/
torchlib.cc
1269 lines (1149 loc) · 44.4 KB
/
torchlib.cc
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
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/**
* DeepDetect
* Copyright (c) 2019-2020 Jolibrain
* Author: Louis Jean <ljean@etud.insa-toulouse.fr>
* Guillaume Infantes <guillaume.infantes@jolibrain.com>
*
* This file is part of deepdetect.
*
* deepdetect is free software: you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* deepdetect is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public License
* along with deepdetect. If not, see <http://www.gnu.org/licenses/>.
*/
#include "torchlib.h"
#if !defined(CPU_ONLY)
#include <c10/cuda/CUDACachingAllocator.h>
#endif
#include "outputconnectorstrategy.h"
#include "generators/net_caffe.h"
#include "generators/net_caffe_recurrent.h"
#include <sys/stat.h>
#include <sys/types.h>
#include <fcntl.h>
#include "native/native.h"
#include "torchsolver.h"
#include "torchutils.h"
using namespace torch;
namespace dd
{
template <class TInputConnectorStrategy, class TOutputConnectorStrategy,
class TMLModel>
TorchLib<TInputConnectorStrategy, TOutputConnectorStrategy,
TMLModel>::TorchLib(const TorchModel &tmodel)
: MLLib<TInputConnectorStrategy, TOutputConnectorStrategy, TorchModel>(
tmodel)
{
this->_libname = "torch";
}
template <class TInputConnectorStrategy, class TOutputConnectorStrategy,
class TMLModel>
TorchLib<TInputConnectorStrategy, TOutputConnectorStrategy,
TMLModel>::TorchLib(TorchLib &&tl) noexcept
: MLLib<TInputConnectorStrategy, TOutputConnectorStrategy, TorchModel>(
std::move(tl))
{
this->_libname = "torch";
_module = std::move(tl._module);
_template = tl._template;
_nclasses = tl._nclasses;
_main_device = tl._main_device;
_devices = tl._devices;
_masked_lm = tl._masked_lm;
_seq_training = tl._seq_training;
_finetuning = tl._finetuning;
_best_metrics = tl._best_metrics;
_best_metric_value = tl._best_metric_value;
_classification = tl._classification;
_regression = tl._regression;
_timeserie = tl._timeserie;
_loss = tl._loss;
_template_params = tl._template_params;
}
template <class TInputConnectorStrategy, class TOutputConnectorStrategy,
class TMLModel>
TorchLib<TInputConnectorStrategy, TOutputConnectorStrategy,
TMLModel>::~TorchLib()
{
_module.free();
torch_utils::empty_cuda_cache();
}
template <class TInputConnectorStrategy, class TOutputConnectorStrategy,
class TMLModel>
double TorchLib<TInputConnectorStrategy, TOutputConnectorStrategy,
TMLModel>::unscale(double val, unsigned int k,
const TInputConnectorStrategy &inputc)
{
(void)inputc;
(void)k;
// unscaling is input connector specific
return val;
}
// full template specialization
template <>
double
TorchLib<CSVTSTorchInputFileConn, SupervisedOutput, TorchModel>::unscale(
double val, unsigned int k, const CSVTSTorchInputFileConn &inputc)
{
if (inputc._min_vals.empty() || inputc._max_vals.empty())
{
this->_logger->info("not unscaling output because no bounds "
"data found");
return val;
}
else
{
if (!inputc._dont_scale_labels)
{
double max = inputc._max_vals[inputc._label_pos[k]];
double min = inputc._min_vals[inputc._label_pos[k]];
if (inputc._scale_between_minus1_and_1)
val += 0.5;
val = val * (max - min) + min;
}
return val;
}
}
/*- from mllib -*/
template <class TInputConnectorStrategy, class TOutputConnectorStrategy,
class TMLModel>
void TorchLib<TInputConnectorStrategy, TOutputConnectorStrategy,
TMLModel>::init_mllib(const APIData &lib_ad)
{
bool gpu = false;
std::vector<int> gpuids;
bool freeze_traced = false;
int embedding_size = 768;
std::string self_supervised = "";
if (lib_ad.has("template"))
_template = lib_ad.get("template").get<std::string>();
if (lib_ad.has("gpu"))
gpu = lib_ad.get("gpu").get<bool>();
if (gpu && !torch::cuda::is_available())
{
throw MLLibBadParamException(
"GPU is not available, service could not be created");
}
if (lib_ad.has("gpuid"))
{
if (lib_ad.get("gpuid").is<int>())
gpuids = { lib_ad.get("gpuid").get<int>() };
else
gpuids = lib_ad.get("gpuid").get<std::vector<int>>();
}
if (lib_ad.has("nclasses"))
{
_classification = true;
_nclasses = lib_ad.get("nclasses").get<int>();
}
else if (lib_ad.has("ntargets"))
{
_regression = true;
_nclasses = lib_ad.get("ntargets").get<int>();
}
if (lib_ad.has("self_supervised"))
self_supervised = lib_ad.get("self_supervised").get<std::string>();
if (lib_ad.has("embedding_size"))
embedding_size = lib_ad.get("embedding_size").get<int>();
if (lib_ad.has("finetuning"))
_finetuning = lib_ad.get("finetuning").get<bool>();
if (lib_ad.has("freeze_traced"))
freeze_traced = lib_ad.get("freeze_traced").get<bool>();
if (lib_ad.has("loss"))
_loss = lib_ad.get("loss").get<std::string>();
if (!gpu)
{
_main_device = torch::Device(DeviceType::CPU);
_devices = { _main_device };
}
else
{
// if gpuids = -1, we use all gpus
if (gpuids.size() == 1 && gpuids[0] == -1)
{
gpuids.resize(torch::cuda::device_count());
std::iota(gpuids.begin(), gpuids.end(), 0);
}
if (gpuids.empty())
gpuids.push_back(0);
for (int gpuid : gpuids)
_devices.push_back(torch::Device(DeviceType::CUDA, gpuid));
_main_device = _devices[0];
if (_devices.size() > 1)
{
std::string devices_str = std::to_string(gpuids[0]);
for (size_t i = 1; i < _devices.size(); ++i)
devices_str += "," + std::to_string(gpuids[i]);
this->_logger->info("Running on multiple devices: " + devices_str);
}
}
_module._device = _main_device;
_module._logger = this->_logger;
if (_template.find("recurrent") != std::string::npos)
{
// call caffe net generator before verything else
std::string dest_net
= this->_mlmodel._repo + '/' + _template + ".prototxt";
if (!fileops::file_exists(dest_net))
{
caffe::NetParameter net_param;
configure_recurrent_template(lib_ad, this->_inputc, net_param,
this->_logger, true);
torch_utils::torch_write_proto_to_text_file(net_param, dest_net);
}
else
{
this->_logger->info("prototxt net definition already present in "
"repo : using old file");
}
this->_mlmodel._proto = dest_net;
}
bool unsupported_model_configuration
= this->_mlmodel._traced.empty() && this->_mlmodel._proto.empty()
&& !NativeFactory::valid_template_def(_template);
if (unsupported_model_configuration)
throw MLLibInternalException("Use of libtorch backend needs either: "
"traced net, protofile or native template");
// Create the model
this->_inputc._input_format = "bert";
if (_template == "bert")
{
if (_classification)
{
_module._linear = nn::Linear(embedding_size, _nclasses);
_module._linear->to(_main_device);
_module._hidden_states = true;
_module._linear_in = 1;
}
else if (!self_supervised.empty())
{
if (self_supervised != "mask")
{
throw MLLibBadParamException("self_supervised");
}
this->_logger->info("Masked Language model");
_masked_lm = true;
_seq_training = true;
}
else
{
throw MLLibBadParamException(
"BERT only supports self-supervised or classification");
}
}
else if (_template == "gpt2")
{
this->_inputc._input_format = "gpt2";
_seq_training = true;
}
else if (_template.find("recurrent") != std::string::npos
|| NativeFactory::is_timeserie(_template))
{
_timeserie = true;
}
else if (!_template.empty())
{
if (NativeFactory::valid_template_def(_template))
_module.create_native_template<TInputConnectorStrategy>(
_template, lib_ad, this->_inputc, this->_mlmodel, _main_device);
else
throw MLLibBadParamException("invalid torch model template "
+ _template);
}
if (_classification)
{
this->_mltype = "classification";
_module._nclasses = _nclasses;
if (_finetuning)
{
_module._require_linear_layer = true;
this->_logger->info(
"Add classification layer on top of the traced model");
}
}
else if (_regression)
{
this->_mltype = "regression";
_module._nclasses = _nclasses;
if (_finetuning)
{
_module._require_linear_layer = true;
this->_logger->info(
"Add regression layer on top of the traced model");
}
}
if (!_regression && !_timeserie && self_supervised.empty())
_classification = true; // classification is default
// Load weights
_module.load(this->_mlmodel);
_module.freeze_traced(freeze_traced);
_best_metrics = { "map", "meaniou", "mlacc", "delta_score_0.1", "bacc",
"f1", "net_meas", "acc", "L1_mean_error", "eucll" };
_best_metric_value = std::numeric_limits<double>::infinity();
if (lib_ad.has("template_params"))
_template_params = lib_ad;
}
template <class TInputConnectorStrategy, class TOutputConnectorStrategy,
class TMLModel>
void TorchLib<TInputConnectorStrategy, TOutputConnectorStrategy,
TMLModel>::clear_mllib(__attribute__((unused))
const APIData &ad)
{
std::vector<std::string> extensions{ ".json", ".pt", ".ptw" };
fileops::remove_directory_files(this->_mlmodel._repo, extensions);
this->_logger->info("Torchlib service cleared");
}
template <class TInputConnectorStrategy, class TOutputConnectorStrategy,
class TMLModel>
bool TorchLib<TInputConnectorStrategy, TOutputConnectorStrategy,
TMLModel>::is_better(double v1, double v2,
std::string metric_name)
{
if (metric_name == "eucll" || metric_name == "delta_score_0.1"
|| metric_name == "L1_mean_error")
return (v2 > v1);
return (v1 > v2);
}
template <class TInputConnectorStrategy, class TOutputConnectorStrategy,
class TMLModel>
int64_t TorchLib<TInputConnectorStrategy, TOutputConnectorStrategy,
TMLModel>::save_if_best(APIData &meas_out,
int64_t elapsed_it,
TorchSolver &tsolver,
int64_t best_iteration_number)
{
double cur_meas = std::numeric_limits<double>::infinity();
std::string meas;
for (auto m : _best_metrics)
{
if (meas_out.has(m))
{
cur_meas = meas_out.get(m).get<double>();
meas = m;
break;
}
}
if (cur_meas == std::numeric_limits<double>::infinity())
{
// could not find value for measuring best
this->_logger->info(
"could not find any value for measuring best model");
return false;
}
if (_best_metric_value == std::numeric_limits<double>::infinity()
|| is_better(cur_meas, _best_metric_value, meas))
{
if (best_iteration_number != -1)
{
remove_model(best_iteration_number);
}
_best_metric_value = cur_meas;
this->snapshot(elapsed_it, tsolver);
try
{
std::ofstream bestfile;
std::string bestfilename
= this->_mlmodel._repo + this->_mlmodel._best_model_filename;
bestfile.open(bestfilename, std::ios::out);
bestfile << "iteration:" << elapsed_it << std::endl;
bestfile << meas << ":" << cur_meas << std::endl;
bestfile.close();
}
catch (std::exception &e)
{
this->_logger->error("could not write best model file");
}
return elapsed_it;
}
return best_iteration_number;
}
template <class TInputConnectorStrategy, class TOutputConnectorStrategy,
class TMLModel>
void TorchLib<TInputConnectorStrategy, TOutputConnectorStrategy,
TMLModel>::remove_model(int64_t elapsed_it)
{
this->_logger->info("Deleting superseeded model {} ", elapsed_it);
std::remove((this->_mlmodel._repo + "/solver-" + std::to_string(elapsed_it)
+ ".pt")
.c_str());
std::remove((this->_mlmodel._repo + "/checkpoint-"
+ std::to_string(elapsed_it) + ".pt")
.c_str());
std::remove((this->_mlmodel._repo + "/checkpoint-"
+ std::to_string(elapsed_it) + ".ptw")
.c_str());
}
template <class TInputConnectorStrategy, class TOutputConnectorStrategy,
class TMLModel>
void TorchLib<TInputConnectorStrategy, TOutputConnectorStrategy,
TMLModel>::snapshot(int64_t elapsed_it, TorchSolver &tsolver)
{
this->_logger->info("Saving checkpoint after {} iterations", elapsed_it);
this->_module.save_checkpoint(this->_mlmodel, std::to_string(elapsed_it));
// Save optimizer
tsolver.save(this->_mlmodel._repo + "/solver-" + std::to_string(elapsed_it)
+ ".pt");
}
template <class TInputConnectorStrategy, class TOutputConnectorStrategy,
class TMLModel>
int TorchLib<TInputConnectorStrategy, TOutputConnectorStrategy,
TMLModel>::train(const APIData &ad, APIData &out)
{
using namespace std::chrono;
this->_tjob_running.store(true);
TInputConnectorStrategy inputc(this->_inputc);
inputc._train = true;
APIData ad_input = ad.getobj("parameters").getobj("input");
if (ad_input.has("shuffle"))
inputc._dataset.set_shuffle(ad_input.get("shuffle").get<bool>());
try
{
inputc.transform(ad);
_module.post_transform_train<TInputConnectorStrategy>(
_template, _template_params, inputc, this->_mlmodel, _main_device);
}
catch (...)
{
throw;
}
APIData ad_mllib = ad.getobj("parameters").getobj("mllib");
// solver params
int64_t iterations = 1;
int64_t batch_size = 1;
int64_t iter_size = 1;
int64_t test_batch_size = 1;
int64_t test_interval = 1;
int64_t save_period = 0;
TorchSolver tsolver(this->_logger);
Tensor class_weights = {};
// logging parameters
int64_t log_batch_period = 20;
if (ad_mllib.has("template_params"))
_template_params = ad_mllib;
if (ad_mllib.has("solver"))
{
APIData ad_solver = ad_mllib.getobj("solver");
tsolver.configure(ad_solver);
if (ad_solver.has("iterations"))
iterations = ad_solver.get("iterations").get<int>();
if (ad_solver.has("test_interval"))
test_interval = ad_solver.get("test_interval").get<int>();
if (ad_solver.has("iter_size"))
iter_size = ad_solver.get("iter_size").get<int>();
if (ad_solver.has("snapshot"))
save_period = ad_solver.get("snapshot").get<int>();
}
if (ad_mllib.has("net"))
{
APIData ad_net = ad_mllib.getobj("net");
if (ad_net.has("batch_size"))
batch_size = ad_net.get("batch_size").get<int>();
if (ad_net.has("test_batch_size"))
test_batch_size = ad_net.get("test_batch_size").get<int>();
}
if (ad_mllib.has("class_weights"))
{
std::vector<double> cwv
= ad_mllib.get("class_weights").get<std::vector<double>>();
if (cwv.size() != _nclasses)
{
this->_logger->error("class weights given, but number of "
"weights {} do not match "
"number of classes {}, ignoring",
cwv.size(), _nclasses);
}
else
{
this->_logger->info("using class weights");
auto options = torch::TensorOptions().dtype(torch::kFloat64);
class_weights
= torch::from_blob(cwv.data(), { _nclasses }, options)
.to(torch::kFloat32)
.to(_main_device);
}
}
if (iter_size <= 0)
iter_size = 1;
size_t gpu_count = _devices.size();
if (gpu_count > 1)
batch_size *= gpu_count;
// create dataset for evaluation during training
TorchDataset eval_dataset;
if (!inputc._test_dataset.empty())
{
eval_dataset = inputc._test_dataset; //.split(0, 0.1);
}
// create solver
tsolver.create(_module);
int64_t best_iteration_number = -1;
int it = tsolver.resume(ad_mllib, this->_mlmodel, _main_device,
_best_metric_value, best_iteration_number);
bool skip_training = it >= iterations;
if (skip_training)
{
this->_logger->info(
"Model was trained for {} iterations, skipping training", it);
}
else
{
this->_logger->info("Training for {} iterations", iterations - it);
}
tsolver.zero_grad();
_module.train();
// create dataloader
inputc._dataset.reset();
auto dataloader = torch::data::make_data_loader(
std::move(inputc._dataset), data::DataLoaderOptions(batch_size));
int batch_id = 0;
// it is the iteration count (not epoch)
while (it < iterations)
{
if (!this->_tjob_running.load())
break;
double train_loss = 0;
double avg_it_time = 0;
for (TorchBatch batch : *dataloader)
{
auto tstart = system_clock::now();
if (_masked_lm)
{
batch = inputc.generate_masked_lm_batch(batch);
}
std::vector<c10::IValue> in_vals;
for (Tensor tensor : batch.data)
{
in_vals.push_back(tensor.to(_main_device));
}
if (batch.target.size() == 0)
{
throw MLLibInternalException(
"Batch " + std::to_string(batch_id) + ": no target");
}
Tensor y = batch.target.at(0).to(_main_device);
Tensor y_pred;
try
{
y_pred = torch_utils::to_tensor_safe(
_module.forward_on_devices(in_vals, _devices));
}
catch (std::exception &e)
{
this->_logger->error(std::string("Libtorch error: ")
+ e.what());
throw MLLibInternalException(std::string("Libtorch error: ")
+ e.what());
}
Tensor loss;
// As CrossEntropy is not available (Libtorch 1.1) we use
// nllloss
// + log_softmax
if (_seq_training)
{
// Convert [n_batch, sequence_length, vocab_size] to
// [n_batch
// * sequence_length, vocab_size]
// + ignore non-masked tokens (== -1 in target)
loss = torch::nll_loss(
torch::log_softmax(
y_pred.view(IntList{ -1, y_pred.size(2) }), 1),
y.view(IntList{ -1 }), class_weights, Reduction::Mean, -1);
}
else if (_timeserie)
{
if (_module._native != nullptr)
loss = _module._native->loss(_loss, in_vals[0].toTensor(),
y_pred, y);
else
{
if (_loss.empty() || _loss == "L1" || _loss == "l1")
loss = torch::l1_loss(y_pred, y);
else if (_loss == "L2" || _loss == "l2" || _loss == "eucl")
loss = torch::mse_loss(y_pred, y);
else
throw MLLibBadParamException("unknown loss " + _loss);
}
}
else if (_regression)
{
if (_loss.empty() || _loss == "L1" || _loss == "l1")
loss = torch::l1_loss(y_pred, y);
else if (_loss == "L2" || _loss == "l2" || _loss == "eucl")
loss = torch::mse_loss(y_pred, y);
else
throw MLLibBadParamException("unknown loss " + _loss);
}
else if (_classification)
{
loss = torch::nll_loss(torch::log_softmax(y_pred, 1),
y.view(IntList{ -1 }), class_weights);
}
else
{
throw MLLibBadParamException("unexpected model type");
}
if (iter_size > 1)
loss /= iter_size;
double loss_val = loss.item<double>();
train_loss += loss_val;
loss.backward();
auto tstop = system_clock::now();
avg_it_time += duration_cast<milliseconds>(tstop - tstart).count();
if ((batch_id + 1) % iter_size == 0)
{
if (!this->_tjob_running.load())
{
break;
}
try
{
tsolver.step();
tsolver.zero_grad();
}
catch (std::exception &e)
{
this->_logger->error(std::string("Libtorch error: ")
+ e.what());
throw MLLibInternalException(
std::string("Libtorch error: ") + e.what());
}
avg_it_time /= iter_size;
this->add_meas("learning_rate", tsolver.base_lr());
this->add_meas("iteration", it);
this->add_meas("iter_time", avg_it_time);
this->add_meas("remain_time", avg_it_time * iter_size
* (iterations - it)
/ 1000.0);
this->add_meas("train_loss", train_loss);
this->add_meas_per_iter("learning_rate", tsolver.base_lr());
this->add_meas_per_iter("train_loss", train_loss);
int64_t elapsed_it = it + 1;
if (log_batch_period != 0
&& elapsed_it % log_batch_period == 0)
{
this->_logger->info("Iteration {}/{}: loss is {}",
elapsed_it, iterations, train_loss);
}
avg_it_time = 0;
train_loss = 0;
if ((elapsed_it % test_interval == 0 && !eval_dataset.empty())
|| elapsed_it == iterations)
{
// Free memory
loss = torch::empty(1);
y_pred = torch::empty(1);
y = torch::empty(1);
in_vals.clear();
APIData meas_out;
this->_logger->info("Start test");
test(ad, inputc, eval_dataset, test_batch_size, meas_out);
APIData meas_obj = meas_out.getobj("measure");
std::vector<std::string> meas_names = meas_obj.list_keys();
best_iteration_number = save_if_best(
meas_obj, elapsed_it, tsolver, best_iteration_number);
for (auto name : meas_names)
{
if (name != "cmdiag" && name != "cmfull"
&& name != "labels")
{
double mval = meas_obj.get(name).get<double>();
this->_logger->info("{}={}", name, mval);
this->add_meas(name, mval);
this->add_meas_per_iter(name, mval);
}
else if (name == "cmdiag")
{
std::vector<double> mdiag
= meas_obj.get(name)
.get<std::vector<double>>();
std::vector<std::string> cnames;
std::string mdiag_str;
for (size_t i = 0; i < mdiag.size(); i++)
{
mdiag_str
+= this->_mlmodel.get_hcorresp(i) + ":"
+ std::to_string(mdiag.at(i)) + " ";
this->add_meas_per_iter(
name + '_'
+ this->_mlmodel.get_hcorresp(i),
mdiag.at(i));
cnames.push_back(
this->_mlmodel.get_hcorresp(i));
}
this->_logger->info("{}=[{}]", name, mdiag_str);
this->add_meas(name, mdiag, cnames);
}
}
if (elapsed_it == iterations)
out = meas_out;
}
if ((save_period != 0 && elapsed_it % save_period == 0)
|| elapsed_it == iterations)
{
if (best_iteration_number == elapsed_it)
// current model already snapshoted as best model,
// do not remove regular snapshot if it is best
// model
best_iteration_number = -1;
else
snapshot(elapsed_it, tsolver);
}
++it;
if (it >= iterations)
break;
}
++batch_id;
}
if (batch_id == 0
&& iterations > 1) // no batch has run, empty dataset ?
throw MLLibBadParamException(
"couldn't fetch any data bach while training");
}
if (!this->_tjob_running.load())
{
this->_logger->info("Training job interrupted at iteration {}", it);
torch_utils::empty_cuda_cache();
return -1;
}
if (skip_training)
test(ad, inputc, inputc._test_dataset, test_batch_size, out);
torch_utils::empty_cuda_cache();
// Update model after training
this->_mlmodel.read_from_repository(this->_logger);
this->_mlmodel.read_corresp_file();
inputc.response_params(out);
this->_logger->info("Training done.");
return 0;
}
template <class TInputConnectorStrategy, class TOutputConnectorStrategy,
class TMLModel>
int TorchLib<TInputConnectorStrategy, TOutputConnectorStrategy,
TMLModel>::predict(const APIData &ad, APIData &out)
{
int64_t predict_batch_size = 1;
APIData params = ad.getobj("parameters");
APIData output_params = params.getobj("output");
std::string extract_layer;
if (params.has("mllib"))
{
APIData ad_mllib = ad.getobj("parameters").getobj("mllib");
if (ad_mllib.has("net"))
{
APIData ad_net = ad_mllib.getobj("net");
if (ad_net.has("test_batch_size"))
predict_batch_size = ad_net.get("test_batch_size").get<int>();
}
if (ad_mllib.has("extract_layer"))
extract_layer = ad_mllib.get("extract_layer").get<std::string>();
}
bool lstm_continuation = false;
TInputConnectorStrategy inputc(this->_inputc);
if (_module._native != nullptr)
_module._native->update_input_connector(inputc);
this->_stats.transform_start();
TOutputConnectorStrategy outputc(this->_outputc);
try
{
inputc.transform(ad);
_module.post_transform_predict(_template, _template_params, inputc,
this->_mlmodel, _main_device, ad);
if (ad.getobj("parameters").getobj("input").has("continuation")
&& ad.getobj("parameters")
.getobj("input")
.get("continuation")
.get<bool>())
lstm_continuation = true;
else
lstm_continuation = false;
}
catch (...)
{
throw;
}
this->_stats.transform_end();
torch::Device cpu("cpu");
_module.eval();
if (!extract_layer.empty() && !_module.extractable(extract_layer))
{
std::string els;
for (const auto &el : _module.extractable_layers())
els += el + " ";
this->_logger->error("Unknown extract layer " + extract_layer
+ " candidates are " + els);
// or we could set back extract_layer to empty string and continue
throw MLLibBadParamException("unknown extract layer " + extract_layer);
}
if (output_params.has("measure"))
{
APIData meas_out;
test(ad, inputc, inputc._dataset, 1, meas_out);
meas_out.erase("iteration");
meas_out.erase("train_loss");
out.add("measure", meas_out.getobj("measure"));
torch_utils::empty_cuda_cache();
return 0;
}
inputc._dataset.reset(false);
int batch_size = predict_batch_size;
if (lstm_continuation)
batch_size = 1;
auto dataloader = torch::data::make_data_loader(
std::move(inputc._dataset), data::DataLoaderOptions(batch_size));
std::vector<APIData> results_ads;
int nsample = 0;
for (TorchBatch batch : *dataloader)
{
std::vector<c10::IValue> in_vals;
for (Tensor tensor : batch.data)
{
in_vals.push_back(tensor.to(_main_device));
}
this->_stats.inc_inference_count(batch.data[0].size(0));
Tensor output;
try
{
if (extract_layer.empty())
output = torch_utils::to_tensor_safe(_module.forward(in_vals));
else
output = torch_utils::to_tensor_safe(
_module.extract(in_vals, extract_layer));
if (_timeserie)
{
// DO NOTHING
}
else
{
if (_template == "gpt2")
{
// Keep only the prediction for the last token
Tensor input_ids = batch.data[0];
std::vector<Tensor> outputs;
for (int i = 0; i < input_ids.size(0); ++i)
{
// output is (n_batch * sequence_length * vocab_size)
// With gpt2, last token is endoftext so we need to
// take the previous output.
outputs.push_back(
output[i][inputc._lengths.at(i) - 2]);
}
output = torch::stack(outputs);
}
if (_classification)
output = torch::softmax(output, 1).to(cpu);
else
output = output.to(cpu);
}
}
catch (std::exception &e)
{
throw MLLibInternalException(std::string("Libtorch error:")
+ e.what());
}
// Output
if (!extract_layer.empty())
{
for (int j = 0; j < batch_size; j++)
{
APIData rad;
if (!inputc._ids.empty())
rad.add("uri", inputc._ids.at(results_ads.size()));
else
rad.add("uri", std::to_string(results_ads.size()));
if (!inputc._meta_uris.empty())
rad.add("meta_uri",
inputc._meta_uris.at(results_ads.size()));
if (!inputc._index_uris.empty())
rad.add("index_uri",
inputc._index_uris.at(results_ads.size()));
rad.add("loss", static_cast<double>(0.0));
torch::Tensor fo = torch::flatten(output)
.contiguous()
.to(torch::kFloat64)
.to(torch::Device("cpu"));
double *startout = fo.data_ptr<double>();
std::vector<double> vals(startout,
startout + torch ::numel(fo));
rad.add("vals", vals);
results_ads.push_back(rad);
}
}
else
{
if (_module._native != nullptr)
output = _module._native->cleanup_output(output);
if (_classification)
{
int best_count = _nclasses;
if (output_params.has("best"))
best_count = output_params.get("best").get<int>();
std::tuple<Tensor, Tensor> sorted_output
= output.sort(1, true);
auto probs_acc
= std::get<0>(sorted_output).accessor<float, 2>();
auto indices_acc
= std::get<1>(sorted_output).accessor<int64_t, 2>();
for (int i = 0; i < output.size(0); ++i)
{
APIData results_ad;
std::vector<double> probs;
std::vector<std::string> cats;
for (int j = 0; j < best_count; ++j)
{
probs.push_back(probs_acc[i][j]);
int index = indices_acc[i][j];