PyTorch
gpt2
File size: 10,865 Bytes
064b576
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
# syllabic_pretokenizer.py

import re
from typing import List, Tuple, Optional

###############################################################################
# Configurable text preprocessor (spacing, lowercase) with alignment tracking
###############################################################################

class Preprocessor:
    def __init__(
        self,
        lowercase: bool = False,
        separate_apostrophes: bool = True,
        separate_digits: bool = True,
        separate_punctuation: bool = True,
    ):
        self.lowercase = lowercase
        self.separate_apostrophes = separate_apostrophes
        self.separate_digits = separate_digits
        self.separate_punctuation = separate_punctuation

        # Precompiled regexes
        self._apos_re = re.compile(r"[’'`]")
        self._punct_re = re.compile(r"[^A-Za-z0-9\s’'`]")
        self._digit_re = re.compile(r"\d")

    def preprocess_with_alignment(self, line: str) -> Tuple[str, List[Optional[int]]]:
        """
        Apply normalization equivalent to your previous Preprocessor, but also
        return a map from each character in the preprocessed string back to the
        raw string index. Inserted spaces get None in the map.
        """
        raw = line
        if self.lowercase:
            raw = raw.lower()

        out_chars: List[str] = []
        out2raw: List[Optional[int]] = []

        def emit(ch: str, raw_idx: Optional[int]):
            out_chars.append(ch)
            out2raw.append(raw_idx)

        i = 0
        n = len(raw)
        while i < n:
            c = raw[i]
            # Decide if we should isolate this char with spaces
            isolate = (
                (self.separate_apostrophes and self._apos_re.match(c) is not None)
                or (self.separate_punctuation and self._punct_re.match(c) is not None)
                or (self.separate_digits and self._digit_re.match(c) is not None)
            )

            if isolate:
                emit(" ", None)
                emit(c, i)
                emit(" ", None)
            else:
                emit(c, i)
            i += 1

        # Collapse whitespace to single spaces and strip, keeping alignment
        pre, pre_map = _collapse_whitespace_with_map(out_chars, out2raw)
        return pre, pre_map


def _collapse_whitespace_with_map(
    chars: List[str], idx_map: List[Optional[int]]
) -> Tuple[str, List[Optional[int]]]:
    """
    Collapse runs of whitespace to a single space and trim leading and trailing
    whitespace, while preserving a per-char map back to the original.
    """
    assert len(chars) == len(idx_map)
    result_chars: List[str] = []
    result_map: List[Optional[int]] = []

    def is_space(ch: str) -> bool:
        return ch.isspace()

    # First pass: collapse runs to single spaces
    prev_space = False
    for ch, m in zip(chars, idx_map):
        if is_space(ch):
            if not prev_space:
                result_chars.append(" ")
                result_map.append(None)  # inserted or collapsed space has no single origin
                prev_space = True
            # else skip extra spaces
        else:
            result_chars.append(ch)
            result_map.append(m)
            prev_space = False

    # Strip leading space
    if result_chars and result_chars[0] == " ":
        result_chars.pop(0)
        result_map.pop(0)

    # Strip trailing space
    if result_chars and result_chars[-1] == " ":
        result_chars.pop()
        result_map.pop()

    return "".join(result_chars), result_map


###############################################################################
# Syllabifier (faithful to your logic)
###############################################################################

class Syllabifier:
    def __init__(self):
        # Case insensitive vowel matcher
        self._vowel_re = re.compile(r"[aeiou]+", re.I)
        self._consonant_clusters = {
            "bl", "br", "cl", "cr", "dr", "fl", "fr", "gl", "gr", "pl", "pr",
            "sc", "sk", "sl", "sm", "sn", "sp", "st", "sw", "tr", "tw", "th",
            "ch", "sh", "ph", "wh", "sch", "str", "spr", "spl", "scr", "thr"
        }

    def syllabify_word(self, word: str) -> List[str]:
        if len(word) <= 2:
            return [word]

        syllables: List[str] = []
        i = 0
        n = len(word)

        while i < n:
            syllable = ""

            # Collect initial consonant cluster
            consonant_start = i
            while i < n and not self._vowel_re.match(word[i]):
                i += 1

            if i > consonant_start:
                cluster = word[consonant_start:i]
                found = False
                # Prefer longer known clusters first
                for known in sorted(self._consonant_clusters, key=len, reverse=True):
                    if cluster.startswith(known):
                        syllable += known
                        cluster_rest = cluster[len(known):]
                        if cluster_rest and syllables:
                            # attach leftover to previous syllable
                            syllables[-1] += cluster_rest
                        found = True
                        break
                if not found:
                    # Split unknown cluster roughly in half
                    split = len(cluster) // 2
                    if syllables:
                        syllables[-1] += cluster[:split]
                    syllable += cluster[split:]

            # Add vowel group
            vowel_start = i
            while i < n and self._vowel_re.match(word[i]):
                i += 1
            syllable += word[vowel_start:i]

            # Trailing consonants
            trailing_start = i
            while i < n and not self._vowel_re.match(word[i]):
                i += 1
            if i > trailing_start:
                consonants = word[trailing_start:i]
                if i < n:
                    # One consonant stays with current syllable, rest go to next
                    syllable += consonants[0]
                    i = trailing_start + 1
                else:
                    # End of word, keep all
                    syllable += consonants

            if syllable:
                syllables.append(syllable)

        # Merge very short syllables
        merged: List[str] = []
        k = 0
        while k < len(syllables):
            cur = syllables[k]
            if len(cur) == 1 and k < len(syllables) - 1:
                merged.append(cur + syllables[k + 1])
                k += 2
            else:
                merged.append(cur)
                k += 1

        return merged if merged else [word]


###############################################################################
# End to end helpers: preprocessing + syllabification + alignment
###############################################################################

def preprocess_and_segment_with_alignment(
    text: str,
    preprocessor: Preprocessor,
    syllabifier: Optional[Syllabifier] = None,
) -> Tuple[str, List[Optional[int]]]:
    """
    1) Apply Preprocessor with alignment tracking
    2) Split by spaces into tokens
    3) Syllabify each token
    4) Rejoin syllables with single spaces
    5) Return segmented text plus a map to raw indices
    """
    pre, pre2raw = preprocessor.preprocess_with_alignment(text)
    if syllabifier is None:
        syllabifier = Syllabifier()

    out_chars: List[str] = []
    out_map: List[Optional[int]] = []

    i = 0
    n = len(pre)
    while i < n:
        # Skip spaces
        while i < n and pre[i].isspace():
            i += 1
        if i >= n:
            break

        # Read one token
        j = i
        while j < n and not pre[j].isspace():
            j += 1
        token = pre[i:j]

        # Map for this token
        token_chars = pre[i:j]
        token_maps = pre2raw[i:j]

        # Syllabify this token
        sylls = syllabifier.syllabify_word(token)

        # --- Guard: ensure syllables cover the whole token exactly ---
        total = sum(len(s) for s in sylls)
        if total != len(token):
            # Rebuild syllables by slicing the original token according to the
            # predicted lengths, and force the last syllable to take any remainder.
            rebuilt = []
            pos2 = 0
            for s in sylls[:-1]:
                L = min(len(s), len(token) - pos2)
                if L <= 0:
                    break
                rebuilt.append(token[pos2:pos2 + L])
                pos2 += L
            # Last syllable gets the rest (can be empty if already exact)
            if pos2 < len(token):
                rebuilt.append(token[pos2:])
            sylls = [s for s in rebuilt if s]
        # --------------------------------------------------------------


        # Emit token chars, inserting a single space between syllables
        pos = 0
        for s_idx, syl in enumerate(sylls):
            L = len(syl)
            # Emit the next L chars from token
            for k in range(L):
                out_chars.append(token_chars[pos + k])
                out_map.append(token_maps[pos + k])
            pos += L
            # Add space between syllables, not after last
            if s_idx < len(sylls) - 1:
                out_chars.append(" ")
                out_map.append(None)

        # Add a space between tokens if there is more input
        i = j
        # Peek ahead to see if there is another token
        while i < n and pre[i].isspace():
            i += 1
        if i < n:
            out_chars.append(" ")
            out_map.append(None)

    # Final collapse (defensive) and strip
    segmented, seg_map = _collapse_whitespace_with_map(out_chars, out_map)
    return segmented, seg_map


def remap_offsets_to_raw(
    offsets: List[Tuple[int, int]],
    pre2raw: List[Optional[int]],
) -> List[Tuple[int, int]]:
    """
    Translate offsets from the preprocessed or segmented string back to raw
    string offsets. If an offset region consists only of inserted spaces,
    return a degenerate (0, 0) span.
    """
    mapped: List[Tuple[int, int]] = []
    L = len(pre2raw)
    for s, e in offsets:
        s = max(0, min(s, L))
        e = max(0, min(e, L))

        rs = None
        re_ = None

        # start: first non-None in [s, e)
        t = s
        while t < e and rs is None:
            if pre2raw[t] is not None:
                rs = pre2raw[t]
            t += 1

        # end: last non-None in [s, e)
        t = e - 1
        while t >= s and re_ is None:
            if pre2raw[t] is not None:
                re_ = pre2raw[t] + 1  # exclusive end
            t -= 1

        if rs is None or re_ is None:
            mapped.append((0, 0))
        else:
            mapped.append((rs, re_))
    return mapped