Initial commit
Browse files- .gitattributes +1 -0
- README.md +526 -0
- benchmark_results.txt +47 -0
- benchmark_translations.zip +3 -0
- config.json +45 -0
- pytorch_model.bin +3 -0
- source.spm +3 -0
- special_tokens_map.json +1 -0
- target.spm +3 -0
- tokenizer_config.json +1 -0
- vocab.json +0 -0
.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 +1,526 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language:
|
3 |
+
- da
|
4 |
+
- fo
|
5 |
+
- gmq
|
6 |
+
- is
|
7 |
+
- nb
|
8 |
+
- nn
|
9 |
+
- no
|
10 |
+
- sv
|
11 |
+
|
12 |
+
tags:
|
13 |
+
- translation
|
14 |
+
- opus-mt-tc
|
15 |
+
|
16 |
+
license: cc-by-4.0
|
17 |
+
model-index:
|
18 |
+
- name: opus-mt-tc-big-gmq-gmq
|
19 |
+
results:
|
20 |
+
- task:
|
21 |
+
name: Translation isl-swe
|
22 |
+
type: translation
|
23 |
+
args: isl-swe
|
24 |
+
dataset:
|
25 |
+
name: europeana2021
|
26 |
+
type: europeana2021
|
27 |
+
args: isl-swe
|
28 |
+
metrics:
|
29 |
+
- name: BLEU
|
30 |
+
type: bleu
|
31 |
+
value: 22.2
|
32 |
+
- name: chr-F
|
33 |
+
type: chrf
|
34 |
+
value: 0.45562
|
35 |
+
- task:
|
36 |
+
name: Translation nob-isl
|
37 |
+
type: translation
|
38 |
+
args: nob-isl
|
39 |
+
dataset:
|
40 |
+
name: europeana2021
|
41 |
+
type: europeana2021
|
42 |
+
args: nob-isl
|
43 |
+
metrics:
|
44 |
+
- name: BLEU
|
45 |
+
type: bleu
|
46 |
+
value: 29.7
|
47 |
+
- name: chr-F
|
48 |
+
type: chrf
|
49 |
+
value: 0.54171
|
50 |
+
- task:
|
51 |
+
name: Translation nob-swe
|
52 |
+
type: translation
|
53 |
+
args: nob-swe
|
54 |
+
dataset:
|
55 |
+
name: europeana2021
|
56 |
+
type: europeana2021
|
57 |
+
args: nob-swe
|
58 |
+
metrics:
|
59 |
+
- name: BLEU
|
60 |
+
type: bleu
|
61 |
+
value: 54.0
|
62 |
+
- name: chr-F
|
63 |
+
type: chrf
|
64 |
+
value: 0.73891
|
65 |
+
- task:
|
66 |
+
name: Translation dan-isl
|
67 |
+
type: translation
|
68 |
+
args: dan-isl
|
69 |
+
dataset:
|
70 |
+
name: flores101-devtest
|
71 |
+
type: flores_101
|
72 |
+
args: dan isl devtest
|
73 |
+
metrics:
|
74 |
+
- name: BLEU
|
75 |
+
type: bleu
|
76 |
+
value: 22.2
|
77 |
+
- name: chr-F
|
78 |
+
type: chrf
|
79 |
+
value: 0.50227
|
80 |
+
- task:
|
81 |
+
name: Translation dan-nob
|
82 |
+
type: translation
|
83 |
+
args: dan-nob
|
84 |
+
dataset:
|
85 |
+
name: flores101-devtest
|
86 |
+
type: flores_101
|
87 |
+
args: dan nob devtest
|
88 |
+
metrics:
|
89 |
+
- name: BLEU
|
90 |
+
type: bleu
|
91 |
+
value: 28.6
|
92 |
+
- name: chr-F
|
93 |
+
type: chrf
|
94 |
+
value: 0.58445
|
95 |
+
- task:
|
96 |
+
name: Translation dan-swe
|
97 |
+
type: translation
|
98 |
+
args: dan-swe
|
99 |
+
dataset:
|
100 |
+
name: flores101-devtest
|
101 |
+
type: flores_101
|
102 |
+
args: dan swe devtest
|
103 |
+
metrics:
|
104 |
+
- name: BLEU
|
105 |
+
type: bleu
|
106 |
+
value: 38.5
|
107 |
+
- name: chr-F
|
108 |
+
type: chrf
|
109 |
+
value: 0.65000
|
110 |
+
- task:
|
111 |
+
name: Translation isl-dan
|
112 |
+
type: translation
|
113 |
+
args: isl-dan
|
114 |
+
dataset:
|
115 |
+
name: flores101-devtest
|
116 |
+
type: flores_101
|
117 |
+
args: isl dan devtest
|
118 |
+
metrics:
|
119 |
+
- name: BLEU
|
120 |
+
type: bleu
|
121 |
+
value: 27.2
|
122 |
+
- name: chr-F
|
123 |
+
type: chrf
|
124 |
+
value: 0.53630
|
125 |
+
- task:
|
126 |
+
name: Translation isl-nob
|
127 |
+
type: translation
|
128 |
+
args: isl-nob
|
129 |
+
dataset:
|
130 |
+
name: flores101-devtest
|
131 |
+
type: flores_101
|
132 |
+
args: isl nob devtest
|
133 |
+
metrics:
|
134 |
+
- name: BLEU
|
135 |
+
type: bleu
|
136 |
+
value: 20.5
|
137 |
+
- name: chr-F
|
138 |
+
type: chrf
|
139 |
+
value: 0.49434
|
140 |
+
- task:
|
141 |
+
name: Translation isl-swe
|
142 |
+
type: translation
|
143 |
+
args: isl-swe
|
144 |
+
dataset:
|
145 |
+
name: flores101-devtest
|
146 |
+
type: flores_101
|
147 |
+
args: isl swe devtest
|
148 |
+
metrics:
|
149 |
+
- name: BLEU
|
150 |
+
type: bleu
|
151 |
+
value: 26.0
|
152 |
+
- name: chr-F
|
153 |
+
type: chrf
|
154 |
+
value: 0.53373
|
155 |
+
- task:
|
156 |
+
name: Translation nob-dan
|
157 |
+
type: translation
|
158 |
+
args: nob-dan
|
159 |
+
dataset:
|
160 |
+
name: flores101-devtest
|
161 |
+
type: flores_101
|
162 |
+
args: nob dan devtest
|
163 |
+
metrics:
|
164 |
+
- name: BLEU
|
165 |
+
type: bleu
|
166 |
+
value: 31.7
|
167 |
+
- name: chr-F
|
168 |
+
type: chrf
|
169 |
+
value: 0.59657
|
170 |
+
- task:
|
171 |
+
name: Translation nob-isl
|
172 |
+
type: translation
|
173 |
+
args: nob-isl
|
174 |
+
dataset:
|
175 |
+
name: flores101-devtest
|
176 |
+
type: flores_101
|
177 |
+
args: nob isl devtest
|
178 |
+
metrics:
|
179 |
+
- name: BLEU
|
180 |
+
type: bleu
|
181 |
+
value: 18.9
|
182 |
+
- name: chr-F
|
183 |
+
type: chrf
|
184 |
+
value: 0.47432
|
185 |
+
- task:
|
186 |
+
name: Translation nob-swe
|
187 |
+
type: translation
|
188 |
+
args: nob-swe
|
189 |
+
dataset:
|
190 |
+
name: flores101-devtest
|
191 |
+
type: flores_101
|
192 |
+
args: nob swe devtest
|
193 |
+
metrics:
|
194 |
+
- name: BLEU
|
195 |
+
type: bleu
|
196 |
+
value: 31.3
|
197 |
+
- name: chr-F
|
198 |
+
type: chrf
|
199 |
+
value: 0.60030
|
200 |
+
- task:
|
201 |
+
name: Translation swe-dan
|
202 |
+
type: translation
|
203 |
+
args: swe-dan
|
204 |
+
dataset:
|
205 |
+
name: flores101-devtest
|
206 |
+
type: flores_101
|
207 |
+
args: swe dan devtest
|
208 |
+
metrics:
|
209 |
+
- name: BLEU
|
210 |
+
type: bleu
|
211 |
+
value: 39.0
|
212 |
+
- name: chr-F
|
213 |
+
type: chrf
|
214 |
+
value: 0.64340
|
215 |
+
- task:
|
216 |
+
name: Translation swe-isl
|
217 |
+
type: translation
|
218 |
+
args: swe-isl
|
219 |
+
dataset:
|
220 |
+
name: flores101-devtest
|
221 |
+
type: flores_101
|
222 |
+
args: swe isl devtest
|
223 |
+
metrics:
|
224 |
+
- name: BLEU
|
225 |
+
type: bleu
|
226 |
+
value: 21.7
|
227 |
+
- name: chr-F
|
228 |
+
type: chrf
|
229 |
+
value: 0.49590
|
230 |
+
- task:
|
231 |
+
name: Translation swe-nob
|
232 |
+
type: translation
|
233 |
+
args: swe-nob
|
234 |
+
dataset:
|
235 |
+
name: flores101-devtest
|
236 |
+
type: flores_101
|
237 |
+
args: swe nob devtest
|
238 |
+
metrics:
|
239 |
+
- name: BLEU
|
240 |
+
type: bleu
|
241 |
+
value: 28.9
|
242 |
+
- name: chr-F
|
243 |
+
type: chrf
|
244 |
+
value: 0.58336
|
245 |
+
- task:
|
246 |
+
name: Translation dan-nob
|
247 |
+
type: translation
|
248 |
+
args: dan-nob
|
249 |
+
dataset:
|
250 |
+
name: tatoeba-test-v2021-08-07
|
251 |
+
type: tatoeba_mt
|
252 |
+
args: dan-nob
|
253 |
+
metrics:
|
254 |
+
- name: BLEU
|
255 |
+
type: bleu
|
256 |
+
value: 78.2
|
257 |
+
- name: chr-F
|
258 |
+
type: chrf
|
259 |
+
value: 0.87556
|
260 |
+
- task:
|
261 |
+
name: Translation dan-swe
|
262 |
+
type: translation
|
263 |
+
args: dan-swe
|
264 |
+
dataset:
|
265 |
+
name: tatoeba-test-v2021-08-07
|
266 |
+
type: tatoeba_mt
|
267 |
+
args: dan-swe
|
268 |
+
metrics:
|
269 |
+
- name: BLEU
|
270 |
+
type: bleu
|
271 |
+
value: 72.5
|
272 |
+
- name: chr-F
|
273 |
+
type: chrf
|
274 |
+
value: 0.83556
|
275 |
+
- task:
|
276 |
+
name: Translation nno-nob
|
277 |
+
type: translation
|
278 |
+
args: nno-nob
|
279 |
+
dataset:
|
280 |
+
name: tatoeba-test-v2021-08-07
|
281 |
+
type: tatoeba_mt
|
282 |
+
args: nno-nob
|
283 |
+
metrics:
|
284 |
+
- name: BLEU
|
285 |
+
type: bleu
|
286 |
+
value: 78.9
|
287 |
+
- name: chr-F
|
288 |
+
type: chrf
|
289 |
+
value: 0.88349
|
290 |
+
- task:
|
291 |
+
name: Translation nob-dan
|
292 |
+
type: translation
|
293 |
+
args: nob-dan
|
294 |
+
dataset:
|
295 |
+
name: tatoeba-test-v2021-08-07
|
296 |
+
type: tatoeba_mt
|
297 |
+
args: nob-dan
|
298 |
+
metrics:
|
299 |
+
- name: BLEU
|
300 |
+
type: bleu
|
301 |
+
value: 73.9
|
302 |
+
- name: chr-F
|
303 |
+
type: chrf
|
304 |
+
value: 0.85345
|
305 |
+
- task:
|
306 |
+
name: Translation nob-nno
|
307 |
+
type: translation
|
308 |
+
args: nob-nno
|
309 |
+
dataset:
|
310 |
+
name: tatoeba-test-v2021-08-07
|
311 |
+
type: tatoeba_mt
|
312 |
+
args: nob-nno
|
313 |
+
metrics:
|
314 |
+
- name: BLEU
|
315 |
+
type: bleu
|
316 |
+
value: 55.2
|
317 |
+
- name: chr-F
|
318 |
+
type: chrf
|
319 |
+
value: 0.74571
|
320 |
+
- task:
|
321 |
+
name: Translation nob-swe
|
322 |
+
type: translation
|
323 |
+
args: nob-swe
|
324 |
+
dataset:
|
325 |
+
name: tatoeba-test-v2021-08-07
|
326 |
+
type: tatoeba_mt
|
327 |
+
args: nob-swe
|
328 |
+
metrics:
|
329 |
+
- name: BLEU
|
330 |
+
type: bleu
|
331 |
+
value: 73.9
|
332 |
+
- name: chr-F
|
333 |
+
type: chrf
|
334 |
+
value: 0.84747
|
335 |
+
- task:
|
336 |
+
name: Translation swe-dan
|
337 |
+
type: translation
|
338 |
+
args: swe-dan
|
339 |
+
dataset:
|
340 |
+
name: tatoeba-test-v2021-08-07
|
341 |
+
type: tatoeba_mt
|
342 |
+
args: swe-dan
|
343 |
+
metrics:
|
344 |
+
- name: BLEU
|
345 |
+
type: bleu
|
346 |
+
value: 72.6
|
347 |
+
- name: chr-F
|
348 |
+
type: chrf
|
349 |
+
value: 0.83392
|
350 |
+
- task:
|
351 |
+
name: Translation swe-nob
|
352 |
+
type: translation
|
353 |
+
args: swe-nob
|
354 |
+
dataset:
|
355 |
+
name: tatoeba-test-v2021-08-07
|
356 |
+
type: tatoeba_mt
|
357 |
+
args: swe-nob
|
358 |
+
metrics:
|
359 |
+
- name: BLEU
|
360 |
+
type: bleu
|
361 |
+
value: 76.3
|
362 |
+
- name: chr-F
|
363 |
+
type: chrf
|
364 |
+
value: 0.85815
|
365 |
+
---
|
366 |
+
# opus-mt-tc-big-gmq-gmq
|
367 |
+
|
368 |
+
## Table of Contents
|
369 |
+
- [Model Details](#model-details)
|
370 |
+
- [Uses](#uses)
|
371 |
+
- [Risks, Limitations and Biases](#risks-limitations-and-biases)
|
372 |
+
- [How to Get Started With the Model](#how-to-get-started-with-the-model)
|
373 |
+
- [Training](#training)
|
374 |
+
- [Evaluation](#evaluation)
|
375 |
+
- [Citation Information](#citation-information)
|
376 |
+
- [Acknowledgements](#acknowledgements)
|
377 |
+
|
378 |
+
## Model Details
|
379 |
+
|
380 |
+
Neural machine translation model for translating from North Germanic languages (gmq) to North Germanic languages (gmq).
|
381 |
+
|
382 |
+
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).
|
383 |
+
**Model Description:**
|
384 |
+
- **Developed by:** Language Technology Research Group at the University of Helsinki
|
385 |
+
- **Model Type:** Translation (transformer-big)
|
386 |
+
- **Release**: 2022-07-29
|
387 |
+
- **License:** CC-BY-4.0
|
388 |
+
- **Language(s):**
|
389 |
+
- Source Language(s): dan fao isl nno nob nor swe
|
390 |
+
- Target Language(s): dan isl nno nob nor swe
|
391 |
+
- Valid Target Language Labels: >>dan<< >>isl<< >>nno<< >>nob<< >>nor<< >>swe<<
|
392 |
+
- **Original Model**: [opusTCv20210807_transformer-big_2022-07-29.zip](https://object.pouta.csc.fi/Tatoeba-MT-models/gmq-gmq/opusTCv20210807_transformer-big_2022-07-29.zip)
|
393 |
+
- **Resources for more information:**
|
394 |
+
- [OPUS-MT-train GitHub Repo](https://github.com/Helsinki-NLP/OPUS-MT-train)
|
395 |
+
- More information about released models for this language pair: [OPUS-MT gmq-gmq README](https://github.com/Helsinki-NLP/Tatoeba-Challenge/tree/master/models/gmq-gmq/README.md)
|
396 |
+
- [More information about MarianNMT models in the transformers library](https://huggingface.co/docs/transformers/model_doc/marian)
|
397 |
+
- [Tatoeba Translation Challenge](https://github.com/Helsinki-NLP/Tatoeba-Challenge/
|
398 |
+
|
399 |
+
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. `>>dan<<`
|
400 |
+
|
401 |
+
## Uses
|
402 |
+
|
403 |
+
This model can be used for translation and text-to-text generation.
|
404 |
+
|
405 |
+
## Risks, Limitations and Biases
|
406 |
+
|
407 |
+
**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.**
|
408 |
+
|
409 |
+
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)).
|
410 |
+
|
411 |
+
## How to Get Started With the Model
|
412 |
+
|
413 |
+
A short example code:
|
414 |
+
|
415 |
+
```python
|
416 |
+
from transformers import MarianMTModel, MarianTokenizer
|
417 |
+
|
418 |
+
src_text = [
|
419 |
+
">>fao<< Jeg er bange for kakerlakker.",
|
420 |
+
">>nob<< Vladivostok är en stad i Ryssland."
|
421 |
+
]
|
422 |
+
|
423 |
+
model_name = "pytorch-models/opus-mt-tc-big-gmq-gmq"
|
424 |
+
tokenizer = MarianTokenizer.from_pretrained(model_name)
|
425 |
+
model = MarianMTModel.from_pretrained(model_name)
|
426 |
+
translated = model.generate(**tokenizer(src_text, return_tensors="pt", padding=True))
|
427 |
+
|
428 |
+
for t in translated:
|
429 |
+
print( tokenizer.decode(t, skip_special_tokens=True) )
|
430 |
+
|
431 |
+
# expected output:
|
432 |
+
# Tað eru uml.
|
433 |
+
# Vladivostok er en by i Russland.
|
434 |
+
```
|
435 |
+
|
436 |
+
You can also use OPUS-MT models with the transformers pipelines, for example:
|
437 |
+
|
438 |
+
```python
|
439 |
+
from transformers import pipeline
|
440 |
+
pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-tc-big-gmq-gmq")
|
441 |
+
print(pipe(">>fao<< Jeg er bange for kakerlakker."))
|
442 |
+
|
443 |
+
# expected output: Tað eru uml.
|
444 |
+
```
|
445 |
+
|
446 |
+
## Training
|
447 |
+
|
448 |
+
- **Data**: opusTCv20210807 ([source](https://github.com/Helsinki-NLP/Tatoeba-Challenge))
|
449 |
+
- **Pre-processing**: SentencePiece (spm32k,spm32k)
|
450 |
+
- **Model Type:** transformer-big
|
451 |
+
- **Original MarianNMT Model**: [opusTCv20210807_transformer-big_2022-07-29.zip](https://object.pouta.csc.fi/Tatoeba-MT-models/gmq-gmq/opusTCv20210807_transformer-big_2022-07-29.zip)
|
452 |
+
- **Training Scripts**: [GitHub Repo](https://github.com/Helsinki-NLP/OPUS-MT-train)
|
453 |
+
|
454 |
+
## Evaluation
|
455 |
+
|
456 |
+
* test set translations: [opusTCv20210807_transformer-big_2022-07-29.test.txt](https://object.pouta.csc.fi/Tatoeba-MT-models/gmq-gmq/opusTCv20210807_transformer-big_2022-07-29.test.txt)
|
457 |
+
* test set scores: [opusTCv20210807_transformer-big_2022-07-29.eval.txt](https://object.pouta.csc.fi/Tatoeba-MT-models/gmq-gmq/opusTCv20210807_transformer-big_2022-07-29.eval.txt)
|
458 |
+
* benchmark results: [benchmark_results.txt](benchmark_results.txt)
|
459 |
+
* benchmark output: [benchmark_translations.zip](benchmark_translations.zip)
|
460 |
+
|
461 |
+
| langpair | testset | chr-F | BLEU | #sent | #words |
|
462 |
+
|----------|---------|-------|-------|-------|--------|
|
463 |
+
| dan-nob | tatoeba-test-v2021-08-07 | 0.87556 | 78.2 | 1299 | 9620 |
|
464 |
+
| dan-swe | tatoeba-test-v2021-08-07 | 0.83556 | 72.5 | 1549 | 10060 |
|
465 |
+
| nno-nob | tatoeba-test-v2021-08-07 | 0.88349 | 78.9 | 467 | 3129 |
|
466 |
+
| nob-dan | tatoeba-test-v2021-08-07 | 0.85345 | 73.9 | 1299 | 9794 |
|
467 |
+
| nob-nno | tatoeba-test-v2021-08-07 | 0.74571 | 55.2 | 466 | 3141 |
|
468 |
+
| nob-swe | tatoeba-test-v2021-08-07 | 0.84747 | 73.9 | 563 | 3698 |
|
469 |
+
| swe-dan | tatoeba-test-v2021-08-07 | 0.83392 | 72.6 | 1549 | 10239 |
|
470 |
+
| swe-nob | tatoeba-test-v2021-08-07 | 0.85815 | 76.3 | 563 | 3708 |
|
471 |
+
| isl-swe | europeana2021 | 0.45562 | 22.2 | 563 | 10293 |
|
472 |
+
| nob-isl | europeana2021 | 0.54171 | 29.7 | 538 | 9932 |
|
473 |
+
| nob-swe | europeana2021 | 0.73891 | 54.0 | 538 | 9885 |
|
474 |
+
| dan-isl | flores101-devtest | 0.50227 | 22.2 | 1012 | 22834 |
|
475 |
+
| dan-nob | flores101-devtest | 0.58445 | 28.6 | 1012 | 23873 |
|
476 |
+
| dan-swe | flores101-devtest | 0.65000 | 38.5 | 1012 | 23121 |
|
477 |
+
| isl-dan | flores101-devtest | 0.53630 | 27.2 | 1012 | 24638 |
|
478 |
+
| isl-nob | flores101-devtest | 0.49434 | 20.5 | 1012 | 23873 |
|
479 |
+
| isl-swe | flores101-devtest | 0.53373 | 26.0 | 1012 | 23121 |
|
480 |
+
| nob-dan | flores101-devtest | 0.59657 | 31.7 | 1012 | 24638 |
|
481 |
+
| nob-isl | flores101-devtest | 0.47432 | 18.9 | 1012 | 22834 |
|
482 |
+
| nob-swe | flores101-devtest | 0.60030 | 31.3 | 1012 | 23121 |
|
483 |
+
| swe-dan | flores101-devtest | 0.64340 | 39.0 | 1012 | 24638 |
|
484 |
+
| swe-isl | flores101-devtest | 0.49590 | 21.7 | 1012 | 22834 |
|
485 |
+
| swe-nob | flores101-devtest | 0.58336 | 28.9 | 1012 | 23873 |
|
486 |
+
|
487 |
+
## Citation Information
|
488 |
+
|
489 |
+
* 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.)
|
490 |
+
|
491 |
+
```
|
492 |
+
@inproceedings{tiedemann-thottingal-2020-opus,
|
493 |
+
title = "{OPUS}-{MT} {--} Building open translation services for the World",
|
494 |
+
author = {Tiedemann, J{\"o}rg and Thottingal, Santhosh},
|
495 |
+
booktitle = "Proceedings of the 22nd Annual Conference of the European Association for Machine Translation",
|
496 |
+
month = nov,
|
497 |
+
year = "2020",
|
498 |
+
address = "Lisboa, Portugal",
|
499 |
+
publisher = "European Association for Machine Translation",
|
500 |
+
url = "https://aclanthology.org/2020.eamt-1.61",
|
501 |
+
pages = "479--480",
|
502 |
+
}
|
503 |
+
|
504 |
+
@inproceedings{tiedemann-2020-tatoeba,
|
505 |
+
title = "The Tatoeba Translation Challenge {--} Realistic Data Sets for Low Resource and Multilingual {MT}",
|
506 |
+
author = {Tiedemann, J{\"o}rg},
|
507 |
+
booktitle = "Proceedings of the Fifth Conference on Machine Translation",
|
508 |
+
month = nov,
|
509 |
+
year = "2020",
|
510 |
+
address = "Online",
|
511 |
+
publisher = "Association for Computational Linguistics",
|
512 |
+
url = "https://aclanthology.org/2020.wmt-1.139",
|
513 |
+
pages = "1174--1182",
|
514 |
+
}
|
515 |
+
```
|
516 |
+
|
517 |
+
## Acknowledgements
|
518 |
+
|
519 |
+
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.
|
520 |
+
|
521 |
+
## Model conversion info
|
522 |
+
|
523 |
+
* transformers version: 4.16.2
|
524 |
+
* OPUS-MT git hash: 8b9f0b0
|
525 |
+
* port time: Fri Aug 12 13:30:22 EEST 2022
|
526 |
+
* port machine: LM0-400-22516.local
|
benchmark_results.txt
ADDED
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
isl-swe europeana2021 0.45562 22.2 563 10293
|
2 |
+
nob-isl europeana2021 0.54171 29.7 538 9932
|
3 |
+
nob-swe europeana2021 0.73891 54.0 538 9885
|
4 |
+
dan-isl flores101-dev 0.50526 22.6 997 21857
|
5 |
+
dan-nob flores101-dev 0.58258 29.0 997 23157
|
6 |
+
dan-swe flores101-dev 0.64830 38.5 997 22417
|
7 |
+
isl-dan flores101-dev 0.53960 27.5 997 23685
|
8 |
+
isl-nob flores101-dev 0.49502 20.9 997 23157
|
9 |
+
isl-swe flores101-dev 0.53173 26.1 997 22417
|
10 |
+
nob-dan flores101-dev 0.59954 32.4 997 23685
|
11 |
+
nob-isl flores101-dev 0.48147 19.7 997 21857
|
12 |
+
nob-swe flores101-dev 0.59719 30.6 997 22417
|
13 |
+
swe-dan flores101-dev 0.64216 38.8 997 23685
|
14 |
+
swe-isl flores101-dev 0.49997 22.4 997 21857
|
15 |
+
swe-nob flores101-dev 0.57853 28.7 997 23157
|
16 |
+
dan-isl flores101-devtest 0.50227 22.2 1012 22834
|
17 |
+
dan-nob flores101-devtest 0.58445 28.6 1012 23873
|
18 |
+
dan-swe flores101-devtest 0.65000 38.5 1012 23121
|
19 |
+
isl-dan flores101-devtest 0.53630 27.2 1012 24638
|
20 |
+
isl-nob flores101-devtest 0.49434 20.5 1012 23873
|
21 |
+
isl-swe flores101-devtest 0.53373 26.0 1012 23121
|
22 |
+
nob-dan flores101-devtest 0.59657 31.7 1012 24638
|
23 |
+
nob-isl flores101-devtest 0.47432 18.9 1012 22834
|
24 |
+
nob-swe flores101-devtest 0.60030 31.3 1012 23121
|
25 |
+
swe-dan flores101-devtest 0.64340 39.0 1012 24638
|
26 |
+
swe-isl flores101-devtest 0.49590 21.7 1012 22834
|
27 |
+
swe-nob flores101-devtest 0.58336 28.9 1012 23873
|
28 |
+
dan-swe tatoeba-test-v2020-07-28 0.83565 72.6 1550 10082
|
29 |
+
nno-nob tatoeba-test-v2020-07-28 0.88417 79.1 474 3167
|
30 |
+
nob-nno tatoeba-test-v2020-07-28 0.74642 55.3 474 3184
|
31 |
+
nob-swe tatoeba-test-v2020-07-28 0.85123 74.5 560 3661
|
32 |
+
swe-dan tatoeba-test-v2020-07-28 0.83413 72.7 1550 10261
|
33 |
+
swe-nob tatoeba-test-v2020-07-28 0.86040 76.7 560 3672
|
34 |
+
dan-swe tatoeba-test-v2021-03-30 0.83565 72.6 1550 10082
|
35 |
+
nno-nob tatoeba-test-v2021-03-30 0.88607 79.2 500 3314
|
36 |
+
nob-nno tatoeba-test-v2021-03-30 0.74623 55.3 488 3321
|
37 |
+
nob-swe tatoeba-test-v2021-03-30 0.84518 73.4 570 3756
|
38 |
+
swe-dan tatoeba-test-v2021-03-30 0.83413 72.7 1550 10261
|
39 |
+
swe-nob tatoeba-test-v2021-03-30 0.85670 76.2 570 3765
|
40 |
+
dan-nob tatoeba-test-v2021-08-07 0.87556 78.2 1299 9620
|
41 |
+
dan-swe tatoeba-test-v2021-08-07 0.83556 72.5 1549 10060
|
42 |
+
nno-nob tatoeba-test-v2021-08-07 0.88349 78.9 467 3129
|
43 |
+
nob-dan tatoeba-test-v2021-08-07 0.85345 73.9 1299 9794
|
44 |
+
nob-nno tatoeba-test-v2021-08-07 0.74571 55.2 466 3141
|
45 |
+
nob-swe tatoeba-test-v2021-08-07 0.84747 73.9 563 3698
|
46 |
+
swe-dan tatoeba-test-v2021-08-07 0.83392 72.6 1549 10239
|
47 |
+
swe-nob tatoeba-test-v2021-08-07 0.85815 76.3 563 3708
|
benchmark_translations.zip
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:0dc88510fb37cbed988cbc7f0eb7314d465c88f373bd448d05b36f7f46cf0722
|
3 |
+
size 4304097
|
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 |
+
34890
|
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": 34890,
|
21 |
+
"decoder_vocab_size": 34891,
|
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": 25972,
|
28 |
+
"forced_eos_token_id": 25972,
|
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": 34890,
|
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": 34891
|
45 |
+
}
|
pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:cf5e0d09da97d87c603652440c65504c7ca9d2dfed3cbdde3a7dda32d4b5426b
|
3 |
+
size 495789763
|
source.spm
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:0c9800926dcbcc092f5d297ca2b8c3af64afc58877287cce0dd7beb94ad2feb9
|
3 |
+
size 802542
|
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 sha256:3f06e9450b1536ecf625fb6e8daa1f7ce9bde964240138b1a5d6df233c9ab3fe
|
3 |
+
size 802737
|
tokenizer_config.json
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"source_lang": "gmq", "target_lang": "gmq", "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_2022-07-29/gmq-gmq", "tokenizer_class": "MarianTokenizer"}
|
vocab.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|