After Change

Epoch 0:
Batch 0: EFE = -36.0038, CE = 9.3680, PPL = 11708.0449
Batch 10: EFE = -36.0043, CE = 9.3664, PPL = 11688.9795
Batch 20: EFE = -36.0049, CE = 9.3651, PPL = 11673.6387
Batch 30: EFE = -36.0049, CE = 9.3633, PPL = 11653.1172
Batch 40: EFE = -36.0052, CE = 9.3610, PPL = 11625.4668
Batch 50: EFE = -36.0062, CE = 9.3594, PPL = 11607.9619
Batch 60: EFE = -36.0082, CE = 9.3571, PPL = 11580.4961
Batch 70: EFE = -36.0097, CE = 9.3555, PPL = 11562.6406
Batch 80: EFE = -36.0123, CE = 9.3515, PPL = 11516.4531
Batch 90: EFE = -36.0120, CE = 9.3530, PPL = 11533.9180
Batch 100: EFE = -36.0179, CE = 9.3482, PPL = 11478.3506
Batch 110: EFE = -36.0203, CE = 9.3442, PPL = 11431.9102
Batch 120: EFE = -36.0253, CE = 9.3427, PPL = 11415.0244
Batch 130: EFE = -36.0284, CE = 9.3400, PPL = 11384.0947
Batch 140: EFE = -36.0303, CE = 9.3355, PPL = 11332.7715
Batch 150: EFE = -36.0379, CE = 9.3364, PPL = 11343.8779
Batch 160: EFE = -36.0465, CE = 9.3319, PPL = 11292.2715
Batch 170: EFE = -36.0520, CE = 9.3258, PPL = 11223.7627
Batch 180: EFE = -36.0597, CE = 9.3238, PPL = 11200.9229
Batch 190: EFE = -36.0606, CE = 9.3249, PPL = 11213.9727
Batch 200: EFE = -36.0837, CE = 9.3119, PPL = 11068.5283
Batch 210: EFE = -36.0863, CE = 9.3127, PPL = 11077.7900
Batch 220: EFE = -36.0948, CE = 9.3099, PPL = 11046.3828
Batch 230: EFE = -36.1154, CE = 9.3057, PPL = 11000.1914
Batch 240: EFE = -36.1308, CE = 9.2994, PPL = 10931.6602
Batch 250: EFE = -36.1341, CE = 9.3003, PPL = 10941.7773
Batch 260: EFE = -36.1523, CE = 9.2991, PPL = 10928.2520
Batch 270: EFE = -36.1841, CE = 9.2888, PPL = 10816.3652
Batch 280: EFE = -36.2266, CE = 9.2800, PPL = 10721.5723
Batch 290: EFE = -36.2163, CE = 9.2908, PPL = 10838.1826
Batch 300: EFE = -36.2557, CE = 9.2756, PPL = 10674.7217
Batch 310: EFE = -36.2997, CE = 9.2618, PPL = 10527.9844
Batch 320: EFE = -36.3285, CE = 9.2665, PPL = 10577.5195
Batch 330: EFE = -36.3446, CE = 9.2653, PPL = 10565.4414
Batch 340: EFE = -36.4261, CE = 9.2486, PPL = 10390.0889
Batch 350: EFE = -36.4724, CE = 9.2462, PPL = 10364.8125
Batch 360: EFE = -36.5088, CE = 9.2401, PPL = 10301.5576
Batch 370: EFE = -36.5918, CE = 9.2282, PPL = 10179.8789
Batch 380: EFE = -36.5884, CE = 9.2378, PPL = 10278.5156
 Batch 390: EFE = -36.6796, CE = 9.2221, PPL = 10118.1006
Batch 400: EFE = -36.7558, CE = 9.2158, PPL = 10054.4609
Batch 410: EFE = -36.8830, CE = 9.2198, PPL = 10095.3047
Batch 420: EFE = -36.8839, CE = 9.2138, PPL = 10034.4404
Batch 430: EFE = -36.9325, CE = 9.2301, PPL = 10199.8496
Batch 440: EFE = -37.0375, CE = 9.2066, PPL = 9962.3740
Batch 450: EFE = -37.0573, CE = 9.1989, PPL = 9886.2217
Batch 460: EFE = -37.1923, CE = 9.1877, PPL = 9776.0400
Batch 470: EFE = -37.3271, CE = 9.2010, PPL = 9907.0234
Batch 480: EFE = -37.4639, CE = 9.1861, PPL = 9760.4268
Batch 490: EFE = -37.5260, CE = 9.1828, PPL = 9728.6260
Batch 500: EFE = -37.6948, CE = 9.1653, PPL = 9559.1455
Batch 510: EFE = -37.6988, CE = 9.1624, PPL = 9532.2266
Batch 520: EFE = -37.9815, CE = 9.1657, PPL = 9563.3857
Batch 530: EFE = -38.0268, CE = 9.1731, PPL = 9634.4863
Batch 540: EFE = -38.2747, CE = 9.1404, PPL = 9324.1914
Batch 550: EFE = -38.2965, CE = 9.1656, PPL = 9562.2812
Batch 560: EFE = -38.4638, CE = 9.1594, PPL = 9503.3447
Batch 570: EFE = -38.4816, CE = 9.1507, PPL = 9421.3262
Batch 580: EFE = -38.8263, CE = 9.1407, PPL = 9327.6689
Batch 590: EFE = -39.1398, CE = 9.1134, PPL = 9076.4434
Batch 600: EFE = -39.6030, CE = 9.1109, PPL = 9053.7666
Batch 610: EFE = -39.7843, CE = 9.1105, PPL = 9050.1934
Batch 620: EFE = -39.7165, CE = 9.1333, PPL = 9258.5059
Batch 630: EFE = -40.4509, CE = 9.1119, PPL = 9062.1113
Batch 640: EFE = -40.6051, CE = 9.1254, PPL = 9185.7725
Batch 650: EFE = -40.8511, CE = 9.1248, PPL = 9179.8350
Batch 660: EFE = -41.1399, CE = 9.0959, PPL = 8918.7070
Batch 670: EFE = -41.3519, CE = 9.1279, PPL = 9208.9639
Batch 680: EFE = -41.6795, CE = 9.0637, PPL = 8635.7383
Batch 690: EFE = -41.8753, CE = 9.0609, PPL = 8611.5508

​
​
Before the Change ​
Epoch 0:
Batch 0: EFE = -0.0113, CE = 9.3667, PPL = 11692.5908
Batch 10: EFE = -0.0152, CE = 9.1084, PPL = 9031.0703
Batch 20: EFE = -0.0222, CE = 8.6934, PPL = 5963.6987
Batch 30: EFE = -0.0479, CE = 8.1123, PPL = 3335.2559
Batch 40: EFE = -0.0552, CE = 8.0690, PPL = 3194.0017
Batch 50: EFE = -0.1298, CE = 7.5283, PPL = 1859.9122
Batch 60: EFE = -0.1821, CE = 7.8294, PPL = 2513.4954
Batch 70: EFE = -0.2400, CE = 7.9055, PPL = 2712.1501
Batch 80: EFE = -0.4333, CE = 7.8995, PPL = 2696.0547
Batch 90: EFE = -0.8242, CE = 7.4210, PPL = 1670.6794
Batch 100: EFE = -1.0631, CE = 7.2459, PPL = 1402.3230
Batch 110: EFE = -1.1139, CE = 7.5383, PPL = 1878.5674
Batch 120: EFE = -0.8454, CE = 8.1745, PPL = 3549.2410
Batch 130: EFE = -1.4507, CE = 7.0658, PPL = 1171.2589
Batch 140: EFE = -1.5779, CE = 7.0789, PPL = 1186.6725
Batch 150: EFE = -1.4714, CE = 7.1637, PPL = 1291.7141
Batch 160: EFE = -2.2473, CE = 6.7651, PPL = 867.0670
Batch 170: EFE = -2.1659, CE = 7.0934, PPL = 1204.0420
Batch 180: EFE = -2.9229, CE = 6.5861, PPL = 724.9772
Batch 190: EFE = -3.1344, CE = 6.7816, PPL = 881.4965
Batch 200: EFE = -3.3949, CE = 6.5889, PPL = 726.9961
Batch 210: EFE = -2.5400, CE = 6.8465, PPL = 940.6009
Batch 220: EFE = -3.1061, CE = 6.4926, PPL = 660.2541
Batch 230: EFE = -3.2989, CE = 6.5086, PPL = 670.8707
Batch 240: EFE = -4.3740, CE = 6.2449, PPL = 515.3690
Batch 250: EFE = -4.6038, CE = 6.3564, PPL = 576.1429
Batch 260: EFE = -4.7351, CE = 6.2699, PPL = 528.4295
Batch 270: EFE = -4.8737, CE = 6.1577, PPL = 472.3288
Batch 280: EFE = -3.2842, CE = 5.8411, PPL = 344.1522
Batch 290: EFE = -4.4774, CE = 6.9072, PPL = 999.4352
Batch 300: EFE = -6.1603, CE = 5.7112, PPL = 302.2408
Batch 310: EFE = -5.2870, CE = 6.3120, PPL = 551.1460
Batch 320: EFE = -9.1360, CE = 6.0869, PPL = 440.0401
Batch 330: EFE = -5.0400, CE = 6.2122, PPL = 498.7918
Batch 340: EFE = -8.3071, CE = 5.7475, PPL = 313.4050
Batch 350: EFE = -5.2906, CE = 5.7943, PPL = 328.4274
Batch 360: EFE = -7.9397, CE = 5.5962, PPL = 269.4111
Batch 370: EFE = -9.2954, CE = 6.2623, PPL = 524.4162
Batch 380: EFE = -6.4545, CE = 5.9934, PPL = 400.7572
Batch 390: EFE = -6.7083, CE = 6.2717, PPL = 529.4030
Batch 400: EFE = -10.7119, CE = 5.2634, PPL = 193.1348
Batch 410: EFE = -6.8738, CE = 5.9100, PPL = 368.7169
Batch 420: EFE = -9.3912, CE = 5.7493, PPL = 313.9828
Batch 430: EFE = -6.5335, CE = 6.5048, PPL = 668.3358
Batch 440: EFE = -10.7935, CE = 5.6155, PPL = 274.6538
Batch 450: EFE = -6.7877, CE = 5.5058, PPL = 246.1243
Batch 460: EFE = -9.1250, CE = 5.6657, PPL = 288.7958
Batch 470: EFE = -9.9068, CE = 5.6510, PPL = 284.5788
Batch 480: EFE = -4.8380, CE = 6.3590, PPL = 577.6520
Batch 490: EFE = -7.0073, CE = 5.4001, PPL = 221.4204
Batch 500: EFE = -5.9315, CE = 5.7897, PPL = 326.9168
Batch 510: EFE = -6.7981, CE = 5.4834, PPL = 240.6567
Batch 520: EFE = -5.4262, CE = 6.2088, PPL = 497.1218
Batch 530: EFE = nan, CE = nan, PPL = nan
Batch 540: EFE = nan, CE = nan, PPL = nan
Batch 550: EFE = nan, CE = nan, PPL = nan
Batch 560: EFE = nan, CE = nan, PPL = nan
Batch 570: EFE = nan, CE = nan, PPL = nan
Batch 580: EFE = nan, CE = nan, PPL = nan
Batch 590: EFE = nan, CE = nan, PPL = nan
Batch 600: EFE = nan, CE = nan, PPL = nan
Batch 610: EFE = nan, CE = nan, PPL = nan
Batch 620: EFE = nan, CE = nan, PPL = nan
Batch 630: EFE = nan, CE = nan, PPL = nan
Batch 640: EFE = nan, CE = nan, PPL = nan
Batch 650: EFE = nan, CE = nan, PPL = nan
Batch 660: EFE = nan, CE = nan, PPL = nan
Batch 670: EFE = nan, CE = nan, PPL = nan
Batch 680: EFE = nan, CE = nan, PPL = nan