File size: 6,923 Bytes
a476bbf |
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 |
import torch
from typing import List, Union, Optional
from rust_trie import Trie
import os
class Tokenizer:
def __init__(self, tokens: List[str], unk_token_id: Optional[int] = None):
self.ids_to_tokens = tokens
self.trie = Trie(unk_token_id)
for token in tokens:
self.trie.add(token)
# If unk_token_id is not provided, add <unk> to the end of the tokens list
if unk_token_id is None:
self.ids_to_tokens += ["<unk>"]
self.pad_token_id = self.ids_to_tokens.index("<pad>")
self.mask_token_id = self.ids_to_tokens.index("<mask>")
def __call__(self, sequences: Union[str, List], *args, **kwargs):
if isinstance(sequences, str):
return self.encode(sequences, *args, **kwargs)
else:
return self.batch_encode(sequences, *args, **kwargs)
def encode(
self,
sequence: str,
add_special_tokens: bool = False,
return_tensor: bool = False,
max_sequence_length: Optional[int] = None,
) -> List[int]:
if max_sequence_length is not None:
if add_special_tokens:
max_sequence_length -= 2
if len(sequence) > max_sequence_length:
# randomly crop the sequence
start_idx = torch.randint(
0, len(sequence) - max_sequence_length + 1, (1,)
)
sequence = sequence[start_idx : start_idx + max_sequence_length]
if add_special_tokens:
sequence = "<cls>" + sequence + "<eos>"
output = self.trie.tokenize(sequence)
if return_tensor:
output = torch.tensor(output, dtype=torch.long)
return output
def batch_encode(
self,
sequences: List[str],
add_special_tokens: bool = False,
return_tensors: bool = False,
max_sequence_length: Optional[int] = None,
) -> List[List[int]]:
output = []
if max_sequence_length is None and return_tensors:
max_sequence_length = max([len(sequence) for sequence in sequences])
if add_special_tokens:
max_sequence_length += 2
# if max_sequence_length is not None:
# sequences = [
# sequence[
# : (max_sequence_length - 2)
# if add_special_tokens
# else max_sequence_length
# ]
# for sequence in sequences
# ]
for sequence in sequences:
output.append(
self.encode(
sequence,
add_special_tokens,
return_tensors,
max_sequence_length=max_sequence_length,
)
)
if return_tensors:
tensor_out = torch.full(
(len(output), max_sequence_length), self.pad_token_id
)
for i, sequence in enumerate(output):
tensor_out[i, : len(sequence)] = sequence
output = tensor_out
return output
def decode(self, tokens: List[int]) -> str:
return "".join([self.ids_to_tokens[idx] for idx in tokens])
class EsmTokenizer(Tokenizer):
def __init__(self):
tokens = [
"<cls>",
"<pad>",
"<eos>",
"<unk>",
"L",
"A",
"G",
"V",
"S",
"E",
"R",
"T",
"I",
"D",
"P",
"K",
"Q",
"N",
"F",
"Y",
"M",
"H",
"W",
"C",
"X",
"B",
"U",
"Z",
"O",
".",
"-",
"<null_1>",
"<mask>",
]
super().__init__(tokens, unk_token_id=3)
class PTMTokenizer(Tokenizer):
def __init__(self):
tokens = [
"<cls>",
"<pad>",
"<eos>",
"<unk>",
".",
"-",
"<null_1>",
"<mask>",
"L",
"A",
"G",
"V",
"S",
"E",
"R",
"T",
"I",
"D",
"P",
"K",
"Q",
"N",
"F",
"Y",
"M",
"H",
"W",
"C",
"X",
"B",
"U",
"Z",
"O",
"PTM",
"<N-linked (GlcNAc...) asparagine>",
"<Pyrrolidone carboxylic acid>",
"<Phosphoserine>",
"<Phosphothreonine>",
"<N-acetylalanine>",
"<N-acetylmethionine>",
"<N6-acetyllysine>",
"<Phosphotyrosine>",
"<S-diacylglycerol cysteine>",
"<N6-(pyridoxal phosphate)lysine>",
"<N-acetylserine>",
"<N6-carboxylysine>",
"<N6-succinyllysine>",
"<S-palmitoyl cysteine>",
"<O-(pantetheine 4'-phosphoryl)serine>",
"<Sulfotyrosine>",
"<O-linked (GalNAc...) threonine>",
"<Omega-N-methylarginine>",
"<N-myristoyl glycine>",
"<4-hydroxyproline>",
"<Asymmetric dimethylarginine>",
"<N5-methylglutamine>",
"<4-aspartylphosphate>",
"<S-geranylgeranyl cysteine>",
"<4-carboxyglutamate>",
]
super().__init__(tokens, unk_token_id=3)
self.ptm_token_start = self.ids_to_tokens.index("PTM")
def is_ptm_token(self, input_ids: torch.tensor):
return input_ids > self.ptm_token_start
def is_special_token(self, input_ids: torch.tensor):
l_id = self.ids_to_tokens.index("L")
return input_ids < l_id
def __len__(self):
return len(self.ids_to_tokens)
def get_vocab_size(self):
return len(self.ids_to_tokens)
class AptTokenizer(Tokenizer):
def __init__(self):
# For our own tokenizers, we don't need to explicitly add the <unk> token
# because it gets added as the last token in the tokens list
# I've also removed X so that it gets translated to <unk>
tokens = [
"<cls>",
"<pad>",
"<eos>",
"L",
"A",
"G",
"V",
"S",
"E",
"R",
"T",
"I",
"D",
"P",
"K",
"Q",
"N",
"F",
"Y",
"M",
"H",
"W",
"C",
"B",
"U",
"Z",
"O",
"<mask>",
]
super().__init__(tokens)
|