File size: 5,957 Bytes
2e30a77 | 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 | from pathlib import Path
import json
import heapq
class SutraTokenizerFast:
def __init__(self, directory):
directory = Path(directory)
with open(directory / "vocab.json", "r", encoding="utf-8") as f:
self.vocab = json.load(f)
with open(directory / "merges.json", "r", encoding="utf-8") as f:
self.merges = [tuple(pair) for pair in json.load(f)]
self.token_to_id = self.vocab
# Invert vocab for fast decode lookup
self.id_to_token = {
int(idx): token
for token, idx in self.vocab.items()
}
# Resolve special token IDs from vocabulary
self.unk_id = self.vocab.get("<|unk|>")
self.pad_token_id = self.vocab.get("<|pad|>")
self.bos_token_id = self.vocab.get("<|bos|>")
self.eos_token_id = self.vocab.get("<|eos|>")
self.unk_token_id = self.unk_id
# Set of special token IDs for filtering during decode
self.special_token_ids = {
self.pad_token_id,
self.bos_token_id,
self.eos_token_id,
self.unk_token_id,
} - {None}
self.merge_rank = {
pair: rank
for rank, pair in enumerate(self.merges)
}
def encode(self, text):
# -------------------------------------------------
# Normalize spaces
# -------------------------------------------------
text = text.replace(" ", "▁")
symbols = list(text)
n = len(symbols)
if n == 0:
return []
if n == 1:
return [
self.token_to_id.get(
symbols[0],
self.unk_id
)
]
merge_rank = self.merge_rank
token_to_id = self.token_to_id
unk_id = self.unk_id
# -------------------------------------------------
# Build initial heap
# -------------------------------------------------
heap = []
for i in range(n - 1):
pair = (
symbols[i],
symbols[i + 1]
)
rank = merge_rank.get(pair)
if rank is not None:
heapq.heappush(
heap,
(rank, i)
)
# -------------------------------------------------
# Linked list
# -------------------------------------------------
previous = [i - 1 for i in range(n)]
following = [i + 1 for i in range(n)]
following[-1] = -1
alive = bytearray(b"\x01") * n
# -------------------------------------------------
# BPE merge loop
# -------------------------------------------------
while heap:
rank, left = heapq.heappop(heap)
if not alive[left]:
continue
right = following[left]
if right == -1 or not alive[right]:
continue
pair = (
symbols[left],
symbols[right]
)
if merge_rank.get(pair) != rank:
continue
# -------------------------------------------------
# Merge
# -------------------------------------------------
symbols[left] += symbols[right]
alive[right] = 0
next_index = following[right]
following[left] = next_index
if next_index != -1:
previous[next_index] = left
# -------------------------------------------------
# Previous pair
# -------------------------------------------------
prev_index = previous[left]
if prev_index != -1:
new_pair = (
symbols[prev_index],
symbols[left]
)
new_rank = merge_rank.get(new_pair)
if new_rank is not None:
heapq.heappush(
heap,
(
new_rank,
prev_index
)
)
# -------------------------------------------------
# Next pair
# -------------------------------------------------
if next_index != -1:
new_pair = (
symbols[left],
symbols[next_index]
)
new_rank = merge_rank.get(new_pair)
if new_rank is not None:
heapq.heappush(
heap,
(
new_rank,
left
)
)
# -------------------------------------------------
# Convert surviving symbols to IDs
# -------------------------------------------------
result = []
index = 0
while index != -1:
if alive[index]:
result.append(
token_to_id.get(
symbols[index],
unk_id
)
)
index = following[index]
return result
def decode(self, ids, skip_special_tokens=False, **kwargs):
if skip_special_tokens:
ids = [t for t in ids if t not in self.special_token_ids]
text = "".join(
self.id_to_token.get(
int(idx),
"<|unk|>"
)
for idx in ids
)
return text.replace("▁", " ") |