NethranjaliSE commited on
Commit
833752e
·
verified ·
1 Parent(s): 0899772

Add from-scratch Transformer (test BLEU 22.96)

Browse files
.gitattributes CHANGED
@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ sample_translations.png filter=lfs diff=lfs merge=lfs -text
37
+ sample_translations_best.png filter=lfs diff=lfs merge=lfs -text
38
+ training_report.png filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,88 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ language:
4
+ - si
5
+ - pi
6
+ tags:
7
+ - translation
8
+ - transformer
9
+ - from-scratch
10
+ - low-resource
11
+ - pali
12
+ - sinhala
13
+ pipeline_tag: translation
14
+ ---
15
+
16
+ # Pāli → Sinhala Transformer (from scratch)
17
+
18
+ An encoder–decoder Transformer implemented **from scratch** in PyTorch — no
19
+ `nn.Transformer`, no pretrained weights — for Pāli → Sinhala translation.
20
+ Both languages are written in Sinhala script in this corpus, so a single shared
21
+ subword vocabulary is used and the embeddings are tied across the encoder,
22
+ decoder and output projection.
23
+
24
+ ## Results
25
+
26
+ | Split | BLEU | chrF |
27
+ |---|---|---|
28
+ | Test | **22.96** | **46.44** |
29
+ | Validation | 22.86 | — |
30
+
31
+ Metrics are sacreBLEU (13a tokenizer) and chrF, single reference.
32
+
33
+ ## Architecture
34
+
35
+ | | |
36
+ |---|---|
37
+ | Layers | 5 encoder + 5 decoder (Pre-LN) |
38
+ | d_model / d_ff / heads | 512 / 2048 / 8 |
39
+ | Activation | GELU |
40
+ | Parameters | ~38.8M |
41
+ | Tokenizer | Shared SentencePiece **unigram**, 4000 vocab |
42
+ | Embeddings | Tied (encoder ⊗ decoder ⊗ output) |
43
+
44
+ ## Training
45
+
46
+ Trained for up to 300 epochs on ~16.5k parallel pairs with:
47
+
48
+ - label smoothing 0.2, residual dropout 0.3
49
+ - source word-dropout 0.1
50
+ - **subword regularisation** (unigram sampling, α=0.1) — a fresh
51
+ segmentation every epoch, acting as data augmentation
52
+ - token-bucketed batching (~3000 tokens × 2 accumulation)
53
+ - AdamW, linear warmup (1000 steps) + inverse-sqrt decay, peak LR 0.0007
54
+ - early stopping on **validation BLEU**, and **top-5 checkpoint averaging**
55
+
56
+ Decoding uses beam search (beam **8**) with a GNMT length penalty
57
+ α=**0.6**, both selected on the validation set.
58
+
59
+ ## Data preprocessing
60
+
61
+ NFC Unicode normalisation → zero-width character stripping → whitespace
62
+ squeezing → exact duplicate-pair removal (which also eliminates train/test
63
+ leakage) → length window 1–80 words → length-ratio cap 4.0.
64
+ Split 80/10/10.
65
+
66
+ ## Files
67
+
68
+ - `model_weights.pt` — state dict (averaged checkpoint)
69
+ - `spm.model` / `spm.vocab` — SentencePiece tokenizer
70
+ - `metadata.json` — architecture + decoding config
71
+ - `test_results.csv` — every test prediction with its sentence BLEU
72
+ - figures: training report, attention map, tokenization analysis, sample translations
73
+
74
+ ## Usage
75
+
76
+ The model classes are not packaged as a `transformers` architecture — see the
77
+ accompanying Space for a self-contained `app.py` that rebuilds the model and
78
+ loads this state dict:
79
+ <https://huggingface.co/spaces/NethranjaliSE/Pali_Sinhala_Translator_From_Scratch>
80
+
81
+ ## Limitations
82
+
83
+ Trained on a small, domain-specific corpus of canonical Buddhist text. Output is
84
+ frequently fluent but **not always faithful** — the model can produce a
85
+ well-formed Sinhala sentence that does not correspond to the Pāli source. It is
86
+ a research prototype and should not be relied on for doctrinal or scholarly
87
+ interpretation. Many Pāli lines have several valid Sinhala renderings, so
88
+ single-reference BLEU understates quality somewhat.
attention_map.png ADDED
history.json ADDED
@@ -0,0 +1,1736 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": [
3
+ 1,
4
+ 2,
5
+ 3,
6
+ 4,
7
+ 5,
8
+ 6,
9
+ 7,
10
+ 8,
11
+ 9,
12
+ 10,
13
+ 11,
14
+ 12,
15
+ 13,
16
+ 14,
17
+ 15,
18
+ 16,
19
+ 17,
20
+ 18,
21
+ 19,
22
+ 20,
23
+ 21,
24
+ 22,
25
+ 23,
26
+ 24,
27
+ 25,
28
+ 26,
29
+ 27,
30
+ 28,
31
+ 29,
32
+ 30,
33
+ 31,
34
+ 32,
35
+ 33,
36
+ 34,
37
+ 35,
38
+ 36,
39
+ 37,
40
+ 38,
41
+ 39,
42
+ 40,
43
+ 41,
44
+ 42,
45
+ 43,
46
+ 44,
47
+ 45,
48
+ 46,
49
+ 47,
50
+ 48,
51
+ 49,
52
+ 50,
53
+ 51,
54
+ 52,
55
+ 53,
56
+ 54,
57
+ 55,
58
+ 56,
59
+ 57,
60
+ 58,
61
+ 59,
62
+ 60,
63
+ 61,
64
+ 62,
65
+ 63,
66
+ 64,
67
+ 65,
68
+ 66,
69
+ 67,
70
+ 68,
71
+ 69,
72
+ 70,
73
+ 71,
74
+ 72,
75
+ 73,
76
+ 74,
77
+ 75,
78
+ 76,
79
+ 77,
80
+ 78,
81
+ 79,
82
+ 80,
83
+ 81,
84
+ 82,
85
+ 83,
86
+ 84,
87
+ 85,
88
+ 86,
89
+ 87,
90
+ 88,
91
+ 89,
92
+ 90,
93
+ 91,
94
+ 92,
95
+ 93,
96
+ 94,
97
+ 95,
98
+ 96,
99
+ 97,
100
+ 98,
101
+ 99,
102
+ 100,
103
+ 101,
104
+ 102,
105
+ 103,
106
+ 104,
107
+ 105,
108
+ 106,
109
+ 107,
110
+ 108,
111
+ 109,
112
+ 110,
113
+ 111,
114
+ 112,
115
+ 113,
116
+ 114,
117
+ 115,
118
+ 116,
119
+ 117,
120
+ 118,
121
+ 119,
122
+ 120,
123
+ 121,
124
+ 122,
125
+ 123,
126
+ 124,
127
+ 125,
128
+ 126,
129
+ 127,
130
+ 128,
131
+ 129,
132
+ 130,
133
+ 131,
134
+ 132,
135
+ 133,
136
+ 134,
137
+ 135,
138
+ 136,
139
+ 137,
140
+ 138,
141
+ 139,
142
+ 140,
143
+ 141,
144
+ 142,
145
+ 143,
146
+ 144,
147
+ 145,
148
+ 146,
149
+ 147,
150
+ 148,
151
+ 149,
152
+ 150,
153
+ 151,
154
+ 152,
155
+ 153,
156
+ 154,
157
+ 155,
158
+ 156,
159
+ 157,
160
+ 158,
161
+ 159,
162
+ 160,
163
+ 161,
164
+ 162,
165
+ 163,
166
+ 164,
167
+ 165,
168
+ 166,
169
+ 167,
170
+ 168,
171
+ 169,
172
+ 170,
173
+ 171,
174
+ 172,
175
+ 173,
176
+ 174,
177
+ 175,
178
+ 176,
179
+ 177,
180
+ 178,
181
+ 179,
182
+ 180,
183
+ 181,
184
+ 182,
185
+ 183,
186
+ 184,
187
+ 185,
188
+ 186,
189
+ 187,
190
+ 188,
191
+ 189,
192
+ 190,
193
+ 191,
194
+ 192,
195
+ 193,
196
+ 194,
197
+ 195,
198
+ 196,
199
+ 197,
200
+ 198,
201
+ 199,
202
+ 200,
203
+ 201,
204
+ 202,
205
+ 203,
206
+ 204,
207
+ 205,
208
+ 206,
209
+ 207,
210
+ 208,
211
+ 209,
212
+ 210,
213
+ 211,
214
+ 212,
215
+ 213,
216
+ 214,
217
+ 215,
218
+ 216,
219
+ 217,
220
+ 218,
221
+ 219,
222
+ 220,
223
+ 221,
224
+ 222,
225
+ 223,
226
+ 224,
227
+ 225,
228
+ 226,
229
+ 227,
230
+ 228,
231
+ 229,
232
+ 230,
233
+ 231,
234
+ 232,
235
+ 233,
236
+ 234,
237
+ 235,
238
+ 236,
239
+ 237,
240
+ 238,
241
+ 239,
242
+ 240,
243
+ 241,
244
+ 242,
245
+ 243,
246
+ 244,
247
+ 245,
248
+ 246,
249
+ 247,
250
+ 248,
251
+ 249,
252
+ 250,
253
+ 251,
254
+ 252,
255
+ 253,
256
+ 254,
257
+ 255,
258
+ 256,
259
+ 257,
260
+ 258,
261
+ 259,
262
+ 260,
263
+ 261,
264
+ 262,
265
+ 263,
266
+ 264,
267
+ 265,
268
+ 266,
269
+ 267,
270
+ 268,
271
+ 269,
272
+ 270,
273
+ 271,
274
+ 272,
275
+ 273,
276
+ 274,
277
+ 275,
278
+ 276,
279
+ 277,
280
+ 278,
281
+ 279,
282
+ 280,
283
+ 281,
284
+ 282,
285
+ 283,
286
+ 284,
287
+ 285,
288
+ 286,
289
+ 287
290
+ ],
291
+ "train_loss": [
292
+ 6.271046748173331,
293
+ 4.831270792010824,
294
+ 4.255707667014782,
295
+ 4.017632595247292,
296
+ 3.923262796621899,
297
+ 3.7968096077625275,
298
+ 3.673768427561082,
299
+ 3.5791950477792542,
300
+ 3.5003681458913496,
301
+ 3.436954984943861,
302
+ 3.3946334478185753,
303
+ 3.3642905755376877,
304
+ 3.306822721639873,
305
+ 3.266119178792471,
306
+ 3.2218770356407127,
307
+ 3.19906583088213,
308
+ 3.147455032420859,
309
+ 3.1071171641611333,
310
+ 3.0717675966471005,
311
+ 3.0336551119970054,
312
+ 2.994713069082757,
313
+ 2.972037556486879,
314
+ 2.9257093299144126,
315
+ 2.8883743255399916,
316
+ 2.8564398696718687,
317
+ 2.8193447527647297,
318
+ 2.783664282604808,
319
+ 2.7506251879289163,
320
+ 2.7265023668258923,
321
+ 2.6792382590575845,
322
+ 2.6443821652318356,
323
+ 2.6151597877752315,
324
+ 2.5786518131276104,
325
+ 2.5480276482464155,
326
+ 2.5057480078295846,
327
+ 2.478781985051796,
328
+ 2.4462704437637948,
329
+ 2.428281964395936,
330
+ 2.39572521775162,
331
+ 2.365361247002458,
332
+ 2.3464059545891147,
333
+ 2.32157974675122,
334
+ 2.3084490590858286,
335
+ 2.2661506782301903,
336
+ 2.26193128305313,
337
+ 2.2424430049163377,
338
+ 2.204703768333692,
339
+ 2.1941667492369787,
340
+ 2.172229020440824,
341
+ 2.158431833016611,
342
+ 2.1336481834913537,
343
+ 2.1204876675563704,
344
+ 2.10527086753768,
345
+ 2.083132828934082,
346
+ 2.0707845425072318,
347
+ 2.05542877300661,
348
+ 2.0374880307848877,
349
+ 2.024156170776884,
350
+ 2.0087931690072063,
351
+ 1.9971267067032459,
352
+ 1.9767825103915564,
353
+ 1.9701500616319034,
354
+ 1.9464237063710261,
355
+ 1.9334995563622535,
356
+ 1.922888878334113,
357
+ 1.9128580893869722,
358
+ 1.8947278010464057,
359
+ 1.8820587447986203,
360
+ 1.8759439571484726,
361
+ 1.8682627158175242,
362
+ 1.8475456707497737,
363
+ 1.8403206967716574,
364
+ 1.8316254374232164,
365
+ 1.817448295290001,
366
+ 1.8080214520900342,
367
+ 1.7939077322786479,
368
+ 1.7808831823370468,
369
+ 1.776127926605112,
370
+ 1.7683815799619138,
371
+ 1.751116892954856,
372
+ 1.7467632222899632,
373
+ 1.7322236924632857,
374
+ 1.7254952955258558,
375
+ 1.7117124709794225,
376
+ 1.7048510314059873,
377
+ 1.6937165838815285,
378
+ 1.6823338015769576,
379
+ 1.675013808878781,
380
+ 1.6655675783854338,
381
+ 1.6613850980470737,
382
+ 1.6496118453537474,
383
+ 1.6476512773850893,
384
+ 1.6292862455302117,
385
+ 1.6267975097742577,
386
+ 1.6149533230520254,
387
+ 1.6069110438281216,
388
+ 1.6063228921128239,
389
+ 1.5937505554603055,
390
+ 1.5788963360978163,
391
+ 1.578705990175345,
392
+ 1.5663100608439662,
393
+ 1.5567790093576237,
394
+ 1.5552735322677003,
395
+ 1.5492618758741787,
396
+ 1.5381295360910614,
397
+ 1.5308890442690444,
398
+ 1.528172701021325,
399
+ 1.5194734331696627,
400
+ 1.514849380685172,
401
+ 1.5084635073984576,
402
+ 1.4995086289063198,
403
+ 1.4913480500759673,
404
+ 1.483556018320436,
405
+ 1.478460942757438,
406
+ 1.4710379813315146,
407
+ 1.465494114105817,
408
+ 1.4594056109959053,
409
+ 1.4532910808370474,
410
+ 1.449126197993718,
411
+ 1.4421139694478995,
412
+ 1.4321991980231266,
413
+ 1.427326060923905,
414
+ 1.4239054368900994,
415
+ 1.4183868057031233,
416
+ 1.406897285783676,
417
+ 1.4059138856963438,
418
+ 1.3986500466556455,
419
+ 1.3890761085870311,
420
+ 1.3853080614995865,
421
+ 1.3827745126601998,
422
+ 1.3787251178647377,
423
+ 1.3700711312979548,
424
+ 1.3657187662868402,
425
+ 1.3609767868315858,
426
+ 1.3527073323262098,
427
+ 1.348151393824382,
428
+ 1.3393912098184828,
429
+ 1.340321779426834,
430
+ 1.3367289975994052,
431
+ 1.3282670769747242,
432
+ 1.3221379447325174,
433
+ 1.3194676107590602,
434
+ 1.3176939344470195,
435
+ 1.3106641687241214,
436
+ 1.3036384075467202,
437
+ 1.2998962675890582,
438
+ 1.2969033420183476,
439
+ 1.2883640898784987,
440
+ 1.2844244529001247,
441
+ 1.2809234223274386,
442
+ 1.2737414176015345,
443
+ 1.274201176882293,
444
+ 1.2669377403480095,
445
+ 1.2631510210207124,
446
+ 1.2621024418140323,
447
+ 1.2547332091056072,
448
+ 1.252503751820779,
449
+ 1.2474942457576286,
450
+ 1.2388351920658947,
451
+ 1.2377693581665508,
452
+ 1.233666173974901,
453
+ 1.2326546863318548,
454
+ 1.22870988331459,
455
+ 1.2219762364978177,
456
+ 1.2190169714696353,
457
+ 1.2174874800699225,
458
+ 1.2117568407349222,
459
+ 1.2059071219020467,
460
+ 1.2051250554059898,
461
+ 1.1978071656767706,
462
+ 1.1965725224181094,
463
+ 1.1904930458322105,
464
+ 1.1874062679024366,
465
+ 1.186812179447507,
466
+ 1.1823858060874364,
467
+ 1.1780288671525228,
468
+ 1.1735446991956067,
469
+ 1.1737225704652163,
470
+ 1.1675755169135902,
471
+ 1.1647423214332486,
472
+ 1.1610140120984658,
473
+ 1.1581154961358606,
474
+ 1.1540897369001908,
475
+ 1.1575263118259491,
476
+ 1.1536281408205258,
477
+ 1.14391873810019,
478
+ 1.1415842288212166,
479
+ 1.1402960677837488,
480
+ 1.1374186444811167,
481
+ 1.131812059596347,
482
+ 1.1271090991337962,
483
+ 1.1265010173948953,
484
+ 1.1241349789157953,
485
+ 1.1238172064574812,
486
+ 1.121695727817151,
487
+ 1.1175067607926736,
488
+ 1.1134037723098122,
489
+ 1.1122707843149968,
490
+ 1.108565347837879,
491
+ 1.1099735511846618,
492
+ 1.1023158255225383,
493
+ 1.1006609687251536,
494
+ 1.0966945600946345,
495
+ 1.0962864319183487,
496
+ 1.090230326315763,
497
+ 1.089362343739573,
498
+ 1.0869473817939221,
499
+ 1.088397431487161,
500
+ 1.078814047859879,
501
+ 1.079169949440363,
502
+ 1.0760323087869472,
503
+ 1.0736110920927355,
504
+ 1.069484271947877,
505
+ 1.066452766962757,
506
+ 1.0667501657798926,
507
+ 1.0613168012843692,
508
+ 1.0610192753509293,
509
+ 1.0588033610059524,
510
+ 1.0556656853285786,
511
+ 1.0520866705365717,
512
+ 1.0500202583955618,
513
+ 1.0492029988344267,
514
+ 1.04579253518464,
515
+ 1.0447212769525853,
516
+ 1.0438316128094507,
517
+ 1.0393227197849717,
518
+ 1.0393742512773647,
519
+ 1.0362167486794824,
520
+ 1.032897441406332,
521
+ 1.0327285994192725,
522
+ 1.029177234128658,
523
+ 1.028031321973655,
524
+ 1.0227550775372345,
525
+ 1.0239170557333517,
526
+ 1.018346294696238,
527
+ 1.0193255552548923,
528
+ 1.0191001811519278,
529
+ 1.0155255265767895,
530
+ 1.0155541461676199,
531
+ 1.0130998749201907,
532
+ 1.0117619818996493,
533
+ 1.0084443590304284,
534
+ 1.0082533192763417,
535
+ 1.003609491415026,
536
+ 1.0003201381184692,
537
+ 0.998461155330531,
538
+ 0.9964239150060761,
539
+ 0.9949075074898845,
540
+ 0.9917866249094238,
541
+ 0.9909628341020358,
542
+ 0.9908968429563436,
543
+ 0.9891882932206478,
544
+ 0.9836452228907945,
545
+ 0.9837336372107515,
546
+ 0.9833572313245159,
547
+ 0.9800462036566073,
548
+ 0.9790241695343068,
549
+ 0.976128654915547,
550
+ 0.9744272046448612,
551
+ 0.9729728588703502,
552
+ 0.9728256431960494,
553
+ 0.9700733737532595,
554
+ 0.9693489810346232,
555
+ 0.9653384947294228,
556
+ 0.9649519179120849,
557
+ 0.9667473037196024,
558
+ 0.9610786104861894,
559
+ 0.9591420211950031,
560
+ 0.9561623810394708,
561
+ 0.9581470181935572,
562
+ 0.9533353512776495,
563
+ 0.9537106611972271,
564
+ 0.954686186615216,
565
+ 0.953792244120756,
566
+ 0.9510382368223419,
567
+ 0.9488015582352214,
568
+ 0.9435153373181392,
569
+ 0.9441851850097729,
570
+ 0.946552116757156,
571
+ 0.9416634161437694,
572
+ 0.9411753108473443,
573
+ 0.9375788797181736,
574
+ 0.9396343061010078,
575
+ 0.9348346679493812,
576
+ 0.9347158157978538,
577
+ 0.9315023044268037,
578
+ 0.9329545814088113
579
+ ],
580
+ "val_loss": [
581
+ 5.804056569250296,
582
+ 5.420388197781151,
583
+ 5.264307954335404,
584
+ 5.289802705547589,
585
+ 5.076968792975378,
586
+ 5.013457190448476,
587
+ 4.8969080980313064,
588
+ 4.848441029963388,
589
+ 4.714395969474642,
590
+ 4.697259353649663,
591
+ 4.658327721342877,
592
+ 4.628471561185472,
593
+ 4.484175609943922,
594
+ 4.465716642696802,
595
+ 4.410384810944443,
596
+ 4.343532033526733,
597
+ 4.299635290584552,
598
+ 4.226750110236001,
599
+ 4.2259080466407974,
600
+ 4.17679486283793,
601
+ 4.109893477286755,
602
+ 4.0653461411126335,
603
+ 4.007909129245172,
604
+ 3.9772642050179843,
605
+ 3.970471250326686,
606
+ 3.902621263475287,
607
+ 3.8606600554066413,
608
+ 3.8251807527197803,
609
+ 3.7870887177786545,
610
+ 3.738227300500703,
611
+ 3.682507830020749,
612
+ 3.6707086195991843,
613
+ 3.599589195731637,
614
+ 3.5640759696393465,
615
+ 3.561816455760725,
616
+ 3.5115351497673295,
617
+ 3.4838539118434246,
618
+ 3.4226678487597804,
619
+ 3.4317281682950873,
620
+ 3.391700855107789,
621
+ 3.384118788871616,
622
+ 3.327868115249619,
623
+ 3.2866118098943473,
624
+ 3.2588514561296074,
625
+ 3.2450205382680823,
626
+ 3.224622161551234,
627
+ 3.1970314629478627,
628
+ 3.2021196854749694,
629
+ 3.149548178284631,
630
+ 3.146527273508777,
631
+ 3.1317515657213435,
632
+ 3.1211452996183926,
633
+ 3.083865949252191,
634
+ 3.0665301066861432,
635
+ 3.0582081841186497,
636
+ 3.0316046289862046,
637
+ 3.037244095463391,
638
+ 3.030443633480637,
639
+ 3.002494706793151,
640
+ 2.9887457569056,
641
+ 2.9837230844057236,
642
+ 2.9696549644222037,
643
+ 2.9498907961832646,
644
+ 2.957230533145382,
645
+ 2.9489960293175383,
646
+ 2.921606620467511,
647
+ 2.909011382191938,
648
+ 2.9072091295699036,
649
+ 2.92339199185494,
650
+ 2.8730097441265205,
651
+ 2.8710114969512706,
652
+ 2.8610005301841612,
653
+ 2.8767702165171825,
654
+ 2.844461136223774,
655
+ 2.83850439477423,
656
+ 2.817113484602557,
657
+ 2.8269662761781422,
658
+ 2.806226945995083,
659
+ 2.7983455861995457,
660
+ 2.7978125160256915,
661
+ 2.7859864644100822,
662
+ 2.802858350478531,
663
+ 2.795093098189011,
664
+ 2.7661418030841243,
665
+ 2.767030306998851,
666
+ 2.7426099189881117,
667
+ 2.7426935048584977,
668
+ 2.7483552589114373,
669
+ 2.746204244087469,
670
+ 2.7430987076071696,
671
+ 2.729823310969666,
672
+ 2.7086419640463384,
673
+ 2.7099485293113506,
674
+ 2.706077773515591,
675
+ 2.7211465775978905,
676
+ 2.695986701195879,
677
+ 2.6862540304158062,
678
+ 2.704642805254182,
679
+ 2.6863538724383904,
680
+ 2.68854290175845,
681
+ 2.699203557829336,
682
+ 2.684399148069569,
683
+ 2.6801295453225404,
684
+ 2.670977227139978,
685
+ 2.6748165400203328,
686
+ 2.6631952103727814,
687
+ 2.6590796525624034,
688
+ 2.653530227637688,
689
+ 2.6282221096407063,
690
+ 2.6475042596664773,
691
+ 2.6402339226325044,
692
+ 2.643476372845647,
693
+ 2.6493627274264475,
694
+ 2.6405724251841693,
695
+ 2.655842510240359,
696
+ 2.639418796400699,
697
+ 2.6350592814264804,
698
+ 2.6278027761034015,
699
+ 2.645291093052525,
700
+ 2.62062748462863,
701
+ 2.616117890603109,
702
+ 2.620834569581814,
703
+ 2.632086639883979,
704
+ 2.6066302238141428,
705
+ 2.6158283922903283,
706
+ 2.624299183325423,
707
+ 2.6076911898263564,
708
+ 2.616625135351027,
709
+ 2.6046203908619137,
710
+ 2.6131700722799946,
711
+ 2.601900559557095,
712
+ 2.60282259903629,
713
+ 2.6130259958124777,
714
+ 2.601894636638271,
715
+ 2.6020425318617115,
716
+ 2.6017687952437534,
717
+ 2.5910851928928316,
718
+ 2.5944399383670875,
719
+ 2.591906640331482,
720
+ 2.5829276221447386,
721
+ 2.588687991524547,
722
+ 2.5806064180596033,
723
+ 2.586710349077846,
724
+ 2.585704231588371,
725
+ 2.5941810130558003,
726
+ 2.5903895310329523,
727
+ 2.578860282395215,
728
+ 2.5791178762949536,
729
+ 2.5908361827731254,
730
+ 2.5804059117499065,
731
+ 2.578184050055498,
732
+ 2.5611949393813487,
733
+ 2.5635378675263647,
734
+ 2.5765972963529125,
735
+ 2.576215044494716,
736
+ 2.5801913082366665,
737
+ 2.5699062008863613,
738
+ 2.564154976178432,
739
+ 2.5713777206876447,
740
+ 2.5773464235423695,
741
+ 2.5757259682504463,
742
+ 2.5651567234004187,
743
+ 2.567392133288216,
744
+ 2.569736526020268,
745
+ 2.5691240664691843,
746
+ 2.5600027404579944,
747
+ 2.5796224455655885,
748
+ 2.5621679842018388,
749
+ 2.569871592082695,
750
+ 2.5633892757096572,
751
+ 2.5709872833129137,
752
+ 2.5597417078079987,
753
+ 2.550717879067574,
754
+ 2.561191658403765,
755
+ 2.552506132286239,
756
+ 2.559284784940505,
757
+ 2.572236192763477,
758
+ 2.5615098560158054,
759
+ 2.5657335143598403,
760
+ 2.54793819723196,
761
+ 2.5586458932983116,
762
+ 2.5713471693037175,
763
+ 2.5567676423386128,
764
+ 2.5571608553205443,
765
+ 2.5544862693730903,
766
+ 2.5669145453763433,
767
+ 2.5587508462383237,
768
+ 2.5634959516420306,
769
+ 2.5478686219722344,
770
+ 2.55375047114817,
771
+ 2.562762023072948,
772
+ 2.5627384788581455,
773
+ 2.5608170494403497,
774
+ 2.566316142834599,
775
+ 2.5637799130989873,
776
+ 2.542344424342501,
777
+ 2.568140642899492,
778
+ 2.550770572332462,
779
+ 2.5497987911525124,
780
+ 2.5658265558370124,
781
+ 2.5459840672297185,
782
+ 2.546463177334598,
783
+ 2.5660210059665647,
784
+ 2.5581093443639236,
785
+ 2.568926992582652,
786
+ 2.552270392605464,
787
+ 2.5574898628927607,
788
+ 2.55946779009541,
789
+ 2.5690907968387906,
790
+ 2.560291921950201,
791
+ 2.5545522760300203,
792
+ 2.5527039380800427,
793
+ 2.5603628399023375,
794
+ 2.5568138627794097,
795
+ 2.5571103594205526,
796
+ 2.557725991481274,
797
+ 2.564639148214819,
798
+ 2.5658901674809758,
799
+ 2.5605194088339047,
800
+ 2.5586463205487218,
801
+ 2.563436088837167,
802
+ 2.572333516456099,
803
+ 2.5619743685495755,
804
+ 2.561225081490676,
805
+ 2.5625286942964425,
806
+ 2.5688623379125834,
807
+ 2.5658997867338402,
808
+ 2.5572125183483916,
809
+ 2.556683779155195,
810
+ 2.5569527521468904,
811
+ 2.5671992420512595,
812
+ 2.552957107171775,
813
+ 2.5651980743710334,
814
+ 2.5598654848291806,
815
+ 2.567096432487407,
816
+ 2.554816717339798,
817
+ 2.554703817068856,
818
+ 2.557441238917041,
819
+ 2.5678872255306877,
820
+ 2.5700128747587514,
821
+ 2.5694466363106017,
822
+ 2.5641568569679904,
823
+ 2.5599414734309702,
824
+ 2.5636929344969066,
825
+ 2.5568736790997626,
826
+ 2.5607473265729896,
827
+ 2.55652409144523,
828
+ 2.5557475057042596,
829
+ 2.5616828269905527,
830
+ 2.565222002380401,
831
+ 2.555111225385574,
832
+ 2.570889169996694,
833
+ 2.5669934910355563,
834
+ 2.570512892765362,
835
+ 2.570241494817367,
836
+ 2.562932703726038,
837
+ 2.56612658932287,
838
+ 2.562775759754836,
839
+ 2.5752340152390674,
840
+ 2.551434288297605,
841
+ 2.5533765495080956,
842
+ 2.5611316545513327,
843
+ 2.5547640129166655,
844
+ 2.554541709368623,
845
+ 2.5587127729713903,
846
+ 2.56865611859394,
847
+ 2.5590366207628774,
848
+ 2.557532721440308,
849
+ 2.5779698616412166,
850
+ 2.5651473760163435,
851
+ 2.5613058827904807,
852
+ 2.5667992113083384,
853
+ 2.5596683720114104,
854
+ 2.5724847983661636,
855
+ 2.573387730369301,
856
+ 2.572874787878387,
857
+ 2.570450218364894,
858
+ 2.5792187850457164,
859
+ 2.562839593018703,
860
+ 2.5658957415145993,
861
+ 2.5747191558475726,
862
+ 2.571571861758347,
863
+ 2.571294846210527,
864
+ 2.5674635469239084,
865
+ 2.5689455264758925,
866
+ 2.566795476987219,
867
+ 2.570247895456199
868
+ ],
869
+ "bleu": [
870
+ 0.0,
871
+ 0.0,
872
+ 0.0,
873
+ 0.0,
874
+ 0.0179,
875
+ 0.0425,
876
+ 0.0283,
877
+ 0.1081,
878
+ 0.2627,
879
+ 0.3712,
880
+ 0.2578,
881
+ 0.2845,
882
+ 0.4214,
883
+ 0.2922,
884
+ 0.5784,
885
+ 0.4902,
886
+ 0.4615,
887
+ 0.453,
888
+ 0.4024,
889
+ 0.4601,
890
+ 0.4643,
891
+ 0.5207,
892
+ 0.5687,
893
+ 0.5294,
894
+ 0.526,
895
+ 0.7393,
896
+ 0.6954,
897
+ 0.5537,
898
+ 0.6267,
899
+ 1.0035,
900
+ 0.9886,
901
+ 0.6205,
902
+ 1.5504,
903
+ 2.1127,
904
+ 1.988,
905
+ 1.8387,
906
+ 2.052,
907
+ 3.08,
908
+ 3.603,
909
+ 3.7596,
910
+ 3.1799,
911
+ 3.8788,
912
+ 3.7185,
913
+ 4.038,
914
+ 4.0552,
915
+ 4.61,
916
+ 4.4592,
917
+ 5.2078,
918
+ 3.8311,
919
+ 5.8869,
920
+ 5.7241,
921
+ 5.3193,
922
+ 5.68,
923
+ 5.9953,
924
+ 6.6062,
925
+ 6.1982,
926
+ 6.4198,
927
+ 6.3648,
928
+ 6.8617,
929
+ 6.4536,
930
+ 7.6901,
931
+ 7.6258,
932
+ 7.3058,
933
+ 7.5982,
934
+ 7.8134,
935
+ 8.1339,
936
+ 8.798,
937
+ 8.9404,
938
+ 8.0351,
939
+ 9.1234,
940
+ 8.9877,
941
+ 8.6618,
942
+ 8.1846,
943
+ 8.9379,
944
+ 9.1654,
945
+ 9.3617,
946
+ 9.0566,
947
+ 9.6525,
948
+ 9.7995,
949
+ 9.2425,
950
+ 10.584,
951
+ 9.5237,
952
+ 10.451,
953
+ 10.922,
954
+ 11.613,
955
+ 11.066,
956
+ 11.1214,
957
+ 10.9408,
958
+ 10.322,
959
+ 10.8183,
960
+ 12.384,
961
+ 12.3028,
962
+ 13.025,
963
+ 12.0998,
964
+ 12.7284,
965
+ 12.5862,
966
+ 12.3682,
967
+ 12.0267,
968
+ 12.8673,
969
+ 13.1772,
970
+ 13.1393,
971
+ 12.7187,
972
+ 13.2015,
973
+ 13.3465,
974
+ 13.5144,
975
+ 13.3798,
976
+ 13.2607,
977
+ 12.5828,
978
+ 12.7969,
979
+ 14.1684,
980
+ 13.3646,
981
+ 13.9724,
982
+ 14.0803,
983
+ 13.7048,
984
+ 14.9395,
985
+ 14.155,
986
+ 14.3952,
987
+ 14.6837,
988
+ 14.6516,
989
+ 15.2779,
990
+ 15.2357,
991
+ 14.1503,
992
+ 15.3263,
993
+ 15.0403,
994
+ 15.7406,
995
+ 14.9066,
996
+ 15.3495,
997
+ 15.3314,
998
+ 14.414,
999
+ 15.3909,
1000
+ 15.84,
1001
+ 16.2761,
1002
+ 15.6907,
1003
+ 16.122,
1004
+ 15.8605,
1005
+ 16.2856,
1006
+ 15.8644,
1007
+ 16.2742,
1008
+ 15.6744,
1009
+ 15.5896,
1010
+ 16.0983,
1011
+ 16.0427,
1012
+ 16.5148,
1013
+ 16.5986,
1014
+ 17.3808,
1015
+ 16.7412,
1016
+ 17.0508,
1017
+ 16.3028,
1018
+ 17.5072,
1019
+ 17.7225,
1020
+ 17.7356,
1021
+ 17.8223,
1022
+ 16.933,
1023
+ 18.2871,
1024
+ 17.5556,
1025
+ 17.2635,
1026
+ 17.2973,
1027
+ 18.2002,
1028
+ 16.8072,
1029
+ 18.5503,
1030
+ 17.0469,
1031
+ 16.7876,
1032
+ 17.3603,
1033
+ 17.8433,
1034
+ 17.5306,
1035
+ 17.6231,
1036
+ 17.4534,
1037
+ 17.8253,
1038
+ 17.6853,
1039
+ 17.9425,
1040
+ 17.6953,
1041
+ 19.203,
1042
+ 19.0538,
1043
+ 17.9083,
1044
+ 18.0494,
1045
+ 18.7251,
1046
+ 18.46,
1047
+ 19.136,
1048
+ 18.7063,
1049
+ 18.3315,
1050
+ 18.3394,
1051
+ 17.9155,
1052
+ 18.8544,
1053
+ 18.7204,
1054
+ 19.0428,
1055
+ 19.5473,
1056
+ 18.4842,
1057
+ 18.8025,
1058
+ 18.1946,
1059
+ 19.3443,
1060
+ 18.6106,
1061
+ 18.8224,
1062
+ 18.9639,
1063
+ 19.6391,
1064
+ 18.7904,
1065
+ 19.0153,
1066
+ 18.9217,
1067
+ 19.7397,
1068
+ 20.2085,
1069
+ 19.2465,
1070
+ 20.3485,
1071
+ 20.0377,
1072
+ 19.7704,
1073
+ 20.1482,
1074
+ 20.3801,
1075
+ 19.3258,
1076
+ 19.4662,
1077
+ 19.7168,
1078
+ 19.4587,
1079
+ 19.3115,
1080
+ 19.5706,
1081
+ 19.8419,
1082
+ 19.3177,
1083
+ 18.9159,
1084
+ 19.8602,
1085
+ 19.4896,
1086
+ 20.3011,
1087
+ 19.7875,
1088
+ 19.3931,
1089
+ 20.5411,
1090
+ 20.9942,
1091
+ 20.3037,
1092
+ 20.2601,
1093
+ 21.2314,
1094
+ 20.5351,
1095
+ 21.0987,
1096
+ 21.4207,
1097
+ 21.2933,
1098
+ 21.1761,
1099
+ 21.0537,
1100
+ 20.9365,
1101
+ 21.0557,
1102
+ 20.8879,
1103
+ 21.1334,
1104
+ 21.2432,
1105
+ 21.5803,
1106
+ 21.938,
1107
+ 21.1681,
1108
+ 21.212,
1109
+ 20.5839,
1110
+ 20.0797,
1111
+ 21.1533,
1112
+ 20.6871,
1113
+ 20.9006,
1114
+ 21.4376,
1115
+ 20.4801,
1116
+ 21.2663,
1117
+ 20.5233,
1118
+ 20.4851,
1119
+ 21.247,
1120
+ 20.8165,
1121
+ 20.6524,
1122
+ 21.3823,
1123
+ 20.9884,
1124
+ 21.041,
1125
+ 20.8447,
1126
+ 21.7294,
1127
+ 20.7953,
1128
+ 20.8748,
1129
+ 22.1623,
1130
+ 21.6543,
1131
+ 22.7495,
1132
+ 21.3308,
1133
+ 22.1107,
1134
+ 20.0613,
1135
+ 21.8006,
1136
+ 21.7231,
1137
+ 20.9757,
1138
+ 20.9985,
1139
+ 21.7217,
1140
+ 21.4047,
1141
+ 20.3734,
1142
+ 21.7985,
1143
+ 20.9503,
1144
+ 21.8382,
1145
+ 21.891,
1146
+ 22.0469,
1147
+ 21.7242,
1148
+ 21.9382,
1149
+ 22.2363,
1150
+ 22.7072,
1151
+ 22.3981,
1152
+ 22.5043,
1153
+ 21.6483,
1154
+ 21.8237,
1155
+ 21.7008,
1156
+ 21.3125
1157
+ ],
1158
+ "chrf": [
1159
+ 0.0,
1160
+ 0.0,
1161
+ 0.0,
1162
+ 0.9621,
1163
+ 2.0718,
1164
+ 2.9281,
1165
+ 4.2183,
1166
+ 5.7807,
1167
+ 5.2123,
1168
+ 7.382,
1169
+ 8.3308,
1170
+ 7.4772,
1171
+ 5.8322,
1172
+ 7.8038,
1173
+ 8.8803,
1174
+ 7.984,
1175
+ 9.824,
1176
+ 7.6279,
1177
+ 6.1017,
1178
+ 8.1084,
1179
+ 8.1937,
1180
+ 8.9945,
1181
+ 10.2126,
1182
+ 9.3963,
1183
+ 11.3469,
1184
+ 11.9229,
1185
+ 10.2126,
1186
+ 11.8987,
1187
+ 12.0584,
1188
+ 12.1155,
1189
+ 13.8199,
1190
+ 12.5443,
1191
+ 15.4592,
1192
+ 15.6002,
1193
+ 16.7653,
1194
+ 16.2653,
1195
+ 17.4464,
1196
+ 18.1673,
1197
+ 19.1163,
1198
+ 19.32,
1199
+ 19.8765,
1200
+ 20.3762,
1201
+ 20.9011,
1202
+ 21.7605,
1203
+ 21.4849,
1204
+ 22.7539,
1205
+ 22.4434,
1206
+ 22.7861,
1207
+ 23.0249,
1208
+ 25.1997,
1209
+ 24.4089,
1210
+ 23.733,
1211
+ 25.7117,
1212
+ 26.0423,
1213
+ 26.2037,
1214
+ 26.4723,
1215
+ 26.1696,
1216
+ 27.2767,
1217
+ 27.2992,
1218
+ 28.3235,
1219
+ 29.7921,
1220
+ 28.9881,
1221
+ 29.3063,
1222
+ 28.9294,
1223
+ 29.9468,
1224
+ 29.0273,
1225
+ 29.7794,
1226
+ 30.4813,
1227
+ 28.737,
1228
+ 31.1487,
1229
+ 30.6825,
1230
+ 31.0892,
1231
+ 30.5545,
1232
+ 31.1347,
1233
+ 31.0182,
1234
+ 31.9171,
1235
+ 31.2572,
1236
+ 32.425,
1237
+ 32.8726,
1238
+ 32.1106,
1239
+ 32.8462,
1240
+ 32.3069,
1241
+ 33.3266,
1242
+ 33.9007,
1243
+ 33.7235,
1244
+ 34.0562,
1245
+ 34.2789,
1246
+ 34.9364,
1247
+ 33.5639,
1248
+ 33.7281,
1249
+ 34.2087,
1250
+ 35.1925,
1251
+ 35.6853,
1252
+ 35.3041,
1253
+ 35.5207,
1254
+ 35.7128,
1255
+ 35.6198,
1256
+ 34.385,
1257
+ 36.1598,
1258
+ 36.7533,
1259
+ 36.1774,
1260
+ 35.9977,
1261
+ 36.6019,
1262
+ 36.1666,
1263
+ 36.089,
1264
+ 36.5192,
1265
+ 36.2647,
1266
+ 35.7986,
1267
+ 36.3632,
1268
+ 37.6835,
1269
+ 36.8083,
1270
+ 38.0955,
1271
+ 36.8394,
1272
+ 38.4754,
1273
+ 37.3669,
1274
+ 37.1739,
1275
+ 37.1864,
1276
+ 38.6535,
1277
+ 37.5414,
1278
+ 38.1588,
1279
+ 38.3332,
1280
+ 37.9641,
1281
+ 38.2424,
1282
+ 38.009,
1283
+ 38.9859,
1284
+ 38.2977,
1285
+ 38.7461,
1286
+ 39.3121,
1287
+ 38.5468,
1288
+ 38.5588,
1289
+ 39.4544,
1290
+ 39.8001,
1291
+ 39.3617,
1292
+ 39.9736,
1293
+ 38.8508,
1294
+ 38.8673,
1295
+ 40.877,
1296
+ 40.2796,
1297
+ 39.4047,
1298
+ 39.1767,
1299
+ 40.1751,
1300
+ 39.7396,
1301
+ 40.2534,
1302
+ 39.9218,
1303
+ 40.7393,
1304
+ 40.5539,
1305
+ 40.265,
1306
+ 39.571,
1307
+ 41.174,
1308
+ 41.1075,
1309
+ 41.4882,
1310
+ 41.0526,
1311
+ 40.7676,
1312
+ 41.5073,
1313
+ 41.123,
1314
+ 40.1413,
1315
+ 40.3533,
1316
+ 41.186,
1317
+ 40.5853,
1318
+ 42.0683,
1319
+ 41.1801,
1320
+ 41.4729,
1321
+ 40.6051,
1322
+ 40.7825,
1323
+ 41.2164,
1324
+ 40.7676,
1325
+ 40.9755,
1326
+ 42.1848,
1327
+ 42.6358,
1328
+ 41.7049,
1329
+ 41.2221,
1330
+ 42.664,
1331
+ 42.1997,
1332
+ 41.5452,
1333
+ 41.4567,
1334
+ 42.0202,
1335
+ 42.1361,
1336
+ 42.5189,
1337
+ 42.3784,
1338
+ 42.2437,
1339
+ 42.2659,
1340
+ 41.7848,
1341
+ 43.1001,
1342
+ 42.9143,
1343
+ 42.6911,
1344
+ 43.3216,
1345
+ 42.2819,
1346
+ 42.6129,
1347
+ 42.6243,
1348
+ 42.7288,
1349
+ 42.2639,
1350
+ 42.0959,
1351
+ 42.1759,
1352
+ 43.1733,
1353
+ 42.9459,
1354
+ 43.8737,
1355
+ 43.3331,
1356
+ 43.5446,
1357
+ 43.8042,
1358
+ 43.0854,
1359
+ 43.4311,
1360
+ 43.5271,
1361
+ 42.906,
1362
+ 43.5409,
1363
+ 43.8748,
1364
+ 43.5499,
1365
+ 43.033,
1366
+ 43.7788,
1367
+ 43.6724,
1368
+ 43.4617,
1369
+ 43.5947,
1370
+ 44.13,
1371
+ 43.309,
1372
+ 43.5078,
1373
+ 43.1661,
1374
+ 43.372,
1375
+ 43.4169,
1376
+ 43.3339,
1377
+ 43.6628,
1378
+ 44.6899,
1379
+ 45.0728,
1380
+ 44.2652,
1381
+ 44.6406,
1382
+ 44.9743,
1383
+ 44.1587,
1384
+ 44.4485,
1385
+ 44.9894,
1386
+ 44.8086,
1387
+ 45.0493,
1388
+ 44.9517,
1389
+ 44.7992,
1390
+ 45.046,
1391
+ 44.2882,
1392
+ 44.9512,
1393
+ 45.3856,
1394
+ 45.1733,
1395
+ 46.024,
1396
+ 44.6306,
1397
+ 45.1784,
1398
+ 44.786,
1399
+ 44.1829,
1400
+ 45.4851,
1401
+ 44.9421,
1402
+ 45.1872,
1403
+ 44.8255,
1404
+ 45.2785,
1405
+ 45.0869,
1406
+ 44.9819,
1407
+ 44.3532,
1408
+ 44.8334,
1409
+ 45.161,
1410
+ 45.1502,
1411
+ 45.582,
1412
+ 45.2771,
1413
+ 44.8298,
1414
+ 45.2939,
1415
+ 45.5606,
1416
+ 46.0719,
1417
+ 45.5214,
1418
+ 46.1862,
1419
+ 45.4903,
1420
+ 45.8888,
1421
+ 45.181,
1422
+ 45.5755,
1423
+ 44.693,
1424
+ 45.944,
1425
+ 45.4897,
1426
+ 45.2955,
1427
+ 45.0721,
1428
+ 45.5201,
1429
+ 45.7761,
1430
+ 45.3752,
1431
+ 46.1321,
1432
+ 45.7825,
1433
+ 46.2748,
1434
+ 45.4683,
1435
+ 46.1643,
1436
+ 45.8866,
1437
+ 46.1443,
1438
+ 46.4845,
1439
+ 46.798,
1440
+ 46.0865,
1441
+ 46.6844,
1442
+ 45.9848,
1443
+ 46.0457,
1444
+ 45.5702,
1445
+ 45.0768
1446
+ ],
1447
+ "lr": [
1448
+ 2.24e-05,
1449
+ 4.48e-05,
1450
+ 6.720000000000001e-05,
1451
+ 8.96e-05,
1452
+ 0.000112,
1453
+ 0.00013440000000000001,
1454
+ 0.0001568,
1455
+ 0.0001792,
1456
+ 0.0002016,
1457
+ 0.000224,
1458
+ 0.00024639999999999997,
1459
+ 0.00026880000000000003,
1460
+ 0.0002912,
1461
+ 0.0003136,
1462
+ 0.000336,
1463
+ 0.0003584,
1464
+ 0.00038080000000000004,
1465
+ 0.0004032,
1466
+ 0.0004256,
1467
+ 0.000448,
1468
+ 0.00047040000000000005,
1469
+ 0.0004927999999999999,
1470
+ 0.0005152,
1471
+ 0.0005376000000000001,
1472
+ 0.0005600000000000001,
1473
+ 0.0005824,
1474
+ 0.0006048,
1475
+ 0.0006272,
1476
+ 0.0006496000000000001,
1477
+ 0.000672,
1478
+ 0.0006944,
1479
+ 0.000691748238161833,
1480
+ 0.0006811865761413326,
1481
+ 0.0006710943652414491,
1482
+ 0.0006614378277661476,
1483
+ 0.0006521864934374386,
1484
+ 0.0006433127943320818,
1485
+ 0.0006347917188462602,
1486
+ 0.0006266005147845038,
1487
+ 0.0006187184335382291,
1488
+ 0.000611126508798381,
1489
+ 0.0006038073644245599,
1490
+ 0.0005967450470366643,
1491
+ 0.0005899248796553369,
1492
+ 0.0005833333333333334,
1493
+ 0.0005769579142213136,
1494
+ 0.0005707870639217619,
1495
+ 0.000564810071321915,
1496
+ 0.0005590169943749475,
1497
+ 0.0005533985905294664,
1498
+ 0.0005479462546995058,
1499
+ 0.000542651963827787,
1500
+ 0.0005375082272297073,
1501
+ 0.0005325080420189115,
1502
+ 0.0005276448530110864,
1503
+ 0.0005229125165837972,
1504
+ 0.0005183052680392059,
1505
+ 0.0005138176920753656,
1506
+ 0.0005094446960221342,
1507
+ 0.0005051814855409226,
1508
+ 0.0005010235425246347,
1509
+ 0.0004969666049661817,
1510
+ 0.0004930066485916347,
1511
+ 0.000489139870078079,
1512
+ 0.00048536267169707543,
1513
+ 0.0004816716472427828,
1514
+ 0.0004780635691196339,
1515
+ 0.0004745353764783104,
1516
+ 0.00047108416430089486,
1517
+ 0.00046770717334674265,
1518
+ 0.00046440178087999085,
1519
+ 0.00046116549210788865,
1520
+ 0.00045799593226643695,
1521
+ 0.00045489083929628184,
1522
+ 0.00045184805705753196,
1523
+ 0.0004488655290372549,
1524
+ 0.0004459412925079224,
1525
+ 0.0004430734730991041,
1526
+ 0.00044026027974829917,
1527
+ 0.0004375,
1528
+ 0.0004347909956249591,
1529
+ 0.0004321316985341955,
1530
+ 0.000429520606964589,
1531
+ 0.00042695628191498326,
1532
+ 0.00042443734381358275,
1533
+ 0.00042196246939911063,
1534
+ 0.00041953038879971106,
1535
+ 0.0004171398827949466,
1536
+ 0.00041478978024748006,
1537
+ 0.0004124789556921527,
1538
+ 0.0004102063270711835,
1539
+ 0.00040797085360513727,
1540
+ 0.00040577153379014726,
1541
+ 0.0004036074035126372,
1542
+ 0.00040147753427348306,
1543
+ 0.0003993810315141836,
1544
+ 0.0003973170330381873,
1545
+ 0.0003952847075210474,
1546
+ 0.0003932832531035579,
1547
+ 0.0003913118960624632,
1548
+ 0.00038936988955373664,
1549
+ 0.0003874565124237917,
1550
+ 0.0003855710680843278,
1551
+ 0.0003837128834468252,
1552
+ 0.00038188130791298665,
1553
+ 0.0003800757124176857,
1554
+ 0.00037829548852122733,
1555
+ 0.0003765400475479433,
1556
+ 0.00037480881976835446,
1557
+ 0.0003731012536223183,
1558
+ 0.00037141681498075775,
1559
+ 0.00036975498644372604,
1560
+ 0.0003681152666727123,
1561
+ 0.0003664971697552336,
1562
+ 0.00036490022459988083,
1563
+ 0.0003633239743601124,
1564
+ 0.00036176797588519127,
1565
+ 0.0003602317991967705,
1566
+ 0.0003587150269897196,
1567
+ 0.00035721725415588016,
1568
+ 0.000355738087329512,
1569
+ 0.00035427714445327574,
1570
+ 0.00035283405436366147,
1571
+ 0.00035140845639484324,
1572
+ 0.00035,
1573
+ 0.00034860834438919813,
1574
+ 0.0003472331581829885,
1575
+ 0.0003458741190809165,
1576
+ 0.000344530913544194,
1577
+ 0.0003432032364918221,
1578
+ 0.00034189079100949653,
1579
+ 0.0003405932880706663,
1580
+ 0.00033931044626914805,
1581
+ 0.0003380419915627369,
1582
+ 0.00033678765702728166,
1583
+ 0.00033554718262072454,
1584
+ 0.00033432031495663064,
1585
+ 0.00033310680708676044,
1586
+ 0.00033190641829226067,
1587
+ 0.0003307189138830738,
1588
+ 0.00032954406500518534,
1589
+ 0.0003283816484553507,
1590
+ 0.00032723144650296,
1591
+ 0.0003260932467187193,
1592
+ 0.00032496684180984074,
1593
+ 0.0003238520294614523,
1594
+ 0.0003227486121839514,
1595
+ 0.0003216563971660409,
1596
+ 0.00032057519613319885,
1597
+ 0.0003195048252113469,
1598
+ 0.00031844510479549205,
1599
+ 0.0003173958594231301,
1600
+ 0.00031635691765220694,
1601
+ 0.00031532811194344564,
1602
+ 0.0003143092785468561,
1603
+ 0.0003133002573922519,
1604
+ 0.00031230089198360926,
1605
+ 0.0003113110292971088,
1606
+ 0.00031033051968271,
1607
+ 0.00030935921676911453,
1608
+ 0.0003083969773719806,
1609
+ 0.0003074436614052591,
1610
+ 0.0003064991317955257,
1611
+ 0.0003055632543991905,
1612
+ 0.0003046358979224712,
1613
+ 0.0003037169338440227,
1614
+ 0.0003028062363401174,
1615
+ 0.00030190368221227995,
1616
+ 0.0003010091508172794,
1617
+ 0.0003001225239993905,
1618
+ 0.0002992436860248366,
1619
+ 0.00029837252351833213,
1620
+ 0.00029750892540164505,
1621
+ 0.0002966527828341045,
1622
+ 0.0002958039891549808,
1623
+ 0.00029496243982766844,
1624
+ 0.00029412803238560624,
1625
+ 0.000293300666379871,
1626
+ 0.00029248024332838307,
1627
+ 0.0002916666666666667,
1628
+ 0.00029085984170010846,
1629
+ 0.0002900596755576607,
1630
+ 0.00028926607714693776,
1631
+ 0.0002884789571106568,
1632
+ 0.00028769822778437526,
1633
+ 0.00028692380315547937,
1634
+ 0.00028615559882338125,
1635
+ 0.00028539353196088094,
1636
+ 0.00028463752127665554,
1637
+ 0.00028388748697883446,
1638
+ 0.0002831433507396259,
1639
+ 0.0002824050356609575,
1640
+ 0.00028167246624109706,
1641
+ 0.0002809455683422218,
1642
+ 0.0002802242691589025,
1643
+ 0.00027950849718747374,
1644
+ 0.00027879818219626016,
1645
+ 0.00027809325519663106,
1646
+ 0.00027739364841485563,
1647
+ 0.0002766992952647332,
1648
+ 0.0002760101303209739,
1649
+ 0.00027532608929330425,
1650
+ 0.0002746471090012757,
1651
+ 0.0002739731273497529,
1652
+ 0.0002733040833050606,
1653
+ 0.00027263991687176815,
1654
+ 0.0002719805690700915,
1655
+ 0.0002713259819138935,
1656
+ 0.0002706760983892639,
1657
+ 0.0002700308624336609,
1658
+ 0.0002693902189155972,
1659
+ 0.00026875411361485365,
1660
+ 0.00026812249320320433,
1661
+ 0.00026749530522563754,
1662
+ 0.00026687249808205816,
1663
+ 0.00026625402100945573,
1664
+ 0.00026563982406452584,
1665
+ 0.00026502985810672995,
1666
+ 0.000264424074781781,
1667
+ 0.0002638224265055432,
1668
+ 0.00026322486644833176,
1669
+ 0.00026263134851960314,
1670
+ 0.0002620418273530224,
1671
+ 0.0002614562582918986,
1672
+ 0.00026087459737497546,
1673
+ 0.0002602968013225692,
1674
+ 0.0002597228275230421,
1675
+ 0.00025915263401960295,
1676
+ 0.0002585861794974245,
1677
+ 0.00025802342327106996,
1678
+ 0.0002574643252722186,
1679
+ 0.0002569088460376828,
1680
+ 0.00025635694669770835,
1681
+ 0.00025580858896455015,
1682
+ 0.0002552637351213155,
1683
+ 0.0002547223480110671,
1684
+ 0.0002541843910261805,
1685
+ 0.00025364982809794617,
1686
+ 0.0002531186236864126,
1687
+ 0.0002525907427704613,
1688
+ 0.0002520661508381089,
1689
+ 0.00025154481387703014,
1690
+ 0.00025102669836529556,
1691
+ 0.00025051177126231733,
1692
+ 0.00025,
1693
+ 0.00024949135247408796,
1694
+ 0.0002489857970357066,
1695
+ 0.00024848330248309085,
1696
+ 0.0002479838380534969,
1697
+ 0.00024748737341529165,
1698
+ 0.000246993878660216,
1699
+ 0.0002465033242958173,
1700
+ 0.00024601568123804583,
1701
+ 0.00024553092080401224,
1702
+ 0.00024504901470490174,
1703
+ 0.0002445699350390395,
1704
+ 0.00024409365428510593,
1705
+ 0.00024362014529549572,
1706
+ 0.00024314938128981866,
1707
+ 0.00024268133584853772,
1708
+ 0.00024221598290674156,
1709
+ 0.0002417532967480477,
1710
+ 0.00024129325199863348,
1711
+ 0.0002408358236213914,
1712
+ 0.0002403809869102058,
1713
+ 0.00023992871748434828,
1714
+ 0.0002394789912829883,
1715
+ 0.00023903178455981694,
1716
+ 0.0002385870738777802,
1717
+ 0.0002381448361039201,
1718
+ 0.00023770504840432054,
1719
+ 0.0002372676882391552,
1720
+ 0.00023683273335783548,
1721
+ 0.0002364001617942559,
1722
+ 0.0002359699518621347,
1723
+ 0.00023554208215044743,
1724
+ 0.00023511653151895082,
1725
+ 0.0002346932790937963,
1726
+ 0.0002342723042632288,
1727
+ 0.00023385358667337133,
1728
+ 0.00023343710622409107,
1729
+ 0.000233022843064947,
1730
+ 0.00023261077759121587,
1731
+ 0.00023220089043999543,
1732
+ 0.00023179316248638266,
1733
+ 0.00023138757483972598,
1734
+ 0.0002309841088399489
1735
+ ]
1736
+ }
metadata.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "Transformer (from scratch, Pre-LN, tied embeddings)",
3
+ "task": "Pali → Sinhala MT",
4
+ "vocab_size": 4000,
5
+ "spm_model_type": "unigram",
6
+ "d_model": 512,
7
+ "n_heads": 8,
8
+ "n_enc_layers": 5,
9
+ "n_dec_layers": 5,
10
+ "d_ff": 2048,
11
+ "dropout": 0.3,
12
+ "label_smoothing": 0.2,
13
+ "beam_size": 8,
14
+ "length_alpha": 0.6,
15
+ "test_bleu": 22.9565,
16
+ "test_chrf": 46.436,
17
+ "checkpoint": "averaged",
18
+ "val_bleu": 22.855
19
+ }
model_weights.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f2b6280352e04431e24cdf303f2421232dafa02e98a22716a9a586678f60c91a
3
+ size 155935495
sample_translations.png ADDED

Git LFS Details

  • SHA256: 6bb9b815faeeabb9eeec089cbc7cae2edd8cf73360bf782bfdbca5e9deaac5c7
  • Pointer size: 131 Bytes
  • Size of remote file: 524 kB
sample_translations_best.png ADDED

Git LFS Details

  • SHA256: e3ab59077b1ce19b246230f6425965486481783500e564bc83a3f8408b82a761
  • Pointer size: 131 Bytes
  • Size of remote file: 781 kB
spm.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2f6e67ad92c2b5066984619c5d9d93fd931ff15f3df0f1fa091544eff7bdbf06
3
+ size 353979
spm.vocab ADDED
The diff for this file is too large to render. See raw diff
 
test_analysis.png ADDED
test_results.csv ADDED
The diff for this file is too large to render. See raw diff
 
tokenization_analysis.png ADDED
training_report.png ADDED

Git LFS Details

  • SHA256: 9b055a7c75480fbf22ca6e8c0948ccfc14a6d859a8fb9215f741170ffa30ccf0
  • Pointer size: 131 Bytes
  • Size of remote file: 280 kB