wietsedv commited on
Commit
2517cba
1 Parent(s): 0db1b87

model dump

Browse files
README.md ADDED
@@ -0,0 +1,348 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ ---
3
+ language:
4
+ - ga
5
+ license: apache-2.0
6
+ library_name: transformers
7
+ tags:
8
+ - part-of-speech
9
+ - token-classification
10
+ datasets:
11
+ - universal_dependencies
12
+ metrics:
13
+ - accuracy
14
+
15
+ model-index:
16
+ - name: xlm-roberta-base-ft-udpos28-ga
17
+ results:
18
+ - task:
19
+ type: token-classification
20
+ name: Part-of-Speech Tagging
21
+ dataset:
22
+ type: universal_dependencies
23
+ name: Universal Dependencies v2.8
24
+ metrics:
25
+ - type: accuracy
26
+ name: English Test accuracy
27
+ value: 75.3
28
+ - type: accuracy
29
+ name: Dutch Test accuracy
30
+ value: 79.5
31
+ - type: accuracy
32
+ name: German Test accuracy
33
+ value: 76.2
34
+ - type: accuracy
35
+ name: Italian Test accuracy
36
+ value: 73.6
37
+ - type: accuracy
38
+ name: French Test accuracy
39
+ value: 76.4
40
+ - type: accuracy
41
+ name: Spanish Test accuracy
42
+ value: 82.4
43
+ - type: accuracy
44
+ name: Russian Test accuracy
45
+ value: 80.7
46
+ - type: accuracy
47
+ name: Swedish Test accuracy
48
+ value: 78.7
49
+ - type: accuracy
50
+ name: Norwegian Test accuracy
51
+ value: 74.5
52
+ - type: accuracy
53
+ name: Danish Test accuracy
54
+ value: 77.9
55
+ - type: accuracy
56
+ name: Low Saxon Test accuracy
57
+ value: 49.0
58
+ - type: accuracy
59
+ name: Akkadian Test accuracy
60
+ value: 46.2
61
+ - type: accuracy
62
+ name: Armenian Test accuracy
63
+ value: 71.8
64
+ - type: accuracy
65
+ name: Welsh Test accuracy
66
+ value: 77.9
67
+ - type: accuracy
68
+ name: Old East Slavic Test accuracy
69
+ value: 67.3
70
+ - type: accuracy
71
+ name: Albanian Test accuracy
72
+ value: 79.8
73
+ - type: accuracy
74
+ name: Slovenian Test accuracy
75
+ value: 65.3
76
+ - type: accuracy
77
+ name: Guajajara Test accuracy
78
+ value: 46.4
79
+ - type: accuracy
80
+ name: Kurmanji Test accuracy
81
+ value: 74.9
82
+ - type: accuracy
83
+ name: Turkish Test accuracy
84
+ value: 73.7
85
+ - type: accuracy
86
+ name: Finnish Test accuracy
87
+ value: 73.8
88
+ - type: accuracy
89
+ name: Indonesian Test accuracy
90
+ value: 78.6
91
+ - type: accuracy
92
+ name: Ukrainian Test accuracy
93
+ value: 79.9
94
+ - type: accuracy
95
+ name: Polish Test accuracy
96
+ value: 82.5
97
+ - type: accuracy
98
+ name: Portuguese Test accuracy
99
+ value: 80.6
100
+ - type: accuracy
101
+ name: Kazakh Test accuracy
102
+ value: 75.6
103
+ - type: accuracy
104
+ name: Latin Test accuracy
105
+ value: 70.0
106
+ - type: accuracy
107
+ name: Old French Test accuracy
108
+ value: 49.1
109
+ - type: accuracy
110
+ name: Buryat Test accuracy
111
+ value: 60.3
112
+ - type: accuracy
113
+ name: Kaapor Test accuracy
114
+ value: 21.2
115
+ - type: accuracy
116
+ name: Korean Test accuracy
117
+ value: 60.5
118
+ - type: accuracy
119
+ name: Estonian Test accuracy
120
+ value: 75.7
121
+ - type: accuracy
122
+ name: Croatian Test accuracy
123
+ value: 77.3
124
+ - type: accuracy
125
+ name: Gothic Test accuracy
126
+ value: 29.1
127
+ - type: accuracy
128
+ name: Swiss German Test accuracy
129
+ value: 44.3
130
+ - type: accuracy
131
+ name: Assyrian Test accuracy
132
+ value: 16.3
133
+ - type: accuracy
134
+ name: North Sami Test accuracy
135
+ value: 45.0
136
+ - type: accuracy
137
+ name: Naija Test accuracy
138
+ value: 32.0
139
+ - type: accuracy
140
+ name: Latvian Test accuracy
141
+ value: 77.7
142
+ - type: accuracy
143
+ name: Chinese Test accuracy
144
+ value: 49.6
145
+ - type: accuracy
146
+ name: Tagalog Test accuracy
147
+ value: 71.1
148
+ - type: accuracy
149
+ name: Bambara Test accuracy
150
+ value: 29.1
151
+ - type: accuracy
152
+ name: Lithuanian Test accuracy
153
+ value: 76.4
154
+ - type: accuracy
155
+ name: Galician Test accuracy
156
+ value: 80.9
157
+ - type: accuracy
158
+ name: Vietnamese Test accuracy
159
+ value: 58.6
160
+ - type: accuracy
161
+ name: Greek Test accuracy
162
+ value: 77.5
163
+ - type: accuracy
164
+ name: Catalan Test accuracy
165
+ value: 79.7
166
+ - type: accuracy
167
+ name: Czech Test accuracy
168
+ value: 78.1
169
+ - type: accuracy
170
+ name: Erzya Test accuracy
171
+ value: 52.5
172
+ - type: accuracy
173
+ name: Bhojpuri Test accuracy
174
+ value: 59.2
175
+ - type: accuracy
176
+ name: Thai Test accuracy
177
+ value: 58.7
178
+ - type: accuracy
179
+ name: Marathi Test accuracy
180
+ value: 79.1
181
+ - type: accuracy
182
+ name: Basque Test accuracy
183
+ value: 68.1
184
+ - type: accuracy
185
+ name: Slovak Test accuracy
186
+ value: 80.0
187
+ - type: accuracy
188
+ name: Kiche Test accuracy
189
+ value: 46.4
190
+ - type: accuracy
191
+ name: Yoruba Test accuracy
192
+ value: 33.1
193
+ - type: accuracy
194
+ name: Warlpiri Test accuracy
195
+ value: 40.5
196
+ - type: accuracy
197
+ name: Tamil Test accuracy
198
+ value: 78.1
199
+ - type: accuracy
200
+ name: Maltese Test accuracy
201
+ value: 36.7
202
+ - type: accuracy
203
+ name: Ancient Greek Test accuracy
204
+ value: 58.5
205
+ - type: accuracy
206
+ name: Icelandic Test accuracy
207
+ value: 71.2
208
+ - type: accuracy
209
+ name: Mbya Guarani Test accuracy
210
+ value: 34.0
211
+ - type: accuracy
212
+ name: Urdu Test accuracy
213
+ value: 65.5
214
+ - type: accuracy
215
+ name: Romanian Test accuracy
216
+ value: 76.8
217
+ - type: accuracy
218
+ name: Persian Test accuracy
219
+ value: 79.7
220
+ - type: accuracy
221
+ name: Apurina Test accuracy
222
+ value: 51.8
223
+ - type: accuracy
224
+ name: Japanese Test accuracy
225
+ value: 36.1
226
+ - type: accuracy
227
+ name: Hungarian Test accuracy
228
+ value: 77.1
229
+ - type: accuracy
230
+ name: Hindi Test accuracy
231
+ value: 69.7
232
+ - type: accuracy
233
+ name: Classical Chinese Test accuracy
234
+ value: 32.1
235
+ - type: accuracy
236
+ name: Komi Permyak Test accuracy
237
+ value: 51.1
238
+ - type: accuracy
239
+ name: Faroese Test accuracy
240
+ value: 70.6
241
+ - type: accuracy
242
+ name: Sanskrit Test accuracy
243
+ value: 35.7
244
+ - type: accuracy
245
+ name: Livvi Test accuracy
246
+ value: 60.6
247
+ - type: accuracy
248
+ name: Arabic Test accuracy
249
+ value: 83.7
250
+ - type: accuracy
251
+ name: Wolof Test accuracy
252
+ value: 40.8
253
+ - type: accuracy
254
+ name: Bulgarian Test accuracy
255
+ value: 78.7
256
+ - type: accuracy
257
+ name: Akuntsu Test accuracy
258
+ value: 43.2
259
+ - type: accuracy
260
+ name: Makurap Test accuracy
261
+ value: 19.9
262
+ - type: accuracy
263
+ name: Kangri Test accuracy
264
+ value: 46.3
265
+ - type: accuracy
266
+ name: Breton Test accuracy
267
+ value: 61.7
268
+ - type: accuracy
269
+ name: Telugu Test accuracy
270
+ value: 76.8
271
+ - type: accuracy
272
+ name: Cantonese Test accuracy
273
+ value: 49.0
274
+ - type: accuracy
275
+ name: Old Church Slavonic Test accuracy
276
+ value: 43.9
277
+ - type: accuracy
278
+ name: Karelian Test accuracy
279
+ value: 64.1
280
+ - type: accuracy
281
+ name: Upper Sorbian Test accuracy
282
+ value: 69.3
283
+ - type: accuracy
284
+ name: South Levantine Arabic Test accuracy
285
+ value: 70.0
286
+ - type: accuracy
287
+ name: Komi Zyrian Test accuracy
288
+ value: 44.9
289
+ - type: accuracy
290
+ name: Irish Test accuracy
291
+ value: 86.0
292
+ - type: accuracy
293
+ name: Nayini Test accuracy
294
+ value: 46.2
295
+ - type: accuracy
296
+ name: Munduruku Test accuracy
297
+ value: 38.9
298
+ - type: accuracy
299
+ name: Manx Test accuracy
300
+ value: 57.2
301
+ - type: accuracy
302
+ name: Skolt Sami Test accuracy
303
+ value: 40.1
304
+ - type: accuracy
305
+ name: Afrikaans Test accuracy
306
+ value: 73.0
307
+ - type: accuracy
308
+ name: Old Turkish Test accuracy
309
+ value: 39.4
310
+ - type: accuracy
311
+ name: Tupinamba Test accuracy
312
+ value: 51.8
313
+ - type: accuracy
314
+ name: Belarusian Test accuracy
315
+ value: 79.1
316
+ - type: accuracy
317
+ name: Serbian Test accuracy
318
+ value: 78.5
319
+ - type: accuracy
320
+ name: Moksha Test accuracy
321
+ value: 49.9
322
+ - type: accuracy
323
+ name: Western Armenian Test accuracy
324
+ value: 68.2
325
+ - type: accuracy
326
+ name: Scottish Gaelic Test accuracy
327
+ value: 77.1
328
+ - type: accuracy
329
+ name: Khunsari Test accuracy
330
+ value: 50.0
331
+ - type: accuracy
332
+ name: Hebrew Test accuracy
333
+ value: 80.2
334
+ - type: accuracy
335
+ name: Uyghur Test accuracy
336
+ value: 70.2
337
+ - type: accuracy
338
+ name: Chukchi Test accuracy
339
+ value: 39.3
340
+ ---
341
+
342
+ # XLM-RoBERTa base Universal Dependencies v2.8 POS tagging: Irish
343
+
344
+ This model is part of our paper called:
345
+
346
+ - Make the Best of Cross-lingual Transfer: Evidence from POS Tagging with over 100 Languages
347
+
348
+ Check the [Space]([Space](https://huggingface.co/spaces/wietsedv/xpos)) for more details.
config.json ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "output/xlm-roberta-base_ft_udpos28-ga/1d6ca3e8",
3
+ "architectures": [
4
+ "XLMRobertaForTokenClassification"
5
+ ],
6
+ "attention_probs_dropout_prob": 0.1,
7
+ "bos_token_id": 0,
8
+ "classifier_dropout": null,
9
+ "eos_token_id": 2,
10
+ "gradient_checkpointing": false,
11
+ "hidden_act": "gelu",
12
+ "hidden_dropout_prob": 0.1,
13
+ "hidden_size": 768,
14
+ "id2label": {
15
+ "0": "ADJ",
16
+ "1": "ADP",
17
+ "2": "ADV",
18
+ "3": "AUX",
19
+ "4": "CCONJ",
20
+ "5": "DET",
21
+ "6": "INTJ",
22
+ "7": "NOUN",
23
+ "8": "NUM",
24
+ "9": "PART",
25
+ "10": "PRON",
26
+ "11": "PROPN",
27
+ "12": "PUNCT",
28
+ "13": "SCONJ",
29
+ "14": "SYM",
30
+ "15": "VERB",
31
+ "16": "X"
32
+ },
33
+ "initializer_range": 0.02,
34
+ "intermediate_size": 3072,
35
+ "label2id": {
36
+ "ADJ": 0,
37
+ "ADP": 1,
38
+ "ADV": 2,
39
+ "AUX": 3,
40
+ "CCONJ": 4,
41
+ "DET": 5,
42
+ "INTJ": 6,
43
+ "NOUN": 7,
44
+ "NUM": 8,
45
+ "PART": 9,
46
+ "PRON": 10,
47
+ "PROPN": 11,
48
+ "PUNCT": 12,
49
+ "SCONJ": 13,
50
+ "SYM": 14,
51
+ "VERB": 15,
52
+ "X": 16
53
+ },
54
+ "layer_norm_eps": 1e-05,
55
+ "max_position_embeddings": 514,
56
+ "model_type": "xlm-roberta",
57
+ "num_attention_heads": 12,
58
+ "num_hidden_layers": 12,
59
+ "output_past": true,
60
+ "pad_token_id": 1,
61
+ "position_embedding_type": "absolute",
62
+ "torch_dtype": "float32",
63
+ "transformers_version": "4.10.2",
64
+ "type_vocab_size": 1,
65
+ "use_cache": true,
66
+ "vocab_size": 250002
67
+ }
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6838098d76d4d0ad28b22e8ecd45873b5223e48d5de1fade5efc3d7d854bfb53
3
+ size 1109946481
special_tokens_map.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"bos_token": "<s>", "eos_token": "</s>", "unk_token": "<unk>", "sep_token": "</s>", "pad_token": "<pad>", "cls_token": "<s>", "mask_token": {"content": "<mask>", "single_word": false, "lstrip": true, "rstrip": false, "normalized": false}}
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"bos_token": "<s>", "eos_token": "</s>", "sep_token": "</s>", "cls_token": "<s>", "unk_token": "<unk>", "pad_token": "<pad>", "mask_token": {"content": "<mask>", "single_word": false, "lstrip": true, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "add_prefix_space": true, "model_max_length": 512, "special_tokens_map_file": null, "name_or_path": "output/xlm-roberta-base_ft_udpos28-ga/1d6ca3e8", "tokenizer_class": "XLMRobertaTokenizer"}
train.args ADDED
@@ -0,0 +1 @@
 
 
1
+ udpos -tt=token-classification -tn=udpos28 -mi=xlm-roberta-base -mt=ft --learning_rate=5e-5 --eval_steps=1000 --eval_batch_size=10 --train_batch_size=10 --multi --max_steps=1000