Source file processors.ml
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
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
(** Post-processing module for tokenization output. *)
type encoding = {
ids : int array;
type_ids : int array;
tokens : string array;
offsets : (int * int) array;
special_tokens_mask : int array;
attention_mask : int array;
overflowing : encoding list;
sequence_ranges : (int * int * int) list;
}
(** Type representing an encoding to be processed *)
(** Main post-processor type *)
type t =
| Bert of { sep : string * int; cls : string * int }
| Roberta of {
sep : string * int;
cls : string * int;
pad : string * int;
trim_offsets : bool;
add_prefix_space : bool;
}
| ByteLevel of { trim_offsets : bool }
| Template of {
single : string;
pair : string option;
special_tokens : (string * int) list;
}
| Sequence of t list
(** Helper functions *)
let create_empty_encoding () =
{
ids = [||];
type_ids = [||];
tokens = [||];
offsets = [||];
special_tokens_mask = [||];
attention_mask = [||];
overflowing = [];
sequence_ranges = [];
}
let add_token encoding token_str token_id type_id is_special =
{
encoding with
ids = Array.append encoding.ids [| token_id |];
tokens = Array.append encoding.tokens [| token_str |];
type_ids = Array.append encoding.type_ids [| type_id |];
special_tokens_mask =
Array.append encoding.special_tokens_mask
[| (if is_special then 1 else 0) |];
attention_mask = Array.append encoding.attention_mask [| 1 |];
offsets = Array.append encoding.offsets [| (0, 0) |];
}
(** Process BERT encoding *)
let process_bert ~sep ~cls encodings ~add_special_tokens =
if not add_special_tokens then encodings
else
let cls_str, cls_id = cls in
let sep_str, sep_id = sep in
match encodings with
| [] -> []
| [ encoding ] ->
let result = create_empty_encoding () in
let result = add_token result cls_str cls_id 0 true in
let result =
{
result with
ids = Array.append result.ids encoding.ids;
tokens = Array.append result.tokens encoding.tokens;
type_ids =
Array.append result.type_ids
(Array.make (Array.length encoding.ids) 0);
special_tokens_mask =
Array.append result.special_tokens_mask
(Array.make (Array.length encoding.ids) 0);
attention_mask =
Array.append result.attention_mask
(Array.make (Array.length encoding.ids) 1);
offsets = Array.append result.offsets encoding.offsets;
}
in
let result = add_token result sep_str sep_id 0 true in
[ result ]
| [ enc1; enc2 ] ->
let result = create_empty_encoding () in
let result = add_token result cls_str cls_id 0 true in
let result =
{
result with
ids = Array.append result.ids enc1.ids;
tokens = Array.append result.tokens enc1.tokens;
type_ids =
Array.append result.type_ids
(Array.make (Array.length enc1.ids) 0);
special_tokens_mask =
Array.append result.special_tokens_mask
(Array.make (Array.length enc1.ids) 0);
attention_mask =
Array.append result.attention_mask
(Array.make (Array.length enc1.ids) 1);
offsets = Array.append result.offsets enc1.offsets;
}
in
let result = add_token result sep_str sep_id 0 true in
let result =
{
result with
ids = Array.append result.ids enc2.ids;
tokens = Array.append result.tokens enc2.tokens;
type_ids =
Array.append result.type_ids
(Array.make (Array.length enc2.ids) 1);
special_tokens_mask =
Array.append result.special_tokens_mask
(Array.make (Array.length enc2.ids) 0);
attention_mask =
Array.append result.attention_mask
(Array.make (Array.length enc2.ids) 1);
offsets = Array.append result.offsets enc2.offsets;
}
in
let result = add_token result sep_str sep_id 1 true in
[ result ]
| _ -> encodings
(** Process RoBERTa encoding *)
let process_roberta ~sep ~cls ~pad:_ ~trim_offsets:_ ~add_prefix_space:_
encodings ~add_special_tokens =
if not add_special_tokens then encodings
else
let cls_str, cls_id = cls in
let sep_str, sep_id = sep in
match encodings with
| [] -> []
| [ encoding ] ->
let result = create_empty_encoding () in
let result = add_token result cls_str cls_id 0 true in
let result =
{
result with
ids = Array.append result.ids encoding.ids;
tokens = Array.append result.tokens encoding.tokens;
type_ids =
Array.append result.type_ids
(Array.make (Array.length encoding.ids) 0);
special_tokens_mask =
Array.append result.special_tokens_mask
(Array.make (Array.length encoding.ids) 0);
attention_mask =
Array.append result.attention_mask
(Array.make (Array.length encoding.ids) 1);
offsets = Array.append result.offsets encoding.offsets;
}
in
let result = add_token result sep_str sep_id 0 true in
[ result ]
| [ enc1; enc2 ] ->
let result = create_empty_encoding () in
let result = add_token result cls_str cls_id 0 true in
let result =
{
result with
ids = Array.append result.ids enc1.ids;
tokens = Array.append result.tokens enc1.tokens;
type_ids =
Array.append result.type_ids
(Array.make (Array.length enc1.ids) 0);
special_tokens_mask =
Array.append result.special_tokens_mask
(Array.make (Array.length enc1.ids) 0);
attention_mask =
Array.append result.attention_mask
(Array.make (Array.length enc1.ids) 1);
offsets = Array.append result.offsets enc1.offsets;
}
in
let result = add_token result sep_str sep_id 0 true in
let result = add_token result sep_str sep_id 0 true in
let result =
{
result with
ids = Array.append result.ids enc2.ids;
tokens = Array.append result.tokens enc2.tokens;
type_ids =
Array.append result.type_ids
(Array.make (Array.length enc2.ids) 0);
special_tokens_mask =
Array.append result.special_tokens_mask
(Array.make (Array.length enc2.ids) 0);
attention_mask =
Array.append result.attention_mask
(Array.make (Array.length enc2.ids) 1);
offsets = Array.append result.offsets enc2.offsets;
}
in
let result = add_token result sep_str sep_id 0 true in
[ result ]
| _ -> encodings
(** Process byte-level encoding *)
let process_byte_level ~trim_offsets encodings ~add_special_tokens:_ =
if not trim_offsets then encodings
else
List.map
(fun encoding ->
{
encoding with
offsets =
Array.map
(fun (start, stop) ->
(start, stop))
encoding.offsets;
})
encodings
(** Process template encoding *)
let process_template ~single ~pair:_ ~special_tokens:_ encodings
~add_special_tokens =
if not add_special_tokens then encodings
else
let _ = single in
encodings
(** Main process function *)
let rec process processor encodings ~add_special_tokens =
match processor with
| Bert { sep; cls } -> process_bert ~sep ~cls encodings ~add_special_tokens
| Roberta { sep; cls; pad; trim_offsets; add_prefix_space } ->
process_roberta ~sep ~cls ~pad ~trim_offsets ~add_prefix_space encodings
~add_special_tokens
| ByteLevel { trim_offsets } ->
process_byte_level ~trim_offsets encodings ~add_special_tokens
| Template { single; pair; special_tokens } ->
process_template ~single ~pair ~special_tokens encodings
~add_special_tokens
| Sequence processors ->
List.fold_left
(fun encs proc -> process proc encs ~add_special_tokens)
encodings processors
(** Get number of added tokens *)
let rec added_tokens processor ~is_pair =
match processor with
| Bert _ ->
if is_pair then 3
else 2
| Roberta _ ->
if is_pair then 4 else 2
| ByteLevel _ -> 0
| Template _ -> 0
| Sequence processors ->
List.fold_left
(fun acc proc -> acc + added_tokens proc ~is_pair)
0 processors
(** Constructors *)
let bert ~sep ~cls () = Bert { sep; cls }
let roberta ~sep ~cls ?(trim_offsets = true) ?(add_prefix_space = true) () =
let pad = ("<pad>", 1) in
Roberta { sep; cls; pad; trim_offsets; add_prefix_space }
let byte_level ?(trim_offsets = true) () = ByteLevel { trim_offsets }
let template ~single ?pair ?(special_tokens = []) () =
Template { single; pair; special_tokens }
let sequence processors = Sequence processors
(** Serialization *)
let rec to_json = function
| Bert { sep = sep_str, sep_id; cls = cls_str, cls_id } ->
`Assoc
[
("type", `String "BertProcessing");
("sep", `List [ `String sep_str; `Int sep_id ]);
("cls", `List [ `String cls_str; `Int cls_id ]);
]
| Roberta
{
sep = sep_str, sep_id;
cls = cls_str, cls_id;
pad = pad_str, pad_id;
trim_offsets;
add_prefix_space;
} ->
`Assoc
[
("type", `String "RobertaProcessing");
("sep", `List [ `String sep_str; `Int sep_id ]);
("cls", `List [ `String cls_str; `Int cls_id ]);
("pad", `List [ `String pad_str; `Int pad_id ]);
("trim_offsets", `Bool trim_offsets);
("add_prefix_space", `Bool add_prefix_space);
]
| ByteLevel { trim_offsets } ->
`Assoc
[ ("type", `String "ByteLevel"); ("trim_offsets", `Bool trim_offsets) ]
| Template { single; pair; special_tokens } ->
`Assoc
[
("type", `String "TemplateProcessing");
("single", `String single);
("pair", match pair with None -> `Null | Some p -> `String p);
( "special_tokens",
`List
(List.map
(fun (s, i) -> `List [ `String s; `Int i ])
special_tokens) );
]
| Sequence processors ->
`Assoc
[
("type", `String "Sequence");
("processors", `List (List.map to_json processors));
]
let rec of_json = function
| `Assoc fields -> (
match List.assoc_opt "type" fields with
| Some (`String "BertProcessing") ->
let sep =
match List.assoc_opt "sep" fields with
| Some (`List [ `String s; `Int i ]) -> (s, i)
| _ -> ("[SEP]", 102)
in
let cls =
match List.assoc_opt "cls" fields with
| Some (`List [ `String s; `Int i ]) -> (s, i)
| _ -> ("[CLS]", 101)
in
Bert { sep; cls }
| Some (`String "RobertaProcessing") ->
let sep =
match List.assoc_opt "sep" fields with
| Some (`List [ `String s; `Int i ]) -> (s, i)
| _ -> ("</s>", 2)
in
let cls =
match List.assoc_opt "cls" fields with
| Some (`List [ `String s; `Int i ]) -> (s, i)
| _ -> ("<s>", 0)
in
let pad =
match List.assoc_opt "pad" fields with
| Some (`List [ `String s; `Int i ]) -> (s, i)
| _ -> ("<pad>", 1)
in
let trim_offsets =
match List.assoc_opt "trim_offsets" fields with
| Some (`Bool b) -> b
| _ -> true
in
let add_prefix_space =
match List.assoc_opt "add_prefix_space" fields with
| Some (`Bool b) -> b
| _ -> true
in
Roberta { sep; cls; pad; trim_offsets; add_prefix_space }
| Some (`String "ByteLevel") ->
let trim_offsets =
match List.assoc_opt "trim_offsets" fields with
| Some (`Bool b) -> b
| _ -> true
in
ByteLevel { trim_offsets }
| Some (`String "TemplateProcessing") ->
let single =
match List.assoc_opt "single" fields with
| Some (`String s) -> s
| _ -> failwith "Missing single template"
in
let pair =
match List.assoc_opt "pair" fields with
| Some (`String p) -> Some p
| _ -> None
in
let special_tokens =
match List.assoc_opt "special_tokens" fields with
| Some (`List tokens) ->
List.map
(function
| `List [ `String s; `Int i ] -> (s, i)
| _ -> failwith "Invalid special token format")
tokens
| _ -> []
in
Template { single; pair; special_tokens }
| Some (`String "Sequence") -> (
match List.assoc_opt "processors" fields with
| Some (`List procs) -> Sequence (List.map of_json procs)
| _ -> failwith "Invalid Sequence processor")
| _ -> failwith "Unknown processor type")
| _ -> failwith "Invalid processor JSON"