Update config.json
Browse files- config.json +224 -208
config.json
CHANGED
@@ -1,58 +1,133 @@
|
|
1 |
{
|
|
|
|
|
|
|
2 |
"architectures": [
|
3 |
"XLMWithLMHeadModel"
|
4 |
],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
"dump_path": "/checkpoint/aconneau/dumped/xlm_17_100_big.3/16656237",
|
6 |
-
"
|
7 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
"exp_id": "16656237",
|
|
|
9 |
"fp16": true,
|
10 |
-
"amp": 2,
|
11 |
-
"encoder_only": true,
|
12 |
-
"emb_dim": 1280,
|
13 |
-
"n_layers": 16,
|
14 |
-
"n_heads": 16,
|
15 |
-
"dropout": 0.1,
|
16 |
-
"attention_dropout": 0.1,
|
17 |
"gelu_activation": true,
|
18 |
-
"
|
19 |
-
"sinusoidal_embeddings": false,
|
20 |
-
"use_lang_emb": false,
|
21 |
-
"use_memory": false,
|
22 |
-
"asm": false,
|
23 |
-
"context_size": 0,
|
24 |
-
"word_pred": 0.15,
|
25 |
-
"sample_alpha": 0.5,
|
26 |
-
"word_mask_keep_rand": "0.8,0.1,0.1",
|
27 |
-
"word_shuffle": 0.0,
|
28 |
-
"word_dropout": 0.0,
|
29 |
-
"word_blank": 0.0,
|
30 |
-
"data_path": "/private/home/aconneau/projects/XLM/data/wiki/17/175k",
|
31 |
-
"lgs": "en-fr-es-de-it-pt-nl-sv-pl-ru-ar-tr-zh-ja-ko-hi-vi",
|
32 |
-
"max_vocab": 200000,
|
33 |
-
"min_count": 0,
|
34 |
-
"lg_sampling_factor": 0.7,
|
35 |
-
"bptt": 256,
|
36 |
-
"max_len": 200,
|
37 |
"group_by_size": true,
|
38 |
-
"
|
39 |
-
"
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
52 |
"lambda_ae": 1.0,
|
53 |
-
"
|
54 |
"lambda_bt": 1.0,
|
55 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
"mlm_steps": [
|
57 |
[
|
58 |
"en",
|
@@ -123,198 +198,94 @@
|
|
123 |
null
|
124 |
]
|
125 |
],
|
126 |
-
"
|
127 |
-
"ae_steps": [],
|
128 |
-
"bt_steps": [],
|
129 |
-
"pc_steps": [],
|
130 |
-
"reload_emb": "",
|
131 |
-
"reload_model": "/checkpoint/aconneau/dumped/xlm_17_100_240_big_model_upper.2/14884510/best-valid_zh_mlm_ppl.pth",
|
132 |
-
"reload_checkpoint": "",
|
133 |
-
"beam_size": 1,
|
134 |
-
"length_penalty": 1,
|
135 |
-
"early_stopping": false,
|
136 |
-
"eval_bleu": false,
|
137 |
-
"eval_only": false,
|
138 |
-
"debug_train": false,
|
139 |
-
"debug_slurm": false,
|
140 |
-
"debug": false,
|
141 |
-
"local_rank": 0,
|
142 |
-
"master_port": 14148,
|
143 |
-
"langs": [
|
144 |
-
"en",
|
145 |
-
"fr",
|
146 |
-
"es",
|
147 |
-
"de",
|
148 |
-
"it",
|
149 |
-
"pt",
|
150 |
-
"nl",
|
151 |
-
"sv",
|
152 |
-
"pl",
|
153 |
-
"ru",
|
154 |
-
"ar",
|
155 |
-
"tr",
|
156 |
-
"zh",
|
157 |
-
"ja",
|
158 |
-
"ko",
|
159 |
-
"hi",
|
160 |
-
"vi"
|
161 |
-
],
|
162 |
-
"id2lang": {
|
163 |
-
"0": "ar",
|
164 |
-
"1": "de",
|
165 |
-
"2": "en",
|
166 |
-
"3": "es",
|
167 |
-
"4": "fr",
|
168 |
-
"5": "hi",
|
169 |
-
"6": "it",
|
170 |
-
"7": "ja",
|
171 |
-
"8": "ko",
|
172 |
-
"9": "nl",
|
173 |
-
"10": "pl",
|
174 |
-
"11": "pt",
|
175 |
-
"12": "ru",
|
176 |
-
"13": "sv",
|
177 |
-
"14": "tr",
|
178 |
-
"15": "vi",
|
179 |
-
"16": "zh"
|
180 |
-
},
|
181 |
-
"lang2id": {
|
182 |
-
"ar": 0,
|
183 |
-
"de": 1,
|
184 |
-
"en": 2,
|
185 |
-
"es": 3,
|
186 |
-
"fr": 4,
|
187 |
-
"hi": 5,
|
188 |
-
"it": 6,
|
189 |
-
"ja": 7,
|
190 |
-
"ko": 8,
|
191 |
-
"nl": 9,
|
192 |
-
"pl": 10,
|
193 |
-
"pt": 11,
|
194 |
-
"ru": 12,
|
195 |
-
"sv": 13,
|
196 |
-
"tr": 14,
|
197 |
-
"vi": 15,
|
198 |
-
"zh": 16
|
199 |
-
},
|
200 |
-
"n_langs": 17,
|
201 |
-
"bt_src_langs": [],
|
202 |
"mono_dataset": {
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
203 |
"en": {
|
|
|
204 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.en.pth",
|
205 |
-
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.en.pth"
|
206 |
-
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.en.pth"
|
207 |
-
},
|
208 |
-
"fr": {
|
209 |
-
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.fr.pth",
|
210 |
-
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.fr.pth",
|
211 |
-
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.fr.pth"
|
212 |
},
|
213 |
"es": {
|
|
|
214 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.es.pth",
|
215 |
-
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.es.pth"
|
216 |
-
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.es.pth"
|
217 |
},
|
218 |
-
"
|
219 |
-
"
|
220 |
-
"
|
221 |
-
"
|
|
|
|
|
|
|
|
|
|
|
222 |
},
|
223 |
"it": {
|
|
|
224 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.it.pth",
|
225 |
-
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.it.pth"
|
226 |
-
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.it.pth"
|
227 |
},
|
228 |
-
"
|
229 |
-
"
|
230 |
-
"
|
231 |
-
"
|
|
|
|
|
|
|
|
|
|
|
232 |
},
|
233 |
"nl": {
|
|
|
234 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.nl.pth",
|
235 |
-
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.nl.pth"
|
236 |
-
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.nl.pth"
|
237 |
-
},
|
238 |
-
"sv": {
|
239 |
-
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.sv.pth",
|
240 |
-
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.sv.pth",
|
241 |
-
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.sv.pth"
|
242 |
},
|
243 |
"pl": {
|
|
|
244 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.pl.pth",
|
245 |
-
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.pl.pth"
|
246 |
-
|
|
|
|
|
|
|
|
|
247 |
},
|
248 |
"ru": {
|
|
|
249 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.ru.pth",
|
250 |
-
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.ru.pth"
|
251 |
-
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.ru.pth"
|
252 |
},
|
253 |
-
"
|
254 |
-
"
|
255 |
-
"
|
256 |
-
"
|
257 |
},
|
258 |
"tr": {
|
|
|
259 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.tr.pth",
|
260 |
-
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.tr.pth"
|
261 |
-
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.tr.pth"
|
262 |
-
},
|
263 |
-
"zh": {
|
264 |
-
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.zh.pth",
|
265 |
-
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.zh.pth",
|
266 |
-
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.zh.pth"
|
267 |
-
},
|
268 |
-
"ja": {
|
269 |
-
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.ja.pth",
|
270 |
-
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.ja.pth",
|
271 |
-
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.ja.pth"
|
272 |
-
},
|
273 |
-
"ko": {
|
274 |
-
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.ko.pth",
|
275 |
-
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.ko.pth",
|
276 |
-
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.ko.pth"
|
277 |
-
},
|
278 |
-
"hi": {
|
279 |
-
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.hi.pth",
|
280 |
-
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.hi.pth",
|
281 |
-
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.hi.pth"
|
282 |
},
|
283 |
"vi": {
|
|
|
284 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.vi.pth",
|
285 |
-
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.vi.pth"
|
286 |
-
|
|
|
|
|
|
|
|
|
287 |
}
|
288 |
},
|
289 |
-
"para_dataset": {},
|
290 |
-
"word_mask": 0.8,
|
291 |
-
"word_keep": 0.1,
|
292 |
-
"word_rand": 0.1,
|
293 |
-
"is_slurm_job": true,
|
294 |
-
"n_nodes": 4,
|
295 |
-
"node_id": 0,
|
296 |
-
"global_rank": 0,
|
297 |
-
"world_size": 32,
|
298 |
-
"n_gpu_per_node": 8,
|
299 |
-
"master_addr": "learnfair1605",
|
300 |
-
"is_master": true,
|
301 |
-
"multi_node": true,
|
302 |
-
"multi_gpu": true,
|
303 |
-
"command": "python /private/home/aconneau/workdir/xlm_17_100_big.3/2019_08_10_19_23_42/train.py --n_heads 16 --bt_steps '' --max_vocab 200000 --word_mask_keep_rand '0.8,0.1,0.1' --use_lang_emb false --data_path '/private/home/aconneau/projects/XLM/data/wiki/17/175k' --save_periodic 0 --max_len 200 --bptt 256 --ae_steps '' --fp16 true --share_inout_emb true --sinusoidal_embeddings false --word_shuffle 0 --tokens_per_batch '-1' --accumulate_gradients 4 --validation_metrics '_valid_en_mlm_ppl,_valid_mlm_ppl,_valid_zh_mlm_ppl' --attention_dropout '0.1' --split_data true --max_epoch 100000 --stopping_criterion '_valid_zh_mlm_ppl,25' --dump_path '/checkpoint/aconneau/dumped' --epoch_size 200000 --word_blank 0 --gelu_activation true --n_layers 16 --optimizer 'adam_inverse_sqrt,lr=0.00005,warmup_updates=30000,beta1=0.9,beta2=0.999,weight_decay=0.01,eps=0.000001' --mlm_steps 'en,fr,es,de,it,pt,nl,sv,pl,ru,ar,tr,zh,ja,ko,hi,vi' --eval_bleu false --dropout '0.1' --mt_steps '' --batch_size 16 --word_dropout 0 --reload_model '/checkpoint/aconneau/dumped/xlm_17_100_240_big_model_upper.2/14884510/best-valid_zh_mlm_ppl.pth' --min_count 0 --amp 2 --group_by_size true --asm false --sample_alpha '0.5' --word_pred '0.15' --clip_grad_norm 1 --emb_dim 1280 --encoder_only true --lgs 'en-fr-es-de-it-pt-nl-sv-pl-ru-ar-tr-zh-ja-ko-hi-vi' --clm_steps '' --exp_name 'xlm_17_100_big.3' --lg_sampling_factor '0.7' --eval_only false --exp_id 16656237 --master_port 14148 --exp_id \"16656237\"",
|
304 |
-
"n_words": 200000,
|
305 |
-
"bos_index": 0,
|
306 |
-
"eos_index": 1,
|
307 |
-
"pad_index": 2,
|
308 |
-
"unk_index": 3,
|
309 |
-
"mask_index": 5,
|
310 |
-
"lambda_clm_config": null,
|
311 |
-
"lambda_mlm_config": null,
|
312 |
-
"lambda_pc_config": null,
|
313 |
-
"lambda_ae_config": null,
|
314 |
-
"lambda_mt_config": null,
|
315 |
-
"lambda_bt_config": null,
|
316 |
-
"hyp_path": "/checkpoint/aconneau/dumped/xlm_17_100_big.3/16656237/hypotheses",
|
317 |
-
"ref_paths": {},
|
318 |
"mono_list": [
|
319 |
"en",
|
320 |
"fr",
|
@@ -334,5 +305,50 @@
|
|
334 |
"hi",
|
335 |
"vi"
|
336 |
],
|
337 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
338 |
}
|
|
|
1 |
{
|
2 |
+
"accumulate_gradients": 4,
|
3 |
+
"ae_steps": [],
|
4 |
+
"amp": 2,
|
5 |
"architectures": [
|
6 |
"XLMWithLMHeadModel"
|
7 |
],
|
8 |
+
"asm": false,
|
9 |
+
"attention_dropout": 0.1,
|
10 |
+
"batch_size": 16,
|
11 |
+
"beam_size": 1,
|
12 |
+
"bos_index": 0,
|
13 |
+
"bos_token_id": 0,
|
14 |
+
"bptt": 256,
|
15 |
+
"bt_src_langs": [],
|
16 |
+
"bt_steps": [],
|
17 |
+
"causal": false,
|
18 |
+
"clip_grad_norm": 1.0,
|
19 |
+
"clm_steps": [],
|
20 |
+
"command": "python /private/home/aconneau/workdir/xlm_17_100_big.3/2019_08_10_19_23_42/train.py --n_heads 16 --bt_steps '' --max_vocab 200000 --word_mask_keep_rand '0.8,0.1,0.1' --use_lang_emb false --data_path '/private/home/aconneau/projects/XLM/data/wiki/17/175k' --save_periodic 0 --max_len 200 --bptt 256 --ae_steps '' --fp16 true --share_inout_emb true --sinusoidal_embeddings false --word_shuffle 0 --tokens_per_batch '-1' --accumulate_gradients 4 --validation_metrics '_valid_en_mlm_ppl,_valid_mlm_ppl,_valid_zh_mlm_ppl' --attention_dropout '0.1' --split_data true --max_epoch 100000 --stopping_criterion '_valid_zh_mlm_ppl,25' --dump_path '/checkpoint/aconneau/dumped' --epoch_size 200000 --word_blank 0 --gelu_activation true --n_layers 16 --optimizer 'adam_inverse_sqrt,lr=0.00005,warmup_updates=30000,beta1=0.9,beta2=0.999,weight_decay=0.01,eps=0.000001' --mlm_steps 'en,fr,es,de,it,pt,nl,sv,pl,ru,ar,tr,zh,ja,ko,hi,vi' --eval_bleu false --dropout '0.1' --mt_steps '' --batch_size 16 --word_dropout 0 --reload_model '/checkpoint/aconneau/dumped/xlm_17_100_240_big_model_upper.2/14884510/best-valid_zh_mlm_ppl.pth' --min_count 0 --amp 2 --group_by_size true --asm false --sample_alpha '0.5' --word_pred '0.15' --clip_grad_norm 1 --emb_dim 1280 --encoder_only true --lgs 'en-fr-es-de-it-pt-nl-sv-pl-ru-ar-tr-zh-ja-ko-hi-vi' --clm_steps '' --exp_name 'xlm_17_100_big.3' --lg_sampling_factor '0.7' --eval_only false --exp_id 16656237 --master_port 14148 --exp_id \"16656237\"",
|
21 |
+
"context_size": 0,
|
22 |
+
"data_path": "/private/home/aconneau/projects/XLM/data/wiki/17/175k",
|
23 |
+
"debug": false,
|
24 |
+
"debug_slurm": false,
|
25 |
+
"debug_train": false,
|
26 |
+
"dropout": 0.1,
|
27 |
"dump_path": "/checkpoint/aconneau/dumped/xlm_17_100_big.3/16656237",
|
28 |
+
"emb_dim": 1280,
|
29 |
+
"embed_init_std": 0.02209708691207961,
|
30 |
+
"encoder_only": true,
|
31 |
+
"end_n_top": 5,
|
32 |
+
"eos_index": 1,
|
33 |
+
"epoch_size": 200000,
|
34 |
+
"eval_bleu": false,
|
35 |
+
"eval_only": false,
|
36 |
"exp_id": "16656237",
|
37 |
+
"exp_name": "xlm_17_100_big.3",
|
38 |
"fp16": true,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
"gelu_activation": true,
|
40 |
+
"global_rank": 0,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
"group_by_size": true,
|
42 |
+
"hyp_path": "/checkpoint/aconneau/dumped/xlm_17_100_big.3/16656237/hypotheses",
|
43 |
+
"id2lang": {
|
44 |
+
"0": "ar",
|
45 |
+
"1": "de",
|
46 |
+
"10": "pl",
|
47 |
+
"11": "pt",
|
48 |
+
"12": "ru",
|
49 |
+
"13": "sv",
|
50 |
+
"14": "tr",
|
51 |
+
"15": "vi",
|
52 |
+
"16": "zh",
|
53 |
+
"2": "en",
|
54 |
+
"3": "es",
|
55 |
+
"4": "fr",
|
56 |
+
"5": "hi",
|
57 |
+
"6": "it",
|
58 |
+
"7": "ja",
|
59 |
+
"8": "ko",
|
60 |
+
"9": "nl"
|
61 |
+
},
|
62 |
+
"init_std": 0.02,
|
63 |
+
"is_encoder": true,
|
64 |
+
"is_master": true,
|
65 |
+
"is_slurm_job": true,
|
66 |
"lambda_ae": 1.0,
|
67 |
+
"lambda_ae_config": null,
|
68 |
"lambda_bt": 1.0,
|
69 |
+
"lambda_bt_config": null,
|
70 |
+
"lambda_clm": 1.0,
|
71 |
+
"lambda_clm_config": null,
|
72 |
+
"lambda_mlm": 1.0,
|
73 |
+
"lambda_mlm_config": null,
|
74 |
+
"lambda_mt": 1.0,
|
75 |
+
"lambda_mt_config": null,
|
76 |
+
"lambda_pc": 1.0,
|
77 |
+
"lambda_pc_config": null,
|
78 |
+
"lang2id": {
|
79 |
+
"ar": 0,
|
80 |
+
"de": 1,
|
81 |
+
"en": 2,
|
82 |
+
"es": 3,
|
83 |
+
"fr": 4,
|
84 |
+
"hi": 5,
|
85 |
+
"it": 6,
|
86 |
+
"ja": 7,
|
87 |
+
"ko": 8,
|
88 |
+
"nl": 9,
|
89 |
+
"pl": 10,
|
90 |
+
"pt": 11,
|
91 |
+
"ru": 12,
|
92 |
+
"sv": 13,
|
93 |
+
"tr": 14,
|
94 |
+
"vi": 15,
|
95 |
+
"zh": 16
|
96 |
+
},
|
97 |
+
"lang_id": 0,
|
98 |
+
"langs": [
|
99 |
+
"en",
|
100 |
+
"fr",
|
101 |
+
"es",
|
102 |
+
"de",
|
103 |
+
"it",
|
104 |
+
"pt",
|
105 |
+
"nl",
|
106 |
+
"sv",
|
107 |
+
"pl",
|
108 |
+
"ru",
|
109 |
+
"ar",
|
110 |
+
"tr",
|
111 |
+
"zh",
|
112 |
+
"ja",
|
113 |
+
"ko",
|
114 |
+
"hi",
|
115 |
+
"vi"
|
116 |
+
],
|
117 |
+
"layer_norm_eps": 1e-12,
|
118 |
+
"lg_sampling_factor": 0.7,
|
119 |
+
"lgs": "en-fr-es-de-it-pt-nl-sv-pl-ru-ar-tr-zh-ja-ko-hi-vi",
|
120 |
+
"local_rank": 0,
|
121 |
+
"mask_index": 5,
|
122 |
+
"mask_token_id": 0,
|
123 |
+
"master_addr": "learnfair1605",
|
124 |
+
"master_port": 14148,
|
125 |
+
"max_batch_size": 0,
|
126 |
+
"max_epoch": 100000,
|
127 |
+
"max_len": 200,
|
128 |
+
"max_position_embeddings": 512,
|
129 |
+
"max_vocab": 200000,
|
130 |
+
"min_count": 0,
|
131 |
"mlm_steps": [
|
132 |
[
|
133 |
"en",
|
|
|
198 |
null
|
199 |
]
|
200 |
],
|
201 |
+
"model_type": "xlm",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
202 |
"mono_dataset": {
|
203 |
+
"ar": {
|
204 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.ar.pth",
|
205 |
+
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.ar.pth",
|
206 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.ar.pth"
|
207 |
+
},
|
208 |
+
"de": {
|
209 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.de.pth",
|
210 |
+
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.de.pth",
|
211 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.de.pth"
|
212 |
+
},
|
213 |
"en": {
|
214 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.en.pth",
|
215 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.en.pth",
|
216 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.en.pth"
|
|
|
|
|
|
|
|
|
|
|
|
|
217 |
},
|
218 |
"es": {
|
219 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.es.pth",
|
220 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.es.pth",
|
221 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.es.pth"
|
|
|
222 |
},
|
223 |
+
"fr": {
|
224 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.fr.pth",
|
225 |
+
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.fr.pth",
|
226 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.fr.pth"
|
227 |
+
},
|
228 |
+
"hi": {
|
229 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.hi.pth",
|
230 |
+
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.hi.pth",
|
231 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.hi.pth"
|
232 |
},
|
233 |
"it": {
|
234 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.it.pth",
|
235 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.it.pth",
|
236 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.it.pth"
|
|
|
237 |
},
|
238 |
+
"ja": {
|
239 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.ja.pth",
|
240 |
+
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.ja.pth",
|
241 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.ja.pth"
|
242 |
+
},
|
243 |
+
"ko": {
|
244 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.ko.pth",
|
245 |
+
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.ko.pth",
|
246 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.ko.pth"
|
247 |
},
|
248 |
"nl": {
|
249 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.nl.pth",
|
250 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.nl.pth",
|
251 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.nl.pth"
|
|
|
|
|
|
|
|
|
|
|
|
|
252 |
},
|
253 |
"pl": {
|
254 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.pl.pth",
|
255 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.pl.pth",
|
256 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.pl.pth"
|
257 |
+
},
|
258 |
+
"pt": {
|
259 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.pt.pth",
|
260 |
+
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.pt.pth",
|
261 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.pt.pth"
|
262 |
},
|
263 |
"ru": {
|
264 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.ru.pth",
|
265 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.ru.pth",
|
266 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.ru.pth"
|
|
|
267 |
},
|
268 |
+
"sv": {
|
269 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.sv.pth",
|
270 |
+
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.sv.pth",
|
271 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.sv.pth"
|
272 |
},
|
273 |
"tr": {
|
274 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.tr.pth",
|
275 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.tr.pth",
|
276 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.tr.pth"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
277 |
},
|
278 |
"vi": {
|
279 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.vi.pth",
|
280 |
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.vi.pth",
|
281 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.vi.pth"
|
282 |
+
},
|
283 |
+
"zh": {
|
284 |
+
"test": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/test.zh.pth",
|
285 |
+
"train": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/train.zh.pth",
|
286 |
+
"valid": "/private/home/aconneau/projects/XLM/data/wiki/17/175k/valid.zh.pth"
|
287 |
}
|
288 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
289 |
"mono_list": [
|
290 |
"en",
|
291 |
"fr",
|
|
|
305 |
"hi",
|
306 |
"vi"
|
307 |
],
|
308 |
+
"mt_steps": [],
|
309 |
+
"multi_gpu": true,
|
310 |
+
"multi_node": true,
|
311 |
+
"n_gpu_per_node": 8,
|
312 |
+
"n_heads": 16,
|
313 |
+
"n_langs": 17,
|
314 |
+
"n_layers": 16,
|
315 |
+
"n_nodes": 4,
|
316 |
+
"node_id": 0,
|
317 |
+
"optimizer": "adam_inverse_sqrt,lr=0.00005,warmup_updates=30000,beta1=0.9,beta2=0.999,weight_decay=0.01,eps=0.000001",
|
318 |
+
"pad_index": 2,
|
319 |
+
"pad_token_id": 2,
|
320 |
+
"para_dataset": {},
|
321 |
+
"para_list": [],
|
322 |
+
"pc_steps": [],
|
323 |
+
"ref_paths": {},
|
324 |
+
"reload_checkpoint": "",
|
325 |
+
"reload_emb": "",
|
326 |
+
"reload_model": "/checkpoint/aconneau/dumped/xlm_17_100_240_big_model_upper.2/14884510/best-valid_zh_mlm_ppl.pth",
|
327 |
+
"sample_alpha": 0.5,
|
328 |
+
"save_periodic": 0,
|
329 |
+
"share_inout_emb": true,
|
330 |
+
"sinusoidal_embeddings": false,
|
331 |
+
"split_data": true,
|
332 |
+
"start_n_top": 5,
|
333 |
+
"stopping_criterion": "_valid_zh_mlm_ppl,25",
|
334 |
+
"summary_activation": null,
|
335 |
+
"summary_first_dropout": 0.1,
|
336 |
+
"summary_proj_to_labels": true,
|
337 |
+
"summary_type": "first",
|
338 |
+
"summary_use_proj": true,
|
339 |
+
"tokens_per_batch": -1,
|
340 |
+
"unk_index": 3,
|
341 |
+
"use_lang_emb": false,
|
342 |
+
"use_memory": false,
|
343 |
+
"validation_metrics": "_valid_en_mlm_ppl,_valid_mlm_ppl,_valid_zh_mlm_ppl",
|
344 |
+
"vocab_size": 200000,
|
345 |
+
"word_blank": 0.0,
|
346 |
+
"word_dropout": 0.0,
|
347 |
+
"word_keep": 0.1,
|
348 |
+
"word_mask": 0.8,
|
349 |
+
"word_mask_keep_rand": "0.8,0.1,0.1",
|
350 |
+
"word_pred": 0.15,
|
351 |
+
"word_rand": 0.1,
|
352 |
+
"word_shuffle": 0.0,
|
353 |
+
"world_size": 32
|
354 |
}
|