Skip to content

Instantly share code, notes, and snippets.

@taylanbil
Created July 20, 2020 19:57
Show Gist options
  • Select an option

  • Save taylanbil/62a1d2eca60ca6b86d906e94e088c347 to your computer and use it in GitHub Desktop.

Select an option

Save taylanbil/62a1d2eca60ca6b86d906e94e088c347 to your computer and use it in GitHub Desktop.
lining up the raw losses.
taylanbil@dlrm-gpu-8:~/kkissmart-fairseq$ paste <( grep RAWLOS gpu-repro.txt ) <( grep RAWLOS tpulog.txt )
RAWLOSS @ 100 tensor(1410.1447, device='cuda:0') RAWLOSS @ 100 tensor(1408.6381, device='xla:1')
RAWLOSS @ 200 tensor(1351.2732, device='cuda:0') RAWLOSS @ 200 tensor(1351.1304, device='xla:1')
RAWLOSS @ 300 tensor(2289.2922, device='cuda:0') RAWLOSS @ 300 tensor(2289.3340, device='xla:1')
RAWLOSS @ 400 tensor(1715.7347, device='cuda:0') RAWLOSS @ 400 tensor(1715.6587, device='xla:1')
RAWLOSS @ 500 tensor(2465.9827, device='cuda:0') RAWLOSS @ 500 tensor(2465.9453, device='xla:1')
RAWLOSS @ 600 tensor(2054.0337, device='cuda:0') RAWLOSS @ 600 tensor(2054.0742, device='xla:1')
RAWLOSS @ 700 tensor(1702.4202, device='cuda:0') RAWLOSS @ 700 tensor(1702.3367, device='xla:1')
RAWLOSS @ 800 tensor(1390.9583, device='cuda:0') RAWLOSS @ 800 tensor(1390.9158, device='xla:1')
RAWLOSS @ 900 tensor(1542.2812, device='cuda:0') RAWLOSS @ 900 tensor(1542.2075, device='xla:1')
RAWLOSS @ 1000 tensor(1371.7253, device='cuda:0') RAWLOSS @ 1000 tensor(1371.6144, device='xla:1')
RAWLOSS @ 1100 tensor(1693.6879, device='cuda:0') RAWLOSS @ 1100 tensor(1693.7521, device='xla:1')
RAWLOSS @ 1200 tensor(1228.1172, device='cuda:0') RAWLOSS @ 1200 tensor(1228.0624, device='xla:1')
RAWLOSS @ 1300 tensor(1328.1652, device='cuda:0') RAWLOSS @ 1300 tensor(1328.1467, device='xla:1')
RAWLOSS @ 1400 tensor(1641.3103, device='cuda:0') RAWLOSS @ 1400 tensor(1641.4343, device='xla:1')
RAWLOSS @ 1500 tensor(1422.3258, device='cuda:0') RAWLOSS @ 1500 tensor(1422.3253, device='xla:1')
RAWLOSS @ 1600 tensor(1923.9146, device='cuda:0') RAWLOSS @ 1600 tensor(1923.9608, device='xla:1')
RAWLOSS @ 1700 tensor(1688.3428, device='cuda:0') RAWLOSS @ 1700 tensor(1688.1075, device='xla:1')
RAWLOSS @ 1800 tensor(1570.4878, device='cuda:0') RAWLOSS @ 1800 tensor(1570.7827, device='xla:1')
RAWLOSS @ 1900 tensor(1828.7078, device='cuda:0') RAWLOSS @ 1900 tensor(1828.8079, device='xla:1')
RAWLOSS @ 2000 tensor(1782.3804, device='cuda:0') RAWLOSS @ 2000 tensor(1782.3601, device='xla:1')
RAWLOSS @ 2100 tensor(2344.2183, device='cuda:0') RAWLOSS @ 2100 tensor(2344.3152, device='xla:1')
RAWLOSS @ 2200 tensor(1709.7332, device='cuda:0') RAWLOSS @ 2200 tensor(1709.8677, device='xla:1')
RAWLOSS @ 2300 tensor(1072.0853, device='cuda:0') RAWLOSS @ 2300 tensor(1072.0686, device='xla:1')
RAWLOSS @ 2400 tensor(1800.6949, device='cuda:0') RAWLOSS @ 2400 tensor(1801.0403, device='xla:1')
RAWLOSS @ 2500 tensor(1392.2799, device='cuda:0') RAWLOSS @ 2500 tensor(1392.2241, device='xla:1')
RAWLOSS @ 2600 tensor(1319.8795, device='cuda:0') RAWLOSS @ 2600 tensor(1319.8892, device='xla:1')
RAWLOSS @ 2700 tensor(1346.2390, device='cuda:0') RAWLOSS @ 2700 tensor(1346.2131, device='xla:1')
RAWLOSS @ 2800 tensor(1433.3136, device='cuda:0') RAWLOSS @ 2800 tensor(1433.2075, device='xla:1')
RAWLOSS @ 2900 tensor(1693.1978, device='cuda:0') RAWLOSS @ 2900 tensor(1693.4897, device='xla:1')
RAWLOSS @ 3000 tensor(867.2125, device='cuda:0') RAWLOSS @ 3000 tensor(867.6174, device='xla:1')
RAWLOSS @ 3100 tensor(1706.2174, device='cuda:0') RAWLOSS @ 3100 tensor(1705.9694, device='xla:1')
RAWLOSS @ 3200 tensor(1676.3699, device='cuda:0') RAWLOSS @ 3200 tensor(1676.2924, device='xla:1')
RAWLOSS @ 3300 tensor(1000.9597, device='cuda:0') RAWLOSS @ 3300 tensor(1001.3185, device='xla:1')
RAWLOSS @ 3400 tensor(1666.6620, device='cuda:0') RAWLOSS @ 3400 tensor(1667.1111, device='xla:1')
RAWLOSS @ 3500 tensor(1426.3322, device='cuda:0') RAWLOSS @ 3500 tensor(1426.0515, device='xla:1')
RAWLOSS @ 3600 tensor(1634.4712, device='cuda:0') RAWLOSS @ 3600 tensor(1634.5762, device='xla:1')
RAWLOSS @ 3700 tensor(1551.6877, device='cuda:0') RAWLOSS @ 3700 tensor(1552.2892, device='xla:1')
RAWLOSS @ 3800 tensor(1543.4656, device='cuda:0') RAWLOSS @ 3800 tensor(1543.8973, device='xla:1')
RAWLOSS @ 3900 tensor(1488.4733, device='cuda:0') RAWLOSS @ 3900 tensor(1488.0979, device='xla:1')
RAWLOSS @ 4000 tensor(1679.9448, device='cuda:0') RAWLOSS @ 4000 tensor(1680.7052, device='xla:1')
RAWLOSS @ 4100 tensor(1940.9502, device='cuda:0') RAWLOSS @ 4100 tensor(1941.4369, device='xla:1')
RAWLOSS @ 4200 tensor(1738.5826, device='cuda:0') RAWLOSS @ 4200 tensor(1739.0806, device='xla:1')
RAWLOSS @ 4300 tensor(1848.1174, device='cuda:0') RAWLOSS @ 4300 tensor(1847.4691, device='xla:1')
RAWLOSS @ 4400 tensor(2160.4170, device='cuda:0') RAWLOSS @ 4400 tensor(2160.4172, device='xla:1')
RAWLOSS @ 4500 tensor(1741.8374, device='cuda:0') RAWLOSS @ 4500 tensor(1743.6643, device='xla:1')
RAWLOSS @ 4600 tensor(1013.6898, device='cuda:0') RAWLOSS @ 4600 tensor(1013.4943, device='xla:1')
RAWLOSS @ 4700 tensor(1215.4442, device='cuda:0') RAWLOSS @ 4700 tensor(1215.8892, device='xla:1')
RAWLOSS @ 4800 tensor(1686.5798, device='cuda:0') RAWLOSS @ 4800 tensor(1687.0696, device='xla:1')
RAWLOSS @ 4900 tensor(1270.6083, device='cuda:0') RAWLOSS @ 4900 tensor(1267.5745, device='xla:1')
RAWLOSS @ 5000 tensor(1934.5712, device='cuda:0') RAWLOSS @ 5000 tensor(1933.2743, device='xla:1')
RAWLOSS @ 5100 tensor(1389.7025, device='cuda:0') RAWLOSS @ 5100 tensor(1389.2589, device='xla:1')
RAWLOSS @ 5200 tensor(1521.7289, device='cuda:0') RAWLOSS @ 5200 tensor(1521.3794, device='xla:1')
RAWLOSS @ 5300 tensor(986.0732, device='cuda:0') RAWLOSS @ 5300 tensor(985.2541, device='xla:1')
RAWLOSS @ 5400 tensor(1238.0280, device='cuda:0') RAWLOSS @ 5400 tensor(1239.3900, device='xla:1')
RAWLOSS @ 5500 tensor(1070.5186, device='cuda:0') RAWLOSS @ 5500 tensor(1070.6313, device='xla:1')
RAWLOSS @ 5600 tensor(2137.5652, device='cuda:0') RAWLOSS @ 5600 tensor(2136.5396, device='xla:1')
RAWLOSS @ 5700 tensor(1708.1031, device='cuda:0') RAWLOSS @ 5700 tensor(1710.4930, device='xla:1')
RAWLOSS @ 5800 tensor(1423.8556, device='cuda:0') RAWLOSS @ 5800 tensor(1422.4799, device='xla:1')
RAWLOSS @ 5900 tensor(1369.5192, device='cuda:0') RAWLOSS @ 5900 tensor(1369.4492, device='xla:1')
RAWLOSS @ 6000 tensor(1245.5112, device='cuda:0') RAWLOSS @ 6000 tensor(1244.3560, device='xla:1')
RAWLOSS @ 6100 tensor(2512.7710, device='cuda:0') RAWLOSS @ 6100 tensor(2512.4678, device='xla:1')
RAWLOSS @ 6200 tensor(1891.8496, device='cuda:0') RAWLOSS @ 6200 tensor(1890.8009, device='xla:1')
RAWLOSS @ 6300 tensor(1130.3253, device='cuda:0') RAWLOSS @ 6300 tensor(1129.4983, device='xla:1')
RAWLOSS @ 6400 tensor(961.2222, device='cuda:0') RAWLOSS @ 6400 tensor(961.3102, device='xla:1')
RAWLOSS @ 6500 tensor(1826.8185, device='cuda:0') RAWLOSS @ 6500 tensor(1826.2983, device='xla:1')
RAWLOSS @ 6600 tensor(668.8315, device='cuda:0') RAWLOSS @ 6600 tensor(669.0168, device='xla:1')
RAWLOSS @ 6700 tensor(1368.3977, device='cuda:0') RAWLOSS @ 6700 tensor(1369.3162, device='xla:1')
RAWLOSS @ 6800 tensor(2059.0720, device='cuda:0') RAWLOSS @ 6800 tensor(2058.0190, device='xla:1')
RAWLOSS @ 6900 tensor(1417.4907, device='cuda:0') RAWLOSS @ 6900 tensor(1416.5652, device='xla:1')
RAWLOSS @ 7000 tensor(1296.0875, device='cuda:0') RAWLOSS @ 7000 tensor(1297.1597, device='xla:1')
RAWLOSS @ 7100 tensor(1533.7075, device='cuda:0') RAWLOSS @ 7100 tensor(1532.3556, device='xla:1')
RAWLOSS @ 7200 tensor(1195.6879, device='cuda:0') RAWLOSS @ 7200 tensor(1194.9662, device='xla:1')
RAWLOSS @ 7300 tensor(1629.7560, device='cuda:0') RAWLOSS @ 7300 tensor(1631.6942, device='xla:1')
RAWLOSS @ 7400 tensor(1123.7363, device='cuda:0') RAWLOSS @ 7400 tensor(1123.8306, device='xla:1')
RAWLOSS @ 7500 tensor(1551.4454, device='cuda:0') RAWLOSS @ 7500 tensor(1551.5527, device='xla:1')
RAWLOSS @ 7600 tensor(1700.6863, device='cuda:0') RAWLOSS @ 7600 tensor(1701.3202, device='xla:1')
RAWLOSS @ 7700 tensor(1107.6963, device='cuda:0') RAWLOSS @ 7700 tensor(1108.4896, device='xla:1')
RAWLOSS @ 7800 tensor(1450.6442, device='cuda:0') RAWLOSS @ 7800 tensor(1451.0496, device='xla:1')
RAWLOSS @ 7900 tensor(1845.3767, device='cuda:0') RAWLOSS @ 7900 tensor(1845.2185, device='xla:1')
RAWLOSS @ 8000 tensor(1751.1384, device='cuda:0') RAWLOSS @ 8000 tensor(1752.1614, device='xla:1')
RAWLOSS @ 8100 tensor(1534.0872, device='cuda:0') RAWLOSS @ 8100 tensor(1533.9989, device='xla:1')
RAWLOSS @ 8200 tensor(1527.2074, device='cuda:0') RAWLOSS @ 8200 tensor(1526.4739, device='xla:1')
RAWLOSS @ 8300 tensor(2236.2222, device='cuda:0') RAWLOSS @ 8300 tensor(2234.0649, device='xla:1')
RAWLOSS @ 8400 tensor(1125.7655, device='cuda:0') RAWLOSS @ 8400 tensor(1127.3406, device='xla:1')
RAWLOSS @ 8500 tensor(1681.7218, device='cuda:0') RAWLOSS @ 8500 tensor(1681.1058, device='xla:1')
RAWLOSS @ 8600 tensor(1828.4143, device='cuda:0') RAWLOSS @ 8600 tensor(1827.9142, device='xla:1')
RAWLOSS @ 8700 tensor(1644.8458, device='cuda:0') RAWLOSS @ 8700 tensor(1644.2002, device='xla:1')
RAWLOSS @ 8800 tensor(1691.0027, device='cuda:0') RAWLOSS @ 8800 tensor(1690.3883, device='xla:1')
RAWLOSS @ 8900 tensor(1959.2830, device='cuda:0') RAWLOSS @ 8900 tensor(1959.0852, device='xla:1')
RAWLOSS @ 9000 tensor(1217.1136, device='cuda:0') RAWLOSS @ 9000 tensor(1217.4624, device='xla:1')
RAWLOSS @ 9100 tensor(1144.0026, device='cuda:0') RAWLOSS @ 9100 tensor(1144.5541, device='xla:1')
RAWLOSS @ 9200 tensor(1399.2024, device='cuda:0') RAWLOSS @ 9200 tensor(1397.6129, device='xla:1')
RAWLOSS @ 9300 tensor(1251.1783, device='cuda:0') RAWLOSS @ 9300 tensor(1250.4963, device='xla:1')
RAWLOSS @ 9400 tensor(1406.9491, device='cuda:0') RAWLOSS @ 9400 tensor(1405.1880, device='xla:1')
RAWLOSS @ 9500 tensor(937.4207, device='cuda:0') RAWLOSS @ 9500 tensor(938.4954, device='xla:1')
RAWLOSS @ 9600 tensor(1411.9232, device='cuda:0') RAWLOSS @ 9600 tensor(1413.3824, device='xla:1')
RAWLOSS @ 9700 tensor(1632.3060, device='cuda:0') RAWLOSS @ 9700 tensor(1633.5691, device='xla:1')
RAWLOSS @ 9800 tensor(2037.6841, device='cuda:0') RAWLOSS @ 9800 tensor(2035.8236, device='xla:1')
RAWLOSS @ 9900 tensor(1467.7316, device='cuda:0') RAWLOSS @ 9900 tensor(1467.5835, device='xla:1')
RAWLOSS @ 10000 tensor(1566.2866, device='cuda:0') RAWLOSS @ 10000 tensor(1565.9857, device='xla:1')
RAWLOSS @ 10100 tensor(1275.3107, device='cuda:0') RAWLOSS @ 10100 tensor(1274.9399, device='xla:1')
RAWLOSS @ 10200 tensor(1482.7819, device='cuda:0') RAWLOSS @ 10200 tensor(1482.6823, device='xla:1')
RAWLOSS @ 10300 tensor(1025.0098, device='cuda:0') RAWLOSS @ 10300 tensor(1025.6306, device='xla:1')
RAWLOSS @ 10400 tensor(1641.3466, device='cuda:0') RAWLOSS @ 10400 tensor(1639.2822, device='xla:1')
RAWLOSS @ 10500 tensor(1740.3737, device='cuda:0') RAWLOSS @ 10500 tensor(1739.4558, device='xla:1')
RAWLOSS @ 10600 tensor(1478.7198, device='cuda:0') RAWLOSS @ 10600 tensor(1478.3591, device='xla:1')
RAWLOSS @ 10700 tensor(1440.9719, device='cuda:0') RAWLOSS @ 10700 tensor(1441.7552, device='xla:1')
RAWLOSS @ 10800 tensor(1178.1041, device='cuda:0') RAWLOSS @ 10800 tensor(1178.0557, device='xla:1')
RAWLOSS @ 10900 tensor(1759.4432, device='cuda:0') RAWLOSS @ 10900 tensor(1758.1354, device='xla:1')
RAWLOSS @ 11000 tensor(2279.4111, device='cuda:0') RAWLOSS @ 11000 tensor(2277.9707, device='xla:1')
RAWLOSS @ 11100 tensor(1960.0864, device='cuda:0') RAWLOSS @ 11100 tensor(1958.7487, device='xla:1')
RAWLOSS @ 11200 tensor(1575.8950, device='cuda:0') RAWLOSS @ 11200 tensor(1572.9771, device='xla:1')
RAWLOSS @ 11300 tensor(1673.9479, device='cuda:0') RAWLOSS @ 11300 tensor(1671.0051, device='xla:1')
RAWLOSS @ 11400 tensor(1403.0437, device='cuda:0') RAWLOSS @ 11400 tensor(1405.1251, device='xla:1')
RAWLOSS @ 11500 tensor(2001.8221, device='cuda:0') RAWLOSS @ 11500 tensor(2001.3026, device='xla:1')
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment