []
            padded = []
            for ids in batch_ids:
                attn.append([1]*len(ids) + [0]*(max_len - len(ids)))
                padded.append(ids + [pad]*(max_len - len(ids)))
            batch_ids = padded
        else:
            attn = [[1]*len(ids) for ids in batch_ids]

        return {"input_ids": batch_ids, "attention_mask": attn}

    def _join_tokens(self, tokens: List[str]) -> str:
        return ''.join(tokens)


SMILES_REGEX_PATTERN = r"""(\[[^\]]+\]|Br?|Cl?|N|O|S|P|F|I|b|c|n|o|s|p|\(|\)|\.|=|#|-|\+|\\|\/|:|~|@|\?|>>?|\*|\$|%[0-9]{2}|[0-9])"""

class SMILESTokenizer(BaseTokenizer):
    def __init__(self, vocabulary_path: str, regex_pattern: str = SMILES_REGEX_PATTERN, **kwargs) -> None:
        self.regex_pattern = regex_pattern
        self.regex = re.compile(self.regex_pattern)
        super().__init__(vocabulary_path=vocabulary_path, **kwargs)

    def _load_vocab(self, vocab_file: str) -> Dict[str, int]:
        vocab: Dict[str,int] = {}
        with open(vocab_file, "r", encoding="utf-8") as f:
            for i, line in enumerate(f):
                tok = line.strip()
                if tok:
                    vocab[tok] = i
        return vocab

    def tokenize(self, text: str) -> List[str]:
        return self.regex.findall(text)

    @classmethod
    def from_config(cls, config, **kwargs) -> 'SMILESTokenizer':
        init_kwargs = {
            'vocabulary_path': config.vocabulary_path,
            'max_length': getattr(config, 'max_length', 512),
            'task_tokens': getattr(config, 'task_tokens', None)
        }
        init_kwargs.update(getattr(config, 'kwargs', {}) or {})
        init_kwargs.update(kwargs)
        return cls(**init_kwargs)


AA_REGEX_PATTERN = r"([ACDEFGHIKLMNPQRSTVWYX]|[BZO]|U|\-|\.)"

class AATokenizer(SMILESTokenizer):
    def __init__(self, vocabulary_path: str, regex_pattern: str = AA_REGEX_PATTERN, **kwargs) -> None:
        super().__init__(vocabulary_path=vocabulary_path, regex_pattern=regex_pattern, **kwargs)


# -------------------------
# HF wrapper
# -------------------------

class HyformerTokenizer(PreTrainedTokenizer):
    """
    HF-compatible wrapper around the above tokenizers.
    Use `mode="aa"` or `mode="smiles"`. Default 'aa'.
    """
    vocab_files_names = {"vocab_file": "aa_vocab.txt"}
    model_input_names = ["input_ids", "attention_mask"]

    def __init__(
        self,
        vocab_file: str,
        mode: str = "aa",
        max_length: int = 512,
        bos_token: str = "<s>",
        eos_token: str = "</s>",
        pad_token: str = "<pad>",
        unk_token: Optional[str] = "<unk>",
        mask_token: Optional[str] = "<mask>",
        **kwargs,
    ):
        tok_kwargs = dict(vocabulary_path=vocab_file, max_length=max_length)
        if mode == "aa":
            self._inner = AATokenizer(**tok_kwargs)
        elif mode == "smiles":
            self._inner = SMILESTokenizer(**tok_kwargs)
        else:
            raise ValueError("mode must be 'aa' or 'smiles'")

        super().__init__(
            bos_token=bos_token, eos_token=eos_token, pad_token=pad_token,
            unk_token=unk_token, mask_token=mask_token, model_max_length=max_length, **kwargs
        )
        self._vocab_file = vocab_file
        self.mode = mode

    @property
    def vocab_size(self) -> int:
        return len(self._inner)

    def get_vocab(self) -> Dict[str, int]:
        return dict(self._inner.vocab)

    def _convert_token_to_id(self, token: str) -> int:
        if token in self._inner.vocab:
            return self._inner.vocab[token]
        uid = self._inner.unk_token_id
        if uid is None:
            raise KeyError(f"Unknown token '{token}' and no <unk>")
        return uid

    def _convert_id_to_token(self, index: int) -> str:
        return self._inner.ids_to_tokens[index]

    def _tokenize(self, text: str) -> List[str]:
        return self._inner.tokenize(text)

    def build_inputs_with_special_tokens(self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None) -> List[int]:
        return [self._inner.bos_token_id] + token_ids_0 + [self._inner.eos_token_id]

    def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None):
        os.makedirs(save_directory, exist_ok=True)
        out = os.path.join(save_directory, ((filename_prefix + "-") if filename_prefix else "") + "vocab.txt")
        inv = sorted(self._inner.vocab.items(), key=lambda kv: kv[1])
        with open(out, "w", encoding="utf-8") as f:
            for tok, _id in inv:
                f.write(tok + "\n")
        return (out,)

    def convert_tokens_to_string(self, tokens: List[str]) -> str:
        return self._inner._join_tokens(tokens)

    # Let HF callers pass 'task' to add task tokens via our BaseTokenizer batching
    def __call__(
        self,
        text: Union[str, List[str]],
        task: str = "lm",
        padding: Union[bool, str] = False,
        truncation: Union[bool, str] = True,
        return_tensors: Optional[str] = None,
        **kwargs: Any,
    ) -> BatchEncoding:
        out = self._inner(
            inputs=text,
            task=task,
            padding=bool(padding) or (isinstance(padding, str) and padding != "do_not_pad"),
            truncation=bool(truncation) or (isinstance(truncation, str) and truncation != "do_not_truncate"),
        )
        input_ids, attention_mask = out["input_ids"], out["attention_mask"]
        if return_tensors == "pt":
            input_ids = torch.tensor(input_ids, dtype=torch.long)
            attention_mask = torch.tensor(attention_mask, dtype=torch.long)
        return BatchEncoding({"input_ids": input_ids, "attention_mask": attention_mask}, tensor_type="pt" if return_tensors == "pt" else None)