tiedeman commited on
Commit
7dab07f
1 Parent(s): 333cda1

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 +1,168 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - fa
4
+ - gmq
5
+
6
+ tags:
7
+ - translation
8
+ - opus-mt-tc
9
+
10
+ license: cc-by-4.0
11
+ model-index:
12
+ - name: opus-mt-tc-big-fa-gmq
13
+ results:
14
+ - task:
15
+ name: Translation fas-dan
16
+ type: translation
17
+ args: fas-dan
18
+ dataset:
19
+ name: flores101-devtest
20
+ type: flores_101
21
+ args: fas dan devtest
22
+ metrics:
23
+ - name: BLEU
24
+ type: bleu
25
+ value: 22.7
26
+ - name: chr-F
27
+ type: chrf
28
+ value: 0.50857
29
+ ---
30
+ # opus-mt-tc-big-fa-gmq
31
+
32
+ ## Table of Contents
33
+ - [Model Details](#model-details)
34
+ - [Uses](#uses)
35
+ - [Risks, Limitations and Biases](#risks-limitations-and-biases)
36
+ - [How to Get Started With the Model](#how-to-get-started-with-the-model)
37
+ - [Training](#training)
38
+ - [Evaluation](#evaluation)
39
+ - [Citation Information](#citation-information)
40
+ - [Acknowledgements](#acknowledgements)
41
+
42
+ ## Model Details
43
+
44
+ Neural machine translation model for translating from Persian (fa) to North Germanic languages (gmq).
45
+
46
+ 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).
47
+ **Model Description:**
48
+ - **Developed by:** Language Technology Research Group at the University of Helsinki
49
+ - **Model Type:** Translation (transformer-big)
50
+ - **Release**: 2022-07-23
51
+ - **License:** CC-BY-4.0
52
+ - **Language(s):**
53
+ - Source Language(s):
54
+ - Target Language(s):
55
+ - Valid Target Language Labels:
56
+ - **Original Model**: [opusTCv20210807_transformer-big_2022-07-23.zip](https://object.pouta.csc.fi/Tatoeba-MT-models/fas-gmq/opusTCv20210807_transformer-big_2022-07-23.zip)
57
+ - **Resources for more information:**
58
+ - [OPUS-MT-train GitHub Repo](https://github.com/Helsinki-NLP/OPUS-MT-train)
59
+ - More information about released models for this language pair: [OPUS-MT fas-gmq README](https://github.com/Helsinki-NLP/Tatoeba-Challenge/tree/master/models/fas-gmq/README.md)
60
+ - [More information about MarianNMT models in the transformers library](https://huggingface.co/docs/transformers/model_doc/marian)
61
+ - [Tatoeba Translation Challenge](https://github.com/Helsinki-NLP/Tatoeba-Challenge/
62
+
63
+ 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. `>><<`
64
+
65
+ ## Uses
66
+
67
+ This model can be used for translation and text-to-text generation.
68
+
69
+ ## Risks, Limitations and Biases
70
+
71
+ **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.**
72
+
73
+ 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)).
74
+
75
+ ## How to Get Started With the Model
76
+
77
+ A short example code:
78
+
79
+ ```python
80
+ from transformers import MarianMTModel, MarianTokenizer
81
+
82
+ src_text = [
83
+ "از سوسک می‌ترسم.",
84
+ "از سوسک می‌ترسم."
85
+ ]
86
+
87
+ model_name = "pytorch-models/opus-mt-tc-big-fa-gmq"
88
+ tokenizer = MarianTokenizer.from_pretrained(model_name)
89
+ model = MarianMTModel.from_pretrained(model_name)
90
+ translated = model.generate(**tokenizer(src_text, return_tensors="pt", padding=True))
91
+
92
+ for t in translated:
93
+ print( tokenizer.decode(t, skip_special_tokens=True) )
94
+
95
+ # expected output:
96
+ # Jeg er bange for kakerlakker.
97
+ # Jeg er bange for kakerlakker.
98
+ ```
99
+
100
+ You can also use OPUS-MT models with the transformers pipelines, for example:
101
+
102
+ ```python
103
+ from transformers import pipeline
104
+ pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-tc-big-fa-gmq")
105
+ print(pipe("از سوسک می‌ترسم."))
106
+
107
+ # expected output: Jeg er bange for kakerlakker.
108
+ ```
109
+
110
+ ## Training
111
+
112
+ - **Data**: opusTCv20210807 ([source](https://github.com/Helsinki-NLP/Tatoeba-Challenge))
113
+ - **Pre-processing**: SentencePiece (spm32k,spm32k)
114
+ - **Model Type:** transformer-big
115
+ - **Original MarianNMT Model**: [opusTCv20210807_transformer-big_2022-07-23.zip](https://object.pouta.csc.fi/Tatoeba-MT-models/fas-gmq/opusTCv20210807_transformer-big_2022-07-23.zip)
116
+ - **Training Scripts**: [GitHub Repo](https://github.com/Helsinki-NLP/OPUS-MT-train)
117
+
118
+ ## Evaluation
119
+
120
+ * test set translations: [opusTCv20210807_transformer-big_2022-07-23.test.txt](https://object.pouta.csc.fi/Tatoeba-MT-models/fas-gmq/opusTCv20210807_transformer-big_2022-07-23.test.txt)
121
+ * test set scores: [opusTCv20210807_transformer-big_2022-07-23.eval.txt](https://object.pouta.csc.fi/Tatoeba-MT-models/fas-gmq/opusTCv20210807_transformer-big_2022-07-23.eval.txt)
122
+ * benchmark results: [benchmark_results.txt](benchmark_results.txt)
123
+ * benchmark output: [benchmark_translations.zip](benchmark_translations.zip)
124
+
125
+ | langpair | testset | chr-F | BLEU | #sent | #words |
126
+ |----------|---------|-------|-------|-------|--------|
127
+ | fas-dan | flores101-devtest | 0.50857 | 22.7 | 1012 | 24638 |
128
+
129
+ ## Citation Information
130
+
131
+ * 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.)
132
+
133
+ ```
134
+ @inproceedings{tiedemann-thottingal-2020-opus,
135
+ title = "{OPUS}-{MT} {--} Building open translation services for the World",
136
+ author = {Tiedemann, J{\"o}rg and Thottingal, Santhosh},
137
+ booktitle = "Proceedings of the 22nd Annual Conference of the European Association for Machine Translation",
138
+ month = nov,
139
+ year = "2020",
140
+ address = "Lisboa, Portugal",
141
+ publisher = "European Association for Machine Translation",
142
+ url = "https://aclanthology.org/2020.eamt-1.61",
143
+ pages = "479--480",
144
+ }
145
+
146
+ @inproceedings{tiedemann-2020-tatoeba,
147
+ title = "The Tatoeba Translation Challenge {--} Realistic Data Sets for Low Resource and Multilingual {MT}",
148
+ author = {Tiedemann, J{\"o}rg},
149
+ booktitle = "Proceedings of the Fifth Conference on Machine Translation",
150
+ month = nov,
151
+ year = "2020",
152
+ address = "Online",
153
+ publisher = "Association for Computational Linguistics",
154
+ url = "https://aclanthology.org/2020.wmt-1.139",
155
+ pages = "1174--1182",
156
+ }
157
+ ```
158
+
159
+ ## Acknowledgements
160
+
161
+ 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.
162
+
163
+ ## Model conversion info
164
+
165
+ * transformers version: 4.16.2
166
+ * OPUS-MT git hash: 8b9f0b0
167
+ * port time: Fri Aug 12 20:17:09 EEST 2022
168
+ * port machine: LM0-400-22516.local
benchmark_results.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ fas-dan flores101-dev 0.51347 23.7 997 23685
2
+ fas-dan flores101-devtest 0.50857 22.7 1012 24638
benchmark_translations.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2bc104b48e740b373c27b4ee2b47be5b3d6d4cc3d1109e418cc1732b8004a782
3
+ size 344829
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
+ 58753
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": 58753,
21
+ "decoder_vocab_size": 58754,
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": 25124,
28
+ "forced_eos_token_id": 25124,
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": 58753,
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": 58754
45
+ }
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e1e5a49c99b6fa44d3375cd09e3e2c259e3cb587ab0da312ef87cd6318343ff6
3
+ size 593580355
source.spm ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b9e065549fde9a9a4b89b2a19c0a66a9a6b4b6292b43a25c1bd85d27080e0811
3
+ size 872053
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:a96c5c0e2bf4d22ff0a6b5f3ff9f30c0d4861a70c2364bac56adc05e451bcb75
3
+ size 807593
tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"source_lang": "fa", "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_2022-07-23/fa-gmq", "tokenizer_class": "MarianTokenizer"}
vocab.json ADDED
The diff for this file is too large to render. See raw diff