tiedeman commited on
Commit
51c5da4
1 Parent(s): 86d9428

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 +1,234 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - da
4
+ - gmq
5
+ - nb
6
+ - nn
7
+ - no
8
+ - sv
9
+ - tr
10
+
11
+ tags:
12
+ - translation
13
+ - opus-mt-tc
14
+
15
+ license: cc-by-4.0
16
+ model-index:
17
+ - name: opus-mt-tc-big-gmq-tr
18
+ results:
19
+ - task:
20
+ name: Translation dan-tur
21
+ type: translation
22
+ args: dan-tur
23
+ dataset:
24
+ name: flores101-devtest
25
+ type: flores_101
26
+ args: dan tur devtest
27
+ metrics:
28
+ - name: BLEU
29
+ type: bleu
30
+ value: 23.4
31
+ - name: chr-F
32
+ type: chrf
33
+ value: 0.56363
34
+ - task:
35
+ name: Translation nob-tur
36
+ type: translation
37
+ args: nob-tur
38
+ dataset:
39
+ name: flores101-devtest
40
+ type: flores_101
41
+ args: nob tur devtest
42
+ metrics:
43
+ - name: BLEU
44
+ type: bleu
45
+ value: 19.0
46
+ - name: chr-F
47
+ type: chrf
48
+ value: 0.52696
49
+ - task:
50
+ name: Translation swe-tur
51
+ type: translation
52
+ args: swe-tur
53
+ dataset:
54
+ name: flores101-devtest
55
+ type: flores_101
56
+ args: swe tur devtest
57
+ metrics:
58
+ - name: BLEU
59
+ type: bleu
60
+ value: 22.2
61
+ - name: chr-F
62
+ type: chrf
63
+ value: 0.54996
64
+ - task:
65
+ name: Translation dan-tur
66
+ type: translation
67
+ args: dan-tur
68
+ dataset:
69
+ name: tatoeba-test-v2021-08-07
70
+ type: tatoeba_mt
71
+ args: dan-tur
72
+ metrics:
73
+ - name: BLEU
74
+ type: bleu
75
+ value: 43.0
76
+ - name: chr-F
77
+ type: chrf
78
+ value: 0.67830
79
+ - task:
80
+ name: Translation swe-tur
81
+ type: translation
82
+ args: swe-tur
83
+ dataset:
84
+ name: tatoeba-test-v2021-08-07
85
+ type: tatoeba_mt
86
+ args: swe-tur
87
+ metrics:
88
+ - name: BLEU
89
+ type: bleu
90
+ value: 34.2
91
+ - name: chr-F
92
+ type: chrf
93
+ value: 0.63653
94
+ ---
95
+ # opus-mt-tc-big-gmq-tr
96
+
97
+ ## Table of Contents
98
+ - [Model Details](#model-details)
99
+ - [Uses](#uses)
100
+ - [Risks, Limitations and Biases](#risks-limitations-and-biases)
101
+ - [How to Get Started With the Model](#how-to-get-started-with-the-model)
102
+ - [Training](#training)
103
+ - [Evaluation](#evaluation)
104
+ - [Citation Information](#citation-information)
105
+ - [Acknowledgements](#acknowledgements)
106
+
107
+ ## Model Details
108
+
109
+ Neural machine translation model for translating from North Germanic languages (gmq) to Turkish (tr).
110
+
111
+ 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).
112
+ **Model Description:**
113
+ - **Developed by:** Language Technology Research Group at the University of Helsinki
114
+ - **Model Type:** Translation (transformer-big)
115
+ - **Release**: 2022-07-26
116
+ - **License:** CC-BY-4.0
117
+ - **Language(s):**
118
+ - Source Language(s): dan nno nob nor swe
119
+ - Target Language(s): tur
120
+ - **Original Model**: [opusTCv20210807_transformer-big_2022-07-26.zip](https://object.pouta.csc.fi/Tatoeba-MT-models/gmq-tur/opusTCv20210807_transformer-big_2022-07-26.zip)
121
+ - **Resources for more information:**
122
+ - [OPUS-MT-train GitHub Repo](https://github.com/Helsinki-NLP/OPUS-MT-train)
123
+ - More information about released models for this language pair: [OPUS-MT gmq-tur README](https://github.com/Helsinki-NLP/Tatoeba-Challenge/tree/master/models/gmq-tur/README.md)
124
+ - [More information about MarianNMT models in the transformers library](https://huggingface.co/docs/transformers/model_doc/marian)
125
+ - [Tatoeba Translation Challenge](https://github.com/Helsinki-NLP/Tatoeba-Challenge/
126
+
127
+ ## Uses
128
+
129
+ This model can be used for translation and text-to-text generation.
130
+
131
+ ## Risks, Limitations and Biases
132
+
133
+ **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.**
134
+
135
+ 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)).
136
+
137
+ ## How to Get Started With the Model
138
+
139
+ A short example code:
140
+
141
+ ```python
142
+ from transformers import MarianMTModel, MarianTokenizer
143
+
144
+ src_text = [
145
+ "Aftensmaden dufter lækkert.",
146
+ "Vi ser vad som händer tillsammans."
147
+ ]
148
+
149
+ model_name = "pytorch-models/opus-mt-tc-big-gmq-tr"
150
+ tokenizer = MarianTokenizer.from_pretrained(model_name)
151
+ model = MarianMTModel.from_pretrained(model_name)
152
+ translated = model.generate(**tokenizer(src_text, return_tensors="pt", padding=True))
153
+
154
+ for t in translated:
155
+ print( tokenizer.decode(t, skip_special_tokens=True) )
156
+
157
+ # expected output:
158
+ # Akşam yemeği çok güzel kokuyor.
159
+ # Birlikte bakalım neler olacak.
160
+ ```
161
+
162
+ You can also use OPUS-MT models with the transformers pipelines, for example:
163
+
164
+ ```python
165
+ from transformers import pipeline
166
+ pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-tc-big-gmq-tr")
167
+ print(pipe("Aftensmaden dufter lækkert."))
168
+
169
+ # expected output: Akşam yemeği çok güzel kokuyor.
170
+ ```
171
+
172
+ ## Training
173
+
174
+ - **Data**: opusTCv20210807 ([source](https://github.com/Helsinki-NLP/Tatoeba-Challenge))
175
+ - **Pre-processing**: SentencePiece (spm32k,spm32k)
176
+ - **Model Type:** transformer-big
177
+ - **Original MarianNMT Model**: [opusTCv20210807_transformer-big_2022-07-26.zip](https://object.pouta.csc.fi/Tatoeba-MT-models/gmq-tur/opusTCv20210807_transformer-big_2022-07-26.zip)
178
+ - **Training Scripts**: [GitHub Repo](https://github.com/Helsinki-NLP/OPUS-MT-train)
179
+
180
+ ## Evaluation
181
+
182
+ * test set translations: [opusTCv20210807_transformer-big_2022-07-26.test.txt](https://object.pouta.csc.fi/Tatoeba-MT-models/gmq-tur/opusTCv20210807_transformer-big_2022-07-26.test.txt)
183
+ * test set scores: [opusTCv20210807_transformer-big_2022-07-26.eval.txt](https://object.pouta.csc.fi/Tatoeba-MT-models/gmq-tur/opusTCv20210807_transformer-big_2022-07-26.eval.txt)
184
+ * benchmark results: [benchmark_results.txt](benchmark_results.txt)
185
+ * benchmark output: [benchmark_translations.zip](benchmark_translations.zip)
186
+
187
+ | langpair | testset | chr-F | BLEU | #sent | #words |
188
+ |----------|---------|-------|-------|-------|--------|
189
+ | dan-tur | tatoeba-test-v2021-08-07 | 0.67830 | 43.0 | 758 | 3436 |
190
+ | swe-tur | tatoeba-test-v2021-08-07 | 0.63653 | 34.2 | 203 | 1008 |
191
+ | dan-tur | flores101-devtest | 0.56363 | 23.4 | 1012 | 20253 |
192
+ | nob-tur | flores101-devtest | 0.52696 | 19.0 | 1012 | 20253 |
193
+ | swe-tur | flores101-devtest | 0.54996 | 22.2 | 1012 | 20253 |
194
+
195
+ ## Citation Information
196
+
197
+ * 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.)
198
+
199
+ ```
200
+ @inproceedings{tiedemann-thottingal-2020-opus,
201
+ title = "{OPUS}-{MT} {--} Building open translation services for the World",
202
+ author = {Tiedemann, J{\"o}rg and Thottingal, Santhosh},
203
+ booktitle = "Proceedings of the 22nd Annual Conference of the European Association for Machine Translation",
204
+ month = nov,
205
+ year = "2020",
206
+ address = "Lisboa, Portugal",
207
+ publisher = "European Association for Machine Translation",
208
+ url = "https://aclanthology.org/2020.eamt-1.61",
209
+ pages = "479--480",
210
+ }
211
+
212
+ @inproceedings{tiedemann-2020-tatoeba,
213
+ title = "The Tatoeba Translation Challenge {--} Realistic Data Sets for Low Resource and Multilingual {MT}",
214
+ author = {Tiedemann, J{\"o}rg},
215
+ booktitle = "Proceedings of the Fifth Conference on Machine Translation",
216
+ month = nov,
217
+ year = "2020",
218
+ address = "Online",
219
+ publisher = "Association for Computational Linguistics",
220
+ url = "https://aclanthology.org/2020.wmt-1.139",
221
+ pages = "1174--1182",
222
+ }
223
+ ```
224
+
225
+ ## Acknowledgements
226
+
227
+ 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.
228
+
229
+ ## Model conversion info
230
+
231
+ * transformers version: 4.16.2
232
+ * OPUS-MT git hash: 8b9f0b0
233
+ * port time: Fri Aug 12 18:05:57 EEST 2022
234
+ * port machine: LM0-400-22516.local
benchmark_results.txt ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ dan-tur flores101-dev 0.56062 22.4 997 19181
2
+ nob-tur flores101-dev 0.53047 18.9 997 19181
3
+ swe-tur flores101-dev 0.55638 22.5 997 19181
4
+ dan-tur flores101-devtest 0.56363 23.4 1012 20253
5
+ nob-tur flores101-devtest 0.52696 19.0 1012 20253
6
+ swe-tur flores101-devtest 0.54996 22.2 1012 20253
7
+ dan-tur tatoeba-test-v2020-07-28 0.67736 42.8 757 3429
8
+ swe-tur tatoeba-test-v2020-07-28 0.63451 33.1 201 993
9
+ dan-tur tatoeba-test-v2021-03-30 0.67803 42.9 764 3465
10
+ swe-tur tatoeba-test-v2021-03-30 0.64067 34.3 212 1063
11
+ dan-tur tatoeba-test-v2021-08-07 0.67830 43.0 758 3436
12
+ swe-tur tatoeba-test-v2021-08-07 0.63653 34.2 203 1008
benchmark_translations.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a8459b7f7ace495323001d9d2b82e3013d6213bb5fcc67457f2eaf816c7ec24d
3
+ size 1012496
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
+ 57176
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": 57176,
21
+ "decoder_vocab_size": 57177,
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": 42578,
28
+ "forced_eos_token_id": 42578,
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": 57176,
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": 57177
45
+ }
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:30846df215eb9bb680abaac71a0eb658f1c01be1d3ca3d3672ed16ffdca0a71c
3
+ size 587117763
source.spm ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3cafc555227e204d2046c12e403605d01d15681ca90fd1f5ca4226ce95b824f6
3
+ size 798335
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:d8f3c5c1adbd60846a654aa96500a154ac9a2dbfc3d7ccb6d701f198a7696cc6
3
+ size 832005
tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
1
+ {"source_lang": "gmq", "target_lang": "tr", "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_2022-07-26/gmq-tr", "tokenizer_class": "MarianTokenizer"}
vocab.json ADDED
The diff for this file is too large to render. See raw diff