-
Notifications
You must be signed in to change notification settings - Fork 2
/
nametag3_server.py
executable file
·711 lines (577 loc) · 34.6 KB
/
nametag3_server.py
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
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
#!/usr/bin/env python3
# Copyright 2024 Institute of Formal and Applied Linguistics, Faculty of
# Mathematics and Physics, Charles University in Prague, Czech Republic.
#
# This Source Code Form is subject to the terms of the Mozilla Public
# License, v. 2.0. If a copy of the MPL was not distributed with this
# file, You can obtain one at https://mozilla.org/MPL/2.0/.
"""NameTag 3 server.
Starting NameTag 3 server
-------------------------
The mandatory arguments are given in this order:
- port
- default model name
- each following triple of arguments defines a model, of which
- first argument is the model name
- second argument is the model directory
- third argument are the acknowledgements to append
A single instance of a trained model physically stored on a disc can be listed
under several variants, just like in the following example, in which one model
(models/nametag3-multilingual-conll-240830/) is served as
a nametag3-multilingual-conll-240830 model and also as
a nametag3-english-CoNLL2003-conll-240830 model. The first model is also known as
multilingual-conll, and the second one is also named eng and en:
$ venv/bin/python3 nametag3_server.py 8001 models/nametag3-multilingual-conll-240830/ \
nametag3-multilingual-conll-240830/ models/nametag3-multilingual-conll-240830/ multilingual_acknowledgements \
nametag3-english-CoNLL2003-conll-240830:eng:en models/nametag3-multilingual-conll-240830/ english_acknowledgements \
Example server usage with three monolingual models:
$ venv/bin/python3 nametag3_server.py 8001 cs \
czech-cnec2.0-240830:cs:ces models/nametag3-czech-cnec2.0-240830/ czech-cnec2_acknowledgements \
english-CoNLL2003-conll-240830:en:eng models/nametag3-english-CoNLL2003-conll-240830/ english-CoNLL2003-conll_acknowledgements \
spanish-CoNLL2002-conll-240830:es:spa models/nametag3-spanish-CoNLL2002-conll-240830/ spanish-CoNLL2002-conll_acknowledgements
Sending requests to the NameTag 3 server
----------------------------------------
Example commandline call with curl:
$ curl -F data=@examples/cs_input.conll -F input="vertical" -F output="conll" localhost:8001/recognize| jq -j .result
Expected commandline output:
Jmenuji O
se O
Jan B-P|B-pf
Novák I-P|B-ps
. O
"""
import argparse
import collections
import email.parser
import http.server
import itertools
import io
import json
import os
import pickle
import socketserver
import sys
import time
import unicodedata
import urllib.parse
os.environ.setdefault("KERAS_BACKEND", "torch")
import transformers
from nametag3_dataset_collection import NameTag3DatasetCollection
from nametag3_model import nametag3_model_factory
import ufal.udpipe
SEP = "\t"
class UDPipeTokenizer:
class Token:
def __init__(self, token, spaces_before, spaces_after):
self.token = token
self.spaces_before = spaces_before
self.spaces_after = spaces_after
def __init__(self, path):
self._model = ufal.udpipe.Model.load(path)
if self._model is None:
raise RuntimeError("Cannot load tokenizer from {}".format(path))
def tokenize(self, text, mode="untokenized"):
if mode == "untokenized":
tokenizer = self._model.newTokenizer(self._model.DEFAULT)
elif mode == "vertical":
tokenizer = ufal.udpipe.InputFormat.newVerticalInputFormat()
elif mode.startswith("conllu"):
tokenizer = ufal.udpipe.InputFormat.newConlluInputFormat()
else:
raise ValueError("Unknown tokenizer mode '{}'".format(mode))
if tokenizer is None:
raise RuntimeError("Cannot create the tokenizer")
sentence = ufal.udpipe.Sentence()
processing_error = ufal.udpipe.ProcessingError()
tokenizer.setText(text)
while tokenizer.nextSentence(sentence, processing_error):
yield sentence
sentence = ufal.udpipe.Sentence()
if processing_error.occurred():
raise RuntimeError("Cannot read input data: '{}'".format(processing_error.message))
class Models:
"""Initializes NameTag 3 models, UDPipe tokenizers and HF tokenizers."""
class Model:
"""Initializes NameTag 3 model.
Initializes the NameTag3 model, along with the respective id2label and
label2id mappings, the UDPipe tokenizer and the HF tokenizer. The model
is not leaded to memory until before the first request.
"""
def __init__(self, path, name, acknowledgements, server_args):
self._server_args = server_args
self.name = name
self.acknowledgements = acknowledgements
# Read train args from model
with open("{}/options.json".format(path), mode="r") as options_file:
self._args = argparse.Namespace(**json.load(options_file))
if "max_labels_per_token" not in self._args:
self._args.max_labels_per_token = server_args.max_labels_per_token
self._args.batch_size = self._server_args.batch_size
print("self._args: ", self._args, file=sys.stderr, flush=True)
# Unpickle word mappings of train data
self._train_collection = NameTag3DatasetCollection(self._args)
with open("{}/mappings.pickle".format(path), mode="rb") as mappings_file:
self._train_collection.load_collection_mappings(path)
# Load the HF tokenizer
self.hf_tokenizer = transformers.AutoTokenizer.from_pretrained(self._args.hf_plm,
add_prefix_space = self._args.hf_plm in ["roberta-base", "roberta-large", "ufal/robeczech-base"])
# Construct the network
self.model = nametag3_model_factory(self._args.decoding)(len(self._train_collection.label2id().keys()),
self._args,
self._train_collection.id2label(),
self.hf_tokenizer)
# Load the checkpoint
self.model.load_checkpoint(os.path.join(path, self._args.checkpoint_filename))
# Load the UDPipe tokenizer
tokenizer_path = os.path.join(path, "udpipe.tokenizer")
self._udpipe_tokenizer = UDPipeTokenizer(tokenizer_path)
if self._udpipe_tokenizer is None:
raise RuntimeError("Cannot load tokenizer from {}".format(tokenizer_path))
def yield_predicted_batches(self, dataset):
time_start = time.time()
for batch_output in self.model.yield_predicted_batches("test", dataset, self.args):
yield batch_output
time_end = time.time()
print("Request {:.2f}ms,".format(1000 * (time_end - time_start)), file=sys.stderr, flush=True)
def postprocess(self, text):
return self.model.postprocess(text)
@property
def args(self):
return self._args
def conll_to_conllu(self, ner_output, sentences, encoding, n_nes_in_batches):
def _clean_misc(misc):
return "|".join(field for field in misc.split("|") if not field.startswith("NE="))
output = []
output_writer = ufal.udpipe.OutputFormat.newConlluOutputFormat()
n_sentences, n_words, n_multiwords, in_sentence = 0, 1, 0, False
open_ids = []
for line in (ner_output.split("\n")):
if not line:
if in_sentence:
output.append(output_writer.writeSentence(sentences[n_sentences]))
n_sentences += 1
n_words = 1
n_multiwords = 0
in_sentence = False
else:
in_sentence = True
# This will work for properly nested entities,
# hence model.postprocess is important before conll_to_conllu.
if encoding == "conllu-ne":
nes_encoded = []
words_in_token = 1
form, ne = line.split(SEP)
if ne == "O": # all entities ended
open_ids = []
else:
labels = ne.split("|")
for i in range(len(labels)):
if i < len(open_ids):
if labels[i].startswith("B-"):
# previous open entity ends here
# -> close it and all open nested entities
open_ids = open_ids[:i]
# open new entity
open_ids.append(n_nes_in_batches)
n_nes_in_batches += 1
else: # no running entities, new entity starts here, just append
open_ids.append(n_nes_in_batches)
n_nes_in_batches += 1
for i in range(len(labels)):
nes_encoded.append(labels[i][2:] + "_" + str(open_ids[i]))
# Multiword token starts here -> consume more words
if n_multiwords < len(sentences[n_sentences].multiwordTokens) and sentences[n_sentences].multiwordTokens[n_multiwords].idFirst == n_words:
words_in_token = sentences[n_sentences].multiwordTokens[n_multiwords].idLast - sentences[n_sentences].multiwordTokens[n_multiwords].idFirst + 1
sentences[n_sentences].multiwordTokens[n_multiwords].misc = _clean_misc(sentences[n_sentences].multiwordTokens[n_multiwords].misc)
if sentences[n_sentences].multiwordTokens[n_multiwords].misc and nes_encoded:
sentences[n_sentences].multiwordTokens[n_multiwords].misc += "|"
if nes_encoded:
sentences[n_sentences].multiwordTokens[n_multiwords].misc += "NE="
sentences[n_sentences].multiwordTokens[n_multiwords].misc = sentences[n_sentences].multiwordTokens[n_multiwords].misc + "-".join(nes_encoded)
n_multiwords += 1
# Write NEs to MISC
for i in range(words_in_token): # consume all words in multiword token
sentences[n_sentences].words[n_words].misc = _clean_misc(sentences[n_sentences].words[n_words].misc)
if sentences[n_sentences].words[n_words].misc and nes_encoded:
sentences[n_sentences].words[n_words].misc += "|"
if nes_encoded:
sentences[n_sentences].words[n_words].misc += "NE="
sentences[n_sentences].words[n_words].misc = sentences[n_sentences].words[n_words].misc + "-".join(nes_encoded)
n_words += 1
return "".join(output), n_nes_in_batches
def conll_to_vertical(self, text, n_tokens_in_batch):
output = []
open_ids, open_forms, open_labels = [], [], [] # open entities on i-th line
in_sentence = False
for i, line in enumerate(text.split("\n")):
if not line: # end of sentence
if in_sentence:
for j in range(len(open_ids)): # print all open entities
output.append((open_ids[j], open_labels[j], open_forms[j]))
open_ids, open_forms, open_labels = [], [], []
n_tokens_in_batch += 1
in_sentence = False
else:
in_sentence = True
form, ne = line.split(SEP)
n_tokens_in_batch += 1
if ne == "O": # all entities ended
for j in range(len(open_ids)): # print all open entities
output.append((open_ids[j], open_labels[j], open_forms[j]))
open_ids, open_forms, open_labels = [], [], []
else:
labels = ne.split("|")
for j in range(len(labels)): # for each label line
if j < len(open_ids): # all open entities
# previous open entity ends here, close and replace with new entity instead
if labels[j].startswith("B-") or open_labels[j] != labels[j][2:]:
output.append((open_ids[j], open_labels[j], open_forms[j]))
open_ids[j] = [n_tokens_in_batch]
open_forms[j] = form
# entity continues, append ids and forms
else:
open_ids[j].append(n_tokens_in_batch)
open_forms[j] += " " + form
open_labels[j] = labels[j][2:]
else: # no running entities, new entity starts here, just append
open_ids.append([n_tokens_in_batch])
open_forms.append(form)
open_labels.append(labels[j][2:])
output.sort(key=lambda ids_labels_forms: (ids_labels_forms[0][0], -ids_labels_forms[0][-1]))
output = "".join([",".join(map(str, ids)) + SEP + label + SEP + forms + "\n" for ids, label, forms in output])
return output, n_tokens_in_batch
@staticmethod
def encode_entities(text):
return text.replace('&', '&').replace('<', '<').replace('>', '>').replace('"', '"')
def conll_to_xml(self, text, udpipe_tokens):
"""Converts postprocessed (!) CoNLL output of the model.
This method expects correct bracketing and the IOB2 format of the
encoded named entities. Hence, postprocessing (model.postprocess)
of the model output is important before calling this method.
Rules for whitespaces around the <sentence>, <token> and <ne> XML
elements:
1. There are no whitespaces inside the <token> element. The <token>
element only holds the token.
2. Therefore it follows that all inter-token whitespaces are
printed in between the <token> elements.
3. There are no leading or trailing whitespaces inside the <ne>
element, threfore the whitespaces before the first <token> and
after the last <token> inside the <ne> element are printed
before/after the <ne> element.
4. There are no leading or trailing whitespaces inside the
<sentence> element, therefore the whitespaces before the first
<token> and after the last <token> are printed before/after the
<sentence> element.
"""
output = []
open_labels = []
in_sentence = False
s, t = -1, 0 # indexes to udpipe sentences and tokens inside sentences
delayed_spaces_after = ""
for line in text.split("\n"):
if not line: # end of sentence
for i in range(len(open_labels)): # close all open entities
output.append("</ne>")
open_labels = []
if in_sentence:
output.append("</sentence>") # close sentence
in_sentence = False
output.append(delayed_spaces_after)
delayed_spaces_after = ""
else: # in sentence
if not in_sentence: # sentence starts
s += 1
t = 0
output.append(udpipe_tokens[s][t].spaces_before)
output.append("<sentence>")
in_sentence = True
cols = line.split(SEP)
form = cols[0]
ne = cols[1] if len(cols) == 2 else "O"
# This will work for properly nested entities,
# hence model.postprocess is important before conll_to_xml.
opening_tags = []
if ne == "O": # all entities ended
for i in range(len(open_labels)): # close all open entities
output.append("</ne>")
open_labels = []
else:
labels = ne.split("|")
for i in range(len(labels)):
if i < len(open_labels):
if labels[i].startswith("B-") or open_labels[i] != labels[i][2:]:
# previous open entity ends here
# -> close it and all open nested entities
for _ in range(i, len(open_labels)):
output.append("</ne>")
open_labels = open_labels[:i]
# open new entity
opening_tags.append("<ne type=\"" + self.encode_entities(labels[i][2:]) + "\">")
open_labels.append(labels[i][2:])
else: # no running entities, new entity starts here, just append
opening_tags.append("<ne type=\"" + self.encode_entities(labels[i][2:]) + "\">")
open_labels.append(labels[i][2:])
output.append(delayed_spaces_after)
if t > 0:
output.append(self.encode_entities(udpipe_tokens[s][t].spaces_before))
output.append("".join(opening_tags))
output.append("<token>" + self.encode_entities(form) + "</token>")
delayed_spaces_after = udpipe_tokens[s][t].spaces_after
t += 1
return "".join(output)
def __init__(self, server_args):
self.default_model = server_args.default_model
self.models_list = [] # initialized models
self.models_by_names = {} # model names and language variants
self.models_by_paths = {} # paths to initialized models
self.important_names_list = [] # important names to list
for i in range(0, len(server_args.models), 3):
names, path, acknowledgements = server_args.models[i:i+3]
names = names.split(":")
names = [name.split("-") for name in names]
names = ["-".join(parts[:None if not i else -i]) for parts in names for i in range(len(parts))]
if path in self.models_by_paths:
print("Model in path \"{}\" already exists, sharing it also for model \"{}\"".format(path, names[0]), file=sys.stderr, flush=True)
model = self.models_by_paths[path]
else:
print("Initializing model \"{}\" from path \"{}\"".format(names[0], path), file=sys.stderr, flush=True)
model = self.Model(path, names[0], acknowledgements, server_args)
self.models_list.append(model)
self.models_by_paths[path] = model
self.important_names_list.append(names[0])
for name in names:
self.models_by_names.setdefault(name, model)
# Check the default model exists
assert self.default_model in self.models_by_names
class NameTag3Server(socketserver.ThreadingTCPServer):
class NameTag3ServerRequestHandler(http.server.BaseHTTPRequestHandler):
protocol_version = "HTTP/1.1"
def respond(request, content_type, code=200, additional_headers={}):
request.close_connection = True
request.send_response(code)
request.send_header("Connection", "close")
request.send_header("Content-Type", content_type)
request.send_header("Access-Control-Allow-Origin", "*")
for key, value in additional_headers.items():
request.send_header(key, value)
request.end_headers()
def respond_error(request, message, code=400):
request.respond("text/plain", code)
request.wfile.write(message.encode("utf-8"))
def start_responding(request, url, output_param, model, infclen):
if url.path.startswith("/weblicht"):
request.respond("application/conllu")
else:
request.respond("application/json", additional_headers={"X-Billing-Input-NFC-Len": str(infclen)})
request.wfile.write(json.dumps(collections.OrderedDict([
("model", model.name),
("acknowledgements", ["https://ufal.mff.cuni.cz/nametag/3#acknowledgements", model.acknowledgements]),
("result", ""),
]), indent=1)[:-3].encode("utf-8"))
if output_param == "conllu-ne":
request.wfile.write(json.dumps(
"# generator = NameTag 3, https://lindat.mff.cuni.cz/services/nametag\n"
"# nametag_model = {}\n"
"# nametag_model_licence = CC BY-NC-SA\n".format(model.name))[1:-1].encode("utf-8"))
def do_GET(request):
# Parse the URL
params = {}
try:
request.path = request.path.encode("iso-8859-1").decode("utf-8")
url = urllib.parse.urlparse(request.path)
for name, value in urllib.parse.parse_qsl(url.query, encoding="utf-8", keep_blank_values=True, errors="strict"):
params[name] = value
except:
return request.respond_error("Cannot parse request URL.")
# Parse the body of a POST request
if request.command == "POST":
if request.headers.get("Transfer-Encoding", "identity").lower() != "identity":
return request.respond_error("Only 'identity' Transfer-Encoding of payload is supported for now.")
try:
content_length = int(request.headers["Content-Length"])
except:
return request.respond_error("The Content-Length of payload is required.")
if content_length > request.server._server_args.max_request_size:
return request.respond_error("The payload size is too large.")
# Content-Type
if url.path.startswith("/weblicht"):
try:
params["data"] = request.rfile.read(content_length).decode("utf-8")
except:
return request.respond_error("Payload not in UTF-8.")
params["input"] = "conllu"
params["output"] = "conllu-ne"
elif request.headers.get("Content-Type", "").startswith("multipart/form-data"):
try:
parser = email.parser.BytesFeedParser()
parser.feed(b"Content-Type: " + request.headers["Content-Type"].encode("ascii") + b"\r\n\r\n")
while content_length:
parser.feed(request.rfile.read(min(content_length, 4096)))
content_length -= min(content_length, 4096)
for part in parser.close().get_payload():
name = part.get_param("name", header="Content-Disposition")
if name:
params[name] = part.get_payload(decode=True).decode("utf-8")
except:
return request.respond_error("Cannot parse the multipart/form-data payload.")
elif request.headers.get("Content-Type", "").startswith("application/x-www-form-urlencoded"):
try:
for name, value in urllib.parse.parse_qsl(
request.rfile.read(content_length).decode("utf-8"), encoding="utf-8", keep_blank_values=True, errors="strict"):
params[name] = value
except:
return request.respond_error("Cannot parse the application/x-www-form-urlencoded payload.")
else:
return request.respond_error("Unsupported payload Content-Type '{}'.".format(request.headers.get("Content-Type", "<none>")))
# Handle /models
if url.path == "/models":
response = {
"models": {name: ["tokenize", "recognize"] for name in request.server._models.important_names_list},
"default_model": request.server._models.default_model,
}
request.respond("application/json")
request.wfile.write(json.dumps(response, indent=1).encode("utf-8"))
# Handle /tokenize and /recognize
elif url.path in [ "/recognize", "/tokenize", "/weblicht/recognize" ]:
# Data
if "data" not in params:
return request.respond_error("The parameter 'data' is required.")
params["data"] = unicodedata.normalize("NFC", params["data"])
# Model
model = params.get("model", request.server._models.default_model)
if model not in request.server._models.models_by_names:
return request.respond_error("The requested model '{}' does not exist.".format(model))
model = request.server._models.models_by_names[model]
# Input
input_param = "untokenized" if url.path == "/tokenize" else params.get("input", "untokenized")
if input_param not in ["untokenized", "vertical", "conllu"]:
return request.respond_error("The requested input '{}' does not exist.".format(input_param))
# Output
output_param = params.get("output", "xml")
if output_param not in ["xml", "vertical"] + (["conll", "conllu-ne"] if url.path in ["/recognize", "/weblicht/recognize"] else []):
return request.respond_error("The requested output '{}' does not exist.".format(output_param))
# Sentences
try:
# Convert the generator to a list to raise exceptions early
sentences = list(model._udpipe_tokenizer.tokenize(params["data"], input_param))
except:
return request.respond_error("Cannot parse the input in the '{}' format.".format(input_param))
# Billing info
infclen = sum(sum(len(word.form) for word in sentence.words[1:]) for sentence in sentences)
# Skip multiwords, get tokens from sentences
input_tokens, token_list = [], [] # [input_tokens], [sentences x tokens]
for sentence in sentences:
token_list.append([])
word, multiword_token = 1, 0
while word < len(sentence.words):
if multiword_token < len(sentence.multiwordTokens) and sentence.multiwordTokens[multiword_token].idFirst == word:
token = sentence.multiwordTokens[multiword_token]
word = sentence.multiwordTokens[multiword_token].idLast + 1
multiword_token += 1
else:
token = sentence.words[word]
word += 1
input_tokens.append(token.form)
token_list[-1].append(model._udpipe_tokenizer.Token(token.form, token.getSpacesBefore(), token.getSpacesAfter()))
input_tokens.append("")
# Create NameTag3Collection with only one NameTag3Dataset.
test_collection = NameTag3DatasetCollection(model.args,
tokenizer=model.hf_tokenizer,
text="\n".join(input_tokens),
train_collection=model._train_collection)
# Predict and convert to the requested output format.
started_responding = False
n_tokens_in_batches, n_nes_in_batches, n_sentences_in_batches = 0, 1, 0
try:
# Handle empty requests separately by generating empty output with valid format and headers.
if len(input_tokens) == 0:
request.start_responding(url, output_param, model, infclen)
# Handle non-empty requests by running the neural network.
else:
for batch_output in model.yield_predicted_batches(test_collection.datasets[-1]):
# Sentences and tokens processed in this batch
batch_sentences = sentences[n_sentences_in_batches:n_sentences_in_batches+len(batch_output)]
batch_udpipe_tokens = token_list[n_sentences_in_batches:n_sentences_in_batches+len(batch_output)]
n_sentences_in_batches += len(batch_output)
# Finalize the batch output string by joining the sentence strings.
batch_output = "".join(batch_output)
if url.path == "/recognize" or url.path == "/weblicht/recognize":
batch_output = model.postprocess(batch_output)
if output_param == "vertical":
batch_output, n_tokens_in_batches = model.conll_to_vertical(batch_output, n_tokens_in_batches)
if output_param == "conllu-ne":
batch_output, n_nes_in_batches = model.conll_to_conllu(batch_output, batch_sentences, "conllu-ne", n_nes_in_batches)
if output_param == "xml":
batch_output = model.conll_to_xml(batch_output, batch_udpipe_tokens)
if not started_responding:
# The first batch is ready, we commit to generate batch_output.
request.start_responding(url, output_param, model, infclen)
started_responding=True
if url.path.startswith("/weblicht"):
request.wfile.write(batch_output.encode("utf-8"))
else:
request.wfile.write(json.dumps(batch_output, ensure_ascii=False)[1:-1].encode("utf-8"))
if not url.path.startswith("/weblicht"):
request.wfile.write(b'"\n}\n')
except:
import traceback
traceback.print_exc(file=sys.stderr)
sys.stderr.flush()
if not started_responding:
request.respond_error("An internal error occurred during processing.")
else:
if url.path.startswith("/weblicht"):
request.wfile.write(b'\n\nAn internal error occurred during processing, producing incorrect CoNLL-U!')
else:
request.wfile.write(b'",\n"An internal error occurred during processing, producing incorrect JSON!"')
else:
request.respond_error("No handler for the given URL '{}'".format(url.path), code=404)
def do_POST(request):
return request.do_GET()
daemon_threads = False
def __init__(self, server_args, models):
super().__init__(("", server_args.port), self.NameTag3ServerRequestHandler)
self._server_args = server_args
self._models = models
def server_bind(self):
import socket
self.socket.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
self.socket.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEPORT, 1)
super().server_bind()
def service_actions(self):
if isinstance(getattr(self, "_threads", None), list):
if len(self._threads) >= 1024:
self._threads = [thread for thread in self._threads if thread.is_alive()]
if __name__ == "__main__":
import signal
import threading
# Parse server arguments
parser = argparse.ArgumentParser()
parser.add_argument("port", type=int, help="Port to use")
parser.add_argument("default_model", type=str, help="Default model")
parser.add_argument("models", type=str, nargs="+", help="Models to serve")
parser.add_argument("--batch_size", default=32, type=int, help="Batch size")
parser.add_argument("--logfile", default=None, type=str, help="Log path")
parser.add_argument("--max_labels_per_token", default=5, type=int, help="Maximum labels per token.")
parser.add_argument("--max_request_size", default=4096*1024, type=int, help="Maximum request size")
parser.add_argument("--threads", default=4, type=int, help="Threads to use")
args = parser.parse_args()
# Log stderr to logfile if given
if args.logfile is not None:
sys.stderr = open(args.logfile, "a", encoding="utf-8")
# Load the models
models = Models(args)
# Create the server
server = NameTag3Server(args, models)
server_thread = threading.Thread(target=server.serve_forever, daemon=True)
server_thread.start()
print("Started NameTag 3 server on port {}.".format(args.port), file=sys.stderr)
print("To stop it gracefully, either send SIGINT (Ctrl+C) or SIGUSR1.", file=sys.stderr, flush=True)
# Wait until the server should be closed
signal.pthread_sigmask(signal.SIG_BLOCK, [signal.SIGINT, signal.SIGUSR1])
signal.sigwait([signal.SIGINT, signal.SIGUSR1])
print("Initiating shutdown of the NameTag 3 server.", file=sys.stderr, flush=True)
server.shutdown()
print("Stopped handling new requests, processing all current ones.", file=sys.stderr, flush=True)
server.server_close()
print("Finished shutdown of the NameTag 3 server.", file=sys.stderr, flush=True)