tiedeman commited on
Commit
3754c6e
โ€ข
1 Parent(s): c95f6d9

Initial commit

Browse files
.gitattributes CHANGED
@@ -26,3 +26,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
26
  *.zip filter=lfs diff=lfs merge=lfs -text
27
  *.zstandard filter=lfs diff=lfs merge=lfs -text
28
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
26
  *.zip filter=lfs diff=lfs merge=lfs -text
27
  *.zstandard filter=lfs diff=lfs merge=lfs -text
28
  *tfevents* filter=lfs diff=lfs merge=lfs -text
29
+ *.spm filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,156 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - ar
4
+ - en
5
+
6
+ tags:
7
+ - translation
8
+
9
+ license: cc-by-4.0
10
+ model-index:
11
+ - name: opus-mt-tc-big-en-ar
12
+ results:
13
+ - task:
14
+ name: Translation eng-ara
15
+ type: translation
16
+ args: eng-ara
17
+ dataset:
18
+ name: flores101-devtest
19
+ type: flores_101
20
+ args: eng ara devtest
21
+ metrics:
22
+ - name: BLEU
23
+ type: bleu
24
+ value: 29.4
25
+ - task:
26
+ name: Translation eng-ara
27
+ type: translation
28
+ args: eng-ara
29
+ dataset:
30
+ name: tatoeba-test-v2020-07-28
31
+ type: tatoeba_mt
32
+ args: eng-ara
33
+ metrics:
34
+ - name: BLEU
35
+ type: bleu
36
+ value: 20.0
37
+ - task:
38
+ name: Translation eng-ara
39
+ type: translation
40
+ args: eng-ara
41
+ dataset:
42
+ name: tico19-test
43
+ type: tico19-test
44
+ args: eng-ara
45
+ metrics:
46
+ - name: BLEU
47
+ type: bleu
48
+ value: 30.0
49
+ ---
50
+ # opus-mt-tc-big-en-ar
51
+
52
+ Neural machine translation model for translating from English (en) to Arabic (ar).
53
+
54
+ 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).
55
+
56
+ * Publications: [OPUS-MT โ€“ Building open translation services for the World](https://aclanthology.org/2020.eamt-1.61/) and [The Tatoeba Translation Challenge โ€“ Realistic Data Sets for Low Resource and Multilingual MT](https://aclanthology.org/2020.wmt-1.139/) (Please, cite if you use this model.)
57
+
58
+ ```
59
+ @inproceedings{tiedemann-thottingal-2020-opus,
60
+ title = "{OPUS}-{MT} {--} Building open translation services for the World",
61
+ author = {Tiedemann, J{\"o}rg and Thottingal, Santhosh},
62
+ booktitle = "Proceedings of the 22nd Annual Conference of the European Association for Machine Translation",
63
+ month = nov,
64
+ year = "2020",
65
+ address = "Lisboa, Portugal",
66
+ publisher = "European Association for Machine Translation",
67
+ url = "https://aclanthology.org/2020.eamt-1.61",
68
+ pages = "479--480",
69
+ }
70
+
71
+ @inproceedings{tiedemann-2020-tatoeba,
72
+ title = "The Tatoeba Translation Challenge {--} Realistic Data Sets for Low Resource and Multilingual {MT}",
73
+ author = {Tiedemann, J{\"o}rg},
74
+ booktitle = "Proceedings of the Fifth Conference on Machine Translation",
75
+ month = nov,
76
+ year = "2020",
77
+ address = "Online",
78
+ publisher = "Association for Computational Linguistics",
79
+ url = "https://aclanthology.org/2020.wmt-1.139",
80
+ pages = "1174--1182",
81
+ }
82
+ ```
83
+
84
+ ## Model info
85
+
86
+ * Release: 2022-02-25
87
+ * source language(s): eng
88
+ * target language(s): afb ara
89
+ * valid target language labels: >>afb<< >>ara<<
90
+ * model: transformer-big
91
+ * data: opusTCv20210807+bt ([source](https://github.com/Helsinki-NLP/Tatoeba-Challenge))
92
+ * tokenization: SentencePiece (spm32k,spm32k)
93
+ * original model: [opusTCv20210807+bt_transformer-big_2022-02-25.zip](https://object.pouta.csc.fi/Tatoeba-MT-models/eng-ara/opusTCv20210807+bt_transformer-big_2022-02-25.zip)
94
+ * more information released models: [OPUS-MT eng-ara README](https://github.com/Helsinki-NLP/Tatoeba-Challenge/tree/master/models/eng-ara/README.md)
95
+ * more information about the model: [MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)
96
+
97
+ 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. `>>afb<<`
98
+
99
+ ## Usage
100
+
101
+ A short example code:
102
+
103
+ ```python
104
+ from transformers import MarianMTModel, MarianTokenizer
105
+
106
+ src_text = [
107
+ ">>ara<< I can't help you because I'm busy.",
108
+ ">>ara<< I have to write a letter. Do you have some paper?"
109
+ ]
110
+
111
+ model_name = "pytorch-models/opus-mt-tc-big-en-ar"
112
+ tokenizer = MarianTokenizer.from_pretrained(model_name)
113
+ model = MarianMTModel.from_pretrained(model_name)
114
+ translated = model.generate(**tokenizer(src_text, return_tensors="pt", padding=True))
115
+
116
+ for t in translated:
117
+ print( tokenizer.decode(t, skip_special_tokens=True) )
118
+
119
+ # expected output:
120
+ # ู„ุง ุฃุณุชุทูŠุน ู…ุณุงุนุฏุชูƒ ู„ุฃู†ู†ูŠ ู…ุดุบูˆู„.
121
+ # ูŠุฌุจ ุฃู† ุฃูƒุชุจ ุฑุณุงู„ุฉ ู‡ู„ ู„ุฏูŠูƒ ุจุนุถ ุงู„ุฃูˆุฑุงู‚ุŸ
122
+ ```
123
+
124
+ You can also use OPUS-MT models with the transformers pipelines, for example:
125
+
126
+ ```python
127
+ from transformers import pipeline
128
+ pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-tc-big-en-ar")
129
+ print(pipe(">>ara<< I can't help you because I'm busy."))
130
+
131
+ # expected output: ู„ุง ุฃุณุชุทูŠุน ู…ุณุงุนุฏุชูƒ ู„ุฃู†ู†ูŠ ู…ุดุบูˆู„.
132
+ ```
133
+
134
+ ## Benchmarks
135
+
136
+ * test set translations: [opusTCv20210807+bt_transformer-big_2022-02-25.test.txt](https://object.pouta.csc.fi/Tatoeba-MT-models/eng-ara/opusTCv20210807+bt_transformer-big_2022-02-25.test.txt)
137
+ * test set scores: [opusTCv20210807+bt_transformer-big_2022-02-25.eval.txt](https://object.pouta.csc.fi/Tatoeba-MT-models/eng-ara/opusTCv20210807+bt_transformer-big_2022-02-25.eval.txt)
138
+ * benchmark results: [benchmark_results.txt](benchmark_results.txt)
139
+ * benchmark output: [benchmark_translations.zip](benchmark_translations.zip)
140
+
141
+ | langpair | testset | chr-F | BLEU | #sent | #words |
142
+ |----------|---------|-------|-------|-------|--------|
143
+ | eng-ara | tatoeba-test-v2021-08-07 | 0.48813 | 19.8 | 10305 | 61356 |
144
+ | eng-ara | flores101-devtest | 0.61154 | 29.4 | 1012 | 21357 |
145
+ | eng-ara | tico19-test | 0.60075 | 30.0 | 2100 | 51339 |
146
+
147
+ ## Acknowledgements
148
+
149
+ 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 Unionโ€™s Horizon 2020 research and innovation programme (grant agreement No 771113), and the [MeMAD project](https://memad.eu/), funded by the European Unionโ€™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.
150
+
151
+ ## Model conversion info
152
+
153
+ * transformers version: 4.16.2
154
+ * OPUS-MT git hash: 3405783
155
+ * port time: Wed Apr 13 16:37:31 EEST 2022
156
+ * port machine: LM0-400-22516.local
benchmark_results.txt ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ eng-ara flores101-dev 0.60808 29.9 997 20512
2
+ eng-ara flores101-devtest 0.61154 29.4 1012 21357
3
+ eng-ara tatoeba-test-v2020-07-28 0.48938 20.0 10000 58935
4
+ eng-arq tatoeba-test-v2020-07-28 0.16786 0.9 403 2272
5
+ eng-ara tatoeba-test-v2021-03-30 0.48790 19.8 10267 61124
6
+ eng-arq tatoeba-test-v2021-03-30 0.16797 0.9 405 2285
7
+ eng-ara tatoeba-test-v2021-08-07 0.48813 19.8 10305 61356
8
+ eng-arq tatoeba-test-v2021-08-07 0.16797 0.9 405 2285
9
+ eng-ara tico19-test 0.60075 30.0 2100 51339
benchmark_translations.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1285375ccee419aa268dec1b666e164c869e7c59dfe264e0bab4d684777c0dae
3
+ size 2110100
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
+ 61246
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": 61246,
21
+ "decoder_vocab_size": 61247,
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": 25897,
28
+ "forced_eos_token_id": 25897,
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": 61246,
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": 61247
45
+ }
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ac090df098d2b66694744ff7d4e1b24288a603aa340730d9b454687053661488
3
+ size 603796611
source.spm ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0a9b220e324e29d4fbab530747ab82968a79d5408f1b3210f0f6d812d148b7d2
3
+ size 806127
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:db09c46631638384f2b13ccdececb1c88103f0e59c14d5def247975007e7eaac
3
+ size 916373
tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"source_lang": "en", "target_lang": "ar", "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+bt_transformer-big_2022-02-25/en-ar", "tokenizer_class": "MarianTokenizer"}
vocab.json ADDED
The diff for this file is too large to render. See raw diff