tiedeman commited on
Commit
81f1f20
1 Parent(s): 39187cd

Initial commit

Browse files
.gitattributes CHANGED
@@ -29,3 +29,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
29
  *.zip filter=lfs diff=lfs merge=lfs -text
30
  *.zst filter=lfs diff=lfs merge=lfs -text
31
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
29
  *.zip filter=lfs diff=lfs merge=lfs -text
30
  *.zst filter=lfs diff=lfs merge=lfs -text
31
  *tfevents* filter=lfs diff=lfs merge=lfs -text
32
+ *.spm filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,724 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - bg
4
+ - bs_Latn
5
+ - es
6
+ - fr
7
+ - hr
8
+ - it
9
+ - itc
10
+ - mk
11
+ - pt
12
+ - sh
13
+ - sl
14
+ - sr_Cyrl
15
+ - sr_Latn
16
+ - zls
17
+
18
+ tags:
19
+ - translation
20
+ - opus-mt-tc
21
+
22
+ license: cc-by-4.0
23
+ model-index:
24
+ - name: opus-mt-tc-big-zls-itc
25
+ results:
26
+ - task:
27
+ name: Translation bul-fra
28
+ type: translation
29
+ args: bul-fra
30
+ dataset:
31
+ name: flores101-devtest
32
+ type: flores_101
33
+ args: bul fra devtest
34
+ metrics:
35
+ - name: BLEU
36
+ type: bleu
37
+ value: 34.4
38
+ - name: chr-F
39
+ type: chrf
40
+ value: 0.60640
41
+ - task:
42
+ name: Translation bul-ita
43
+ type: translation
44
+ args: bul-ita
45
+ dataset:
46
+ name: flores101-devtest
47
+ type: flores_101
48
+ args: bul ita devtest
49
+ metrics:
50
+ - name: BLEU
51
+ type: bleu
52
+ value: 24.0
53
+ - name: chr-F
54
+ type: chrf
55
+ value: 0.54135
56
+ - task:
57
+ name: Translation bul-por
58
+ type: translation
59
+ args: bul-por
60
+ dataset:
61
+ name: flores101-devtest
62
+ type: flores_101
63
+ args: bul por devtest
64
+ metrics:
65
+ - name: BLEU
66
+ type: bleu
67
+ value: 32.4
68
+ - name: chr-F
69
+ type: chrf
70
+ value: 0.59322
71
+ - task:
72
+ name: Translation bul-ron
73
+ type: translation
74
+ args: bul-ron
75
+ dataset:
76
+ name: flores101-devtest
77
+ type: flores_101
78
+ args: bul ron devtest
79
+ metrics:
80
+ - name: BLEU
81
+ type: bleu
82
+ value: 27.1
83
+ - name: chr-F
84
+ type: chrf
85
+ value: 0.55558
86
+ - task:
87
+ name: Translation bul-spa
88
+ type: translation
89
+ args: bul-spa
90
+ dataset:
91
+ name: flores101-devtest
92
+ type: flores_101
93
+ args: bul spa devtest
94
+ metrics:
95
+ - name: BLEU
96
+ type: bleu
97
+ value: 22.4
98
+ - name: chr-F
99
+ type: chrf
100
+ value: 0.50962
101
+ - task:
102
+ name: Translation hrv-fra
103
+ type: translation
104
+ args: hrv-fra
105
+ dataset:
106
+ name: flores101-devtest
107
+ type: flores_101
108
+ args: hrv fra devtest
109
+ metrics:
110
+ - name: BLEU
111
+ type: bleu
112
+ value: 33.1
113
+ - name: chr-F
114
+ type: chrf
115
+ value: 0.59349
116
+ - task:
117
+ name: Translation hrv-ita
118
+ type: translation
119
+ args: hrv-ita
120
+ dataset:
121
+ name: flores101-devtest
122
+ type: flores_101
123
+ args: hrv ita devtest
124
+ metrics:
125
+ - name: BLEU
126
+ type: bleu
127
+ value: 23.5
128
+ - name: chr-F
129
+ type: chrf
130
+ value: 0.52980
131
+ - task:
132
+ name: Translation hrv-por
133
+ type: translation
134
+ args: hrv-por
135
+ dataset:
136
+ name: flores101-devtest
137
+ type: flores_101
138
+ args: hrv por devtest
139
+ metrics:
140
+ - name: BLEU
141
+ type: bleu
142
+ value: 30.2
143
+ - name: chr-F
144
+ type: chrf
145
+ value: 0.57402
146
+ - task:
147
+ name: Translation hrv-ron
148
+ type: translation
149
+ args: hrv-ron
150
+ dataset:
151
+ name: flores101-devtest
152
+ type: flores_101
153
+ args: hrv ron devtest
154
+ metrics:
155
+ - name: BLEU
156
+ type: bleu
157
+ value: 25.9
158
+ - name: chr-F
159
+ type: chrf
160
+ value: 0.53650
161
+ - task:
162
+ name: Translation hrv-spa
163
+ type: translation
164
+ args: hrv-spa
165
+ dataset:
166
+ name: flores101-devtest
167
+ type: flores_101
168
+ args: hrv spa devtest
169
+ metrics:
170
+ - name: BLEU
171
+ type: bleu
172
+ value: 21.5
173
+ - name: chr-F
174
+ type: chrf
175
+ value: 0.50161
176
+ - task:
177
+ name: Translation mkd-fra
178
+ type: translation
179
+ args: mkd-fra
180
+ dataset:
181
+ name: flores101-devtest
182
+ type: flores_101
183
+ args: mkd fra devtest
184
+ metrics:
185
+ - name: BLEU
186
+ type: bleu
187
+ value: 35.2
188
+ - name: chr-F
189
+ type: chrf
190
+ value: 0.60801
191
+ - task:
192
+ name: Translation mkd-ita
193
+ type: translation
194
+ args: mkd-ita
195
+ dataset:
196
+ name: flores101-devtest
197
+ type: flores_101
198
+ args: mkd ita devtest
199
+ metrics:
200
+ - name: BLEU
201
+ type: bleu
202
+ value: 23.9
203
+ - name: chr-F
204
+ type: chrf
205
+ value: 0.53543
206
+ - task:
207
+ name: Translation mkd-por
208
+ type: translation
209
+ args: mkd-por
210
+ dataset:
211
+ name: flores101-devtest
212
+ type: flores_101
213
+ args: mkd por devtest
214
+ metrics:
215
+ - name: BLEU
216
+ type: bleu
217
+ value: 33.9
218
+ - name: chr-F
219
+ type: chrf
220
+ value: 0.59648
221
+ - task:
222
+ name: Translation mkd-ron
223
+ type: translation
224
+ args: mkd-ron
225
+ dataset:
226
+ name: flores101-devtest
227
+ type: flores_101
228
+ args: mkd ron devtest
229
+ metrics:
230
+ - name: BLEU
231
+ type: bleu
232
+ value: 28.0
233
+ - name: chr-F
234
+ type: chrf
235
+ value: 0.54998
236
+ - task:
237
+ name: Translation mkd-spa
238
+ type: translation
239
+ args: mkd-spa
240
+ dataset:
241
+ name: flores101-devtest
242
+ type: flores_101
243
+ args: mkd spa devtest
244
+ metrics:
245
+ - name: BLEU
246
+ type: bleu
247
+ value: 22.8
248
+ - name: chr-F
249
+ type: chrf
250
+ value: 0.51079
251
+ - task:
252
+ name: Translation slv-fra
253
+ type: translation
254
+ args: slv-fra
255
+ dataset:
256
+ name: flores101-devtest
257
+ type: flores_101
258
+ args: slv fra devtest
259
+ metrics:
260
+ - name: BLEU
261
+ type: bleu
262
+ value: 31.5
263
+ - name: chr-F
264
+ type: chrf
265
+ value: 0.58233
266
+ - task:
267
+ name: Translation slv-ita
268
+ type: translation
269
+ args: slv-ita
270
+ dataset:
271
+ name: flores101-devtest
272
+ type: flores_101
273
+ args: slv ita devtest
274
+ metrics:
275
+ - name: BLEU
276
+ type: bleu
277
+ value: 22.4
278
+ - name: chr-F
279
+ type: chrf
280
+ value: 0.52390
281
+ - task:
282
+ name: Translation slv-por
283
+ type: translation
284
+ args: slv-por
285
+ dataset:
286
+ name: flores101-devtest
287
+ type: flores_101
288
+ args: slv por devtest
289
+ metrics:
290
+ - name: BLEU
291
+ type: bleu
292
+ value: 29.0
293
+ - name: chr-F
294
+ type: chrf
295
+ value: 0.56436
296
+ - task:
297
+ name: Translation slv-ron
298
+ type: translation
299
+ args: slv-ron
300
+ dataset:
301
+ name: flores101-devtest
302
+ type: flores_101
303
+ args: slv ron devtest
304
+ metrics:
305
+ - name: BLEU
306
+ type: bleu
307
+ value: 25.0
308
+ - name: chr-F
309
+ type: chrf
310
+ value: 0.53116
311
+ - task:
312
+ name: Translation slv-spa
313
+ type: translation
314
+ args: slv-spa
315
+ dataset:
316
+ name: flores101-devtest
317
+ type: flores_101
318
+ args: slv spa devtest
319
+ metrics:
320
+ - name: BLEU
321
+ type: bleu
322
+ value: 21.1
323
+ - name: chr-F
324
+ type: chrf
325
+ value: 0.49621
326
+ - task:
327
+ name: Translation srp_Cyrl-fra
328
+ type: translation
329
+ args: srp_Cyrl-fra
330
+ dataset:
331
+ name: flores101-devtest
332
+ type: flores_101
333
+ args: srp_Cyrl fra devtest
334
+ metrics:
335
+ - name: BLEU
336
+ type: bleu
337
+ value: 36.0
338
+ - name: chr-F
339
+ type: chrf
340
+ value: 0.62110
341
+ - task:
342
+ name: Translation srp_Cyrl-ita
343
+ type: translation
344
+ args: srp_Cyrl-ita
345
+ dataset:
346
+ name: flores101-devtest
347
+ type: flores_101
348
+ args: srp_Cyrl ita devtest
349
+ metrics:
350
+ - name: BLEU
351
+ type: bleu
352
+ value: 23.9
353
+ - name: chr-F
354
+ type: chrf
355
+ value: 0.54083
356
+ - task:
357
+ name: Translation srp_Cyrl-por
358
+ type: translation
359
+ args: srp_Cyrl-por
360
+ dataset:
361
+ name: flores101-devtest
362
+ type: flores_101
363
+ args: srp_Cyrl por devtest
364
+ metrics:
365
+ - name: BLEU
366
+ type: bleu
367
+ value: 34.9
368
+ - name: chr-F
369
+ type: chrf
370
+ value: 0.61248
371
+ - task:
372
+ name: Translation srp_Cyrl-ron
373
+ type: translation
374
+ args: srp_Cyrl-ron
375
+ dataset:
376
+ name: flores101-devtest
377
+ type: flores_101
378
+ args: srp_Cyrl ron devtest
379
+ metrics:
380
+ - name: BLEU
381
+ type: bleu
382
+ value: 28.8
383
+ - name: chr-F
384
+ type: chrf
385
+ value: 0.56235
386
+ - task:
387
+ name: Translation srp_Cyrl-spa
388
+ type: translation
389
+ args: srp_Cyrl-spa
390
+ dataset:
391
+ name: flores101-devtest
392
+ type: flores_101
393
+ args: srp_Cyrl spa devtest
394
+ metrics:
395
+ - name: BLEU
396
+ type: bleu
397
+ value: 22.8
398
+ - name: chr-F
399
+ type: chrf
400
+ value: 0.51698
401
+ - task:
402
+ name: Translation bul-fra
403
+ type: translation
404
+ args: bul-fra
405
+ dataset:
406
+ name: tatoeba-test-v2021-08-07
407
+ type: tatoeba_mt
408
+ args: bul-fra
409
+ metrics:
410
+ - name: BLEU
411
+ type: bleu
412
+ value: 52.9
413
+ - name: chr-F
414
+ type: chrf
415
+ value: 0.68971
416
+ - task:
417
+ name: Translation bul-ita
418
+ type: translation
419
+ args: bul-ita
420
+ dataset:
421
+ name: tatoeba-test-v2021-08-07
422
+ type: tatoeba_mt
423
+ args: bul-ita
424
+ metrics:
425
+ - name: BLEU
426
+ type: bleu
427
+ value: 45.1
428
+ - name: chr-F
429
+ type: chrf
430
+ value: 0.66412
431
+ - task:
432
+ name: Translation bul-spa
433
+ type: translation
434
+ args: bul-spa
435
+ dataset:
436
+ name: tatoeba-test-v2021-08-07
437
+ type: tatoeba_mt
438
+ args: bul-spa
439
+ metrics:
440
+ - name: BLEU
441
+ type: bleu
442
+ value: 49.7
443
+ - name: chr-F
444
+ type: chrf
445
+ value: 0.66672
446
+ - task:
447
+ name: Translation hbs-fra
448
+ type: translation
449
+ args: hbs-fra
450
+ dataset:
451
+ name: tatoeba-test-v2021-08-07
452
+ type: tatoeba_mt
453
+ args: hbs-fra
454
+ metrics:
455
+ - name: BLEU
456
+ type: bleu
457
+ value: 48.1
458
+ - name: chr-F
459
+ type: chrf
460
+ value: 0.66434
461
+ - task:
462
+ name: Translation hbs-ita
463
+ type: translation
464
+ args: hbs-ita
465
+ dataset:
466
+ name: tatoeba-test-v2021-08-07
467
+ type: tatoeba_mt
468
+ args: hbs-ita
469
+ metrics:
470
+ - name: BLEU
471
+ type: bleu
472
+ value: 53.5
473
+ - name: chr-F
474
+ type: chrf
475
+ value: 0.72381
476
+ - task:
477
+ name: Translation hbs-spa
478
+ type: translation
479
+ args: hbs-spa
480
+ dataset:
481
+ name: tatoeba-test-v2021-08-07
482
+ type: tatoeba_mt
483
+ args: hbs-spa
484
+ metrics:
485
+ - name: BLEU
486
+ type: bleu
487
+ value: 58.0
488
+ - name: chr-F
489
+ type: chrf
490
+ value: 0.73105
491
+ - task:
492
+ name: Translation hrv-fra
493
+ type: translation
494
+ args: hrv-fra
495
+ dataset:
496
+ name: tatoeba-test-v2021-08-07
497
+ type: tatoeba_mt
498
+ args: hrv-fra
499
+ metrics:
500
+ - name: BLEU
501
+ type: bleu
502
+ value: 44.3
503
+ - name: chr-F
504
+ type: chrf
505
+ value: 0.62800
506
+ - task:
507
+ name: Translation hrv-spa
508
+ type: translation
509
+ args: hrv-spa
510
+ dataset:
511
+ name: tatoeba-test-v2021-08-07
512
+ type: tatoeba_mt
513
+ args: hrv-spa
514
+ metrics:
515
+ - name: BLEU
516
+ type: bleu
517
+ value: 57.5
518
+ - name: chr-F
519
+ type: chrf
520
+ value: 0.71370
521
+ - task:
522
+ name: Translation mkd-spa
523
+ type: translation
524
+ args: mkd-spa
525
+ dataset:
526
+ name: tatoeba-test-v2021-08-07
527
+ type: tatoeba_mt
528
+ args: mkd-spa
529
+ metrics:
530
+ - name: BLEU
531
+ type: bleu
532
+ value: 62.1
533
+ - name: chr-F
534
+ type: chrf
535
+ value: 0.75366
536
+ - task:
537
+ name: Translation srp_Latn-ita
538
+ type: translation
539
+ args: srp_Latn-ita
540
+ dataset:
541
+ name: tatoeba-test-v2021-08-07
542
+ type: tatoeba_mt
543
+ args: srp_Latn-ita
544
+ metrics:
545
+ - name: BLEU
546
+ type: bleu
547
+ value: 59.6
548
+ - name: chr-F
549
+ type: chrf
550
+ value: 0.76045
551
+ ---
552
+ # opus-mt-tc-big-zls-itc
553
+
554
+ ## Table of Contents
555
+ - [Model Details](#model-details)
556
+ - [Uses](#uses)
557
+ - [Risks, Limitations and Biases](#risks-limitations-and-biases)
558
+ - [How to Get Started With the Model](#how-to-get-started-with-the-model)
559
+ - [Training](#training)
560
+ - [Evaluation](#evaluation)
561
+ - [Citation Information](#citation-information)
562
+ - [Acknowledgements](#acknowledgements)
563
+
564
+ ## Model Details
565
+
566
+ Neural machine translation model for translating from South Slavic languages (zls) to Italic languages (itc).
567
+
568
+ This model is part of the [OPUS-MT project](https://github.com/Helsinki-NLP/Opus-MT), an effort to make neural machine translation models widely available and accessible for many languages in the world. All models are originally trained using the amazing framework of [Marian NMT](https://marian-nmt.github.io/), an efficient NMT implementation written in pure C++. The models have been converted to pyTorch using the transformers library by huggingface. Training data is taken from [OPUS](https://opus.nlpl.eu/) and training pipelines use the procedures of [OPUS-MT-train](https://github.com/Helsinki-NLP/Opus-MT-train).
569
+ **Model Description:**
570
+ - **Developed by:** Language Technology Research Group at the University of Helsinki
571
+ - **Model Type:** Translation (transformer-big)
572
+ - **Release**: 2022-08-10
573
+ - **License:** CC-BY-4.0
574
+ - **Language(s):**
575
+ - Source Language(s): bos_Latn bul hbs hrv mkd slv srp_Cyrl srp_Latn
576
+ - Target Language(s): fra ita por spa
577
+ - Valid Target Language Labels: >>fra<< >>ita<< >>por<< >>spa<<
578
+ - **Original Model**: [opusTCv20210807_transformer-big_2022-08-10.zip](https://object.pouta.csc.fi/Tatoeba-MT-models/zls-itc/opusTCv20210807_transformer-big_2022-08-10.zip)
579
+ - **Resources for more information:**
580
+ - [OPUS-MT-train GitHub Repo](https://github.com/Helsinki-NLP/OPUS-MT-train)
581
+ - More information about released models for this language pair: [OPUS-MT zls-itc README](https://github.com/Helsinki-NLP/Tatoeba-Challenge/tree/master/models/zls-itc/README.md)
582
+ - [More information about MarianNMT models in the transformers library](https://huggingface.co/docs/transformers/model_doc/marian)
583
+ - [Tatoeba Translation Challenge](https://github.com/Helsinki-NLP/Tatoeba-Challenge/
584
+
585
+ This is a multilingual translation model with multiple target languages. A sentence initial language token is required in the form of `>>id<<` (id = valid target language ID), e.g. `>>fra<<`
586
+
587
+ ## Uses
588
+
589
+ This model can be used for translation and text-to-text generation.
590
+
591
+ ## Risks, Limitations and Biases
592
+
593
+ **CONTENT WARNING: Readers should be aware that the model is trained on various public data sets that may contain content that is disturbing, offensive, and can propagate historical and current stereotypes.**
594
+
595
+ Significant research has explored bias and fairness issues with language models (see, e.g., [Sheng et al. (2021)](https://aclanthology.org/2021.acl-long.330.pdf) and [Bender et al. (2021)](https://dl.acm.org/doi/pdf/10.1145/3442188.3445922)).
596
+
597
+ ## How to Get Started With the Model
598
+
599
+ A short example code:
600
+
601
+ ```python
602
+ from transformers import MarianMTModel, MarianTokenizer
603
+
604
+ src_text = [
605
+ ">>fra<< Dobar dan, kako si?",
606
+ ">>spa<< Znam da je ovo čudno."
607
+ ]
608
+
609
+ model_name = "pytorch-models/opus-mt-tc-big-zls-itc"
610
+ tokenizer = MarianTokenizer.from_pretrained(model_name)
611
+ model = MarianMTModel.from_pretrained(model_name)
612
+ translated = model.generate(**tokenizer(src_text, return_tensors="pt", padding=True))
613
+
614
+ for t in translated:
615
+ print( tokenizer.decode(t, skip_special_tokens=True) )
616
+
617
+ # expected output:
618
+ # Bonjour, comment allez-vous ?
619
+ # Sé que esto es raro.
620
+ ```
621
+
622
+ You can also use OPUS-MT models with the transformers pipelines, for example:
623
+
624
+ ```python
625
+ from transformers import pipeline
626
+ pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-tc-big-zls-itc")
627
+ print(pipe(">>fra<< Dobar dan, kako si?"))
628
+
629
+ # expected output: Bonjour, comment allez-vous ?
630
+ ```
631
+
632
+ ## Training
633
+
634
+ - **Data**: opusTCv20210807 ([source](https://github.com/Helsinki-NLP/Tatoeba-Challenge))
635
+ - **Pre-processing**: SentencePiece (spm32k,spm32k)
636
+ - **Model Type:** transformer-big
637
+ - **Original MarianNMT Model**: [opusTCv20210807_transformer-big_2022-08-10.zip](https://object.pouta.csc.fi/Tatoeba-MT-models/zls-itc/opusTCv20210807_transformer-big_2022-08-10.zip)
638
+ - **Training Scripts**: [GitHub Repo](https://github.com/Helsinki-NLP/OPUS-MT-train)
639
+
640
+ ## Evaluation
641
+
642
+ * test set translations: [opusTCv20210807_transformer-big_2022-08-10.test.txt](https://object.pouta.csc.fi/Tatoeba-MT-models/zls-itc/opusTCv20210807_transformer-big_2022-08-10.test.txt)
643
+ * test set scores: [opusTCv20210807_transformer-big_2022-08-10.eval.txt](https://object.pouta.csc.fi/Tatoeba-MT-models/zls-itc/opusTCv20210807_transformer-big_2022-08-10.eval.txt)
644
+ * benchmark results: [benchmark_results.txt](benchmark_results.txt)
645
+ * benchmark output: [benchmark_translations.zip](benchmark_translations.zip)
646
+
647
+ | langpair | testset | chr-F | BLEU | #sent | #words |
648
+ |----------|---------|-------|-------|-------|--------|
649
+ | bul-fra | tatoeba-test-v2021-08-07 | 0.68971 | 52.9 | 446 | 3669 |
650
+ | bul-ita | tatoeba-test-v2021-08-07 | 0.66412 | 45.1 | 2500 | 16951 |
651
+ | bul-spa | tatoeba-test-v2021-08-07 | 0.66672 | 49.7 | 286 | 1783 |
652
+ | hbs-fra | tatoeba-test-v2021-08-07 | 0.66434 | 48.1 | 474 | 3370 |
653
+ | hbs-ita | tatoeba-test-v2021-08-07 | 0.72381 | 53.5 | 534 | 3208 |
654
+ | hbs-spa | tatoeba-test-v2021-08-07 | 0.73105 | 58.0 | 607 | 3766 |
655
+ | hrv-fra | tatoeba-test-v2021-08-07 | 0.62800 | 44.3 | 258 | 1943 |
656
+ | hrv-spa | tatoeba-test-v2021-08-07 | 0.71370 | 57.5 | 254 | 1702 |
657
+ | mkd-spa | tatoeba-test-v2021-08-07 | 0.75366 | 62.1 | 217 | 1121 |
658
+ | srp_Latn-ita | tatoeba-test-v2021-08-07 | 0.76045 | 59.6 | 212 | 1292 |
659
+ | bul-fra | flores101-devtest | 0.60640 | 34.4 | 1012 | 28343 |
660
+ | bul-ita | flores101-devtest | 0.54135 | 24.0 | 1012 | 27306 |
661
+ | bul-por | flores101-devtest | 0.59322 | 32.4 | 1012 | 26519 |
662
+ | bul-ron | flores101-devtest | 0.55558 | 27.1 | 1012 | 26799 |
663
+ | bul-spa | flores101-devtest | 0.50962 | 22.4 | 1012 | 29199 |
664
+ | hrv-fra | flores101-devtest | 0.59349 | 33.1 | 1012 | 28343 |
665
+ | hrv-ita | flores101-devtest | 0.52980 | 23.5 | 1012 | 27306 |
666
+ | hrv-por | flores101-devtest | 0.57402 | 30.2 | 1012 | 26519 |
667
+ | hrv-ron | flores101-devtest | 0.53650 | 25.9 | 1012 | 26799 |
668
+ | hrv-spa | flores101-devtest | 0.50161 | 21.5 | 1012 | 29199 |
669
+ | mkd-fra | flores101-devtest | 0.60801 | 35.2 | 1012 | 28343 |
670
+ | mkd-ita | flores101-devtest | 0.53543 | 23.9 | 1012 | 27306 |
671
+ | mkd-por | flores101-devtest | 0.59648 | 33.9 | 1012 | 26519 |
672
+ | mkd-ron | flores101-devtest | 0.54998 | 28.0 | 1012 | 26799 |
673
+ | mkd-spa | flores101-devtest | 0.51079 | 22.8 | 1012 | 29199 |
674
+ | slv-fra | flores101-devtest | 0.58233 | 31.5 | 1012 | 28343 |
675
+ | slv-ita | flores101-devtest | 0.52390 | 22.4 | 1012 | 27306 |
676
+ | slv-por | flores101-devtest | 0.56436 | 29.0 | 1012 | 26519 |
677
+ | slv-ron | flores101-devtest | 0.53116 | 25.0 | 1012 | 26799 |
678
+ | slv-spa | flores101-devtest | 0.49621 | 21.1 | 1012 | 29199 |
679
+ | srp_Cyrl-fra | flores101-devtest | 0.62110 | 36.0 | 1012 | 28343 |
680
+ | srp_Cyrl-ita | flores101-devtest | 0.54083 | 23.9 | 1012 | 27306 |
681
+ | srp_Cyrl-por | flores101-devtest | 0.61248 | 34.9 | 1012 | 26519 |
682
+ | srp_Cyrl-ron | flores101-devtest | 0.56235 | 28.8 | 1012 | 26799 |
683
+ | srp_Cyrl-spa | flores101-devtest | 0.51698 | 22.8 | 1012 | 29199 |
684
+
685
+ ## Citation Information
686
+
687
+ * Publications: [OPUS-MT – Building open translation services for the World](https://aclanthology.org/2020.eamt-1.61/) and [The Tatoeba Translation Challenge – Realistic Data Sets for Low Resource and Multilingual MT](https://aclanthology.org/2020.wmt-1.139/) (Please, cite if you use this model.)
688
+
689
+ ```
690
+ @inproceedings{tiedemann-thottingal-2020-opus,
691
+ title = "{OPUS}-{MT} {--} Building open translation services for the World",
692
+ author = {Tiedemann, J{\"o}rg and Thottingal, Santhosh},
693
+ booktitle = "Proceedings of the 22nd Annual Conference of the European Association for Machine Translation",
694
+ month = nov,
695
+ year = "2020",
696
+ address = "Lisboa, Portugal",
697
+ publisher = "European Association for Machine Translation",
698
+ url = "https://aclanthology.org/2020.eamt-1.61",
699
+ pages = "479--480",
700
+ }
701
+
702
+ @inproceedings{tiedemann-2020-tatoeba,
703
+ title = "The Tatoeba Translation Challenge {--} Realistic Data Sets for Low Resource and Multilingual {MT}",
704
+ author = {Tiedemann, J{\"o}rg},
705
+ booktitle = "Proceedings of the Fifth Conference on Machine Translation",
706
+ month = nov,
707
+ year = "2020",
708
+ address = "Online",
709
+ publisher = "Association for Computational Linguistics",
710
+ url = "https://aclanthology.org/2020.wmt-1.139",
711
+ pages = "1174--1182",
712
+ }
713
+ ```
714
+
715
+ ## Acknowledgements
716
+
717
+ The work is supported by the [European Language Grid](https://www.european-language-grid.eu/) as [pilot project 2866](https://live.european-language-grid.eu/catalogue/#/resource/projects/2866), by the [FoTran project](https://www.helsinki.fi/en/researchgroups/natural-language-understanding-with-cross-lingual-grounding), funded by the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 771113), and the [MeMAD project](https://memad.eu/), funded by the European Union’s Horizon 2020 Research and Innovation Programme under grant agreement No 780069. We are also grateful for the generous computational resources and IT infrastructure provided by [CSC -- IT Center for Science](https://www.csc.fi/), Finland.
718
+
719
+ ## Model conversion info
720
+
721
+ * transformers version: 4.16.2
722
+ * OPUS-MT git hash: 8b9f0b0
723
+ * port time: Fri Aug 12 13:57:47 EEST 2022
724
+ * port machine: LM0-400-22516.local
benchmark_results.txt ADDED
@@ -0,0 +1,74 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ bul-fra flores101-dev 0.60515 34.0 997 26706
2
+ bul-ita flores101-dev 0.53356 22.8 997 25840
3
+ bul-por flores101-dev 0.58413 31.3 997 25287
4
+ bul-ron flores101-dev 0.55870 27.9 997 25616
5
+ bul-spa flores101-dev 0.50507 22.2 997 27793
6
+ hrv-fra flores101-dev 0.59160 32.9 997 26706
7
+ hrv-ita flores101-dev 0.52624 22.6 997 25840
8
+ hrv-por flores101-dev 0.56567 29.7 997 25287
9
+ hrv-ron flores101-dev 0.53565 26.0 997 25616
10
+ hrv-spa flores101-dev 0.49356 21.2 997 27793
11
+ mkd-fra flores101-dev 0.60379 34.6 997 26706
12
+ mkd-ita flores101-dev 0.53334 23.5 997 25840
13
+ mkd-por flores101-dev 0.59103 32.9 997 25287
14
+ mkd-ron flores101-dev 0.55096 28.2 997 25616
15
+ mkd-spa flores101-dev 0.50218 22.3 997 27793
16
+ slv-fra flores101-dev 0.57857 31.1 997 26706
17
+ slv-ita flores101-dev 0.52418 22.2 997 25840
18
+ slv-por flores101-dev 0.55881 28.3 997 25287
19
+ slv-ron flores101-dev 0.53118 25.9 997 25616
20
+ slv-spa flores101-dev 0.49112 21.2 997 27793
21
+ srp_Cyrl-fra flores101-dev 0.62093 36.3 997 26706
22
+ srp_Cyrl-ita flores101-dev 0.53829 23.4 997 25840
23
+ srp_Cyrl-por flores101-dev 0.60678 34.7 997 25287
24
+ srp_Cyrl-ron flores101-dev 0.56395 29.0 997 25616
25
+ srp_Cyrl-spa flores101-dev 0.50659 22.1 997 27793
26
+ bul-fra flores101-devtest 0.60640 34.4 1012 28343
27
+ bul-ita flores101-devtest 0.54135 24.0 1012 27306
28
+ bul-por flores101-devtest 0.59322 32.4 1012 26519
29
+ bul-ron flores101-devtest 0.55558 27.1 1012 26799
30
+ bul-spa flores101-devtest 0.50962 22.4 1012 29199
31
+ hrv-fra flores101-devtest 0.59349 33.1 1012 28343
32
+ hrv-ita flores101-devtest 0.52980 23.5 1012 27306
33
+ hrv-por flores101-devtest 0.57402 30.2 1012 26519
34
+ hrv-ron flores101-devtest 0.53650 25.9 1012 26799
35
+ hrv-spa flores101-devtest 0.50161 21.5 1012 29199
36
+ mkd-fra flores101-devtest 0.60801 35.2 1012 28343
37
+ mkd-ita flores101-devtest 0.53543 23.9 1012 27306
38
+ mkd-por flores101-devtest 0.59648 33.9 1012 26519
39
+ mkd-ron flores101-devtest 0.54998 28.0 1012 26799
40
+ mkd-spa flores101-devtest 0.51079 22.8 1012 29199
41
+ slv-fra flores101-devtest 0.58233 31.5 1012 28343
42
+ slv-ita flores101-devtest 0.52390 22.4 1012 27306
43
+ slv-por flores101-devtest 0.56436 29.0 1012 26519
44
+ slv-ron flores101-devtest 0.53116 25.0 1012 26799
45
+ slv-spa flores101-devtest 0.49621 21.1 1012 29199
46
+ srp_Cyrl-fra flores101-devtest 0.62110 36.0 1012 28343
47
+ srp_Cyrl-ita flores101-devtest 0.54083 23.9 1012 27306
48
+ srp_Cyrl-por flores101-devtest 0.61248 34.9 1012 26519
49
+ srp_Cyrl-ron flores101-devtest 0.56235 28.8 1012 26799
50
+ srp_Cyrl-spa flores101-devtest 0.51698 22.8 1012 29199
51
+ hbs-spa tatoeba-test-v2020-07-28 0.73160 58.1 606 3762
52
+ hrv-spa tatoeba-test-v2020-07-28 0.71492 57.6 253 1698
53
+ slv-fra tatoeba-test-v2020-07-28 0.26482 10.3 442 3662
54
+ bul-fra tatoeba-test-v2021-03-30 0.68891 52.8 448 3689
55
+ bul-spa tatoeba-test-v2021-03-30 0.66812 49.8 288 1791
56
+ hbs-fra tatoeba-test-v2021-03-30 0.66430 48.0 479 3404
57
+ hbs-ita tatoeba-test-v2021-03-30 0.72403 53.4 537 3226
58
+ hbs-spa tatoeba-test-v2021-03-30 0.73080 58.1 615 3812
59
+ hrv-fra tatoeba-test-v2021-03-30 0.62774 44.1 261 1969
60
+ hrv-spa tatoeba-test-v2021-03-30 0.71268 57.4 261 1739
61
+ mkd-spa tatoeba-test-v2021-03-30 0.75405 62.1 223 1149
62
+ slv-fra tatoeba-test-v2021-03-30 0.26515 10.2 444 3673
63
+ srp_Latn-ita tatoeba-test-v2021-03-30 0.76034 59.5 213 1297
64
+ bul-fra tatoeba-test-v2021-08-07 0.68971 52.9 446 3669
65
+ bul-ita tatoeba-test-v2021-08-07 0.66412 45.1 2500 16951
66
+ bul-spa tatoeba-test-v2021-08-07 0.66672 49.7 286 1783
67
+ hbs-fra tatoeba-test-v2021-08-07 0.66434 48.1 474 3370
68
+ hbs-ita tatoeba-test-v2021-08-07 0.72381 53.5 534 3208
69
+ hbs-spa tatoeba-test-v2021-08-07 0.73105 58.0 607 3766
70
+ hrv-fra tatoeba-test-v2021-08-07 0.62800 44.3 258 1943
71
+ hrv-spa tatoeba-test-v2021-08-07 0.71370 57.5 254 1702
72
+ mkd-spa tatoeba-test-v2021-08-07 0.75366 62.1 217 1121
73
+ slv-fra tatoeba-test-v2021-08-07 0.26575 11.4 448 3792
74
+ srp_Latn-ita tatoeba-test-v2021-08-07 0.76045 59.6 212 1292
benchmark_translations.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c08e88024b0728240f100e49c837b1ae39616cd81050113f908bbefb6ffc44b7
3
+ size 9194324
config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "activation_dropout": 0.0,
3
+ "activation_function": "relu",
4
+ "architectures": [
5
+ "MarianMTModel"
6
+ ],
7
+ "attention_dropout": 0.0,
8
+ "bad_words_ids": [
9
+ [
10
+ 59412
11
+ ]
12
+ ],
13
+ "bos_token_id": 0,
14
+ "classifier_dropout": 0.0,
15
+ "d_model": 1024,
16
+ "decoder_attention_heads": 16,
17
+ "decoder_ffn_dim": 4096,
18
+ "decoder_layerdrop": 0.0,
19
+ "decoder_layers": 6,
20
+ "decoder_start_token_id": 59412,
21
+ "decoder_vocab_size": 59413,
22
+ "dropout": 0.1,
23
+ "encoder_attention_heads": 16,
24
+ "encoder_ffn_dim": 4096,
25
+ "encoder_layerdrop": 0.0,
26
+ "encoder_layers": 6,
27
+ "eos_token_id": 34288,
28
+ "forced_eos_token_id": 34288,
29
+ "init_std": 0.02,
30
+ "is_encoder_decoder": true,
31
+ "max_length": 512,
32
+ "max_position_embeddings": 1024,
33
+ "model_type": "marian",
34
+ "normalize_embedding": false,
35
+ "num_beams": 4,
36
+ "num_hidden_layers": 6,
37
+ "pad_token_id": 59412,
38
+ "scale_embedding": true,
39
+ "share_encoder_decoder_embeddings": true,
40
+ "static_position_embeddings": true,
41
+ "torch_dtype": "float16",
42
+ "transformers_version": "4.18.0.dev0",
43
+ "use_cache": true,
44
+ "vocab_size": 59413
45
+ }
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:447149996e9bc9e6cc1c946ff230ac6e384bbbebccabe01547fa0ffa1bda3c1b
3
+ size 596280899
source.spm ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4fe699c379148912d2c00945aec2f34716e12701ffbe707fbda59fc229ee0b6f
3
+ size 884536
special_tokens_map.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>"}
target.spm ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6f01392fe95e7d77c0c79eb784837cb3566c8beaeef019d6882babac3c4ad859
3
+ size 807963
tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"source_lang": "zls", "target_lang": "itc", "unk_token": "<unk>", "eos_token": "</s>", "pad_token": "<pad>", "model_max_length": 512, "sp_model_kwargs": {}, "separate_vocabs": false, "special_tokens_map_file": null, "name_or_path": "marian-models/opusTCv20210807_transformer-big_2022-08-10/zls-itc", "tokenizer_class": "MarianTokenizer"}
vocab.json ADDED
The diff for this file is too large to render. See raw diff