KOLM-Alpha / train_native_tf.log
Dbrent's picture
add train_native_tf.log
94eb60b verified
Raw
History Blame Contribute Delete
11 kB
transformer | 17,440,512 params | vocab 8192 | mps
step 1 | loss 9.0659 | gnorm 5.66 | 1s
step 25 | loss 8.4237 | gnorm 2.84 | 21s
step 50 | loss 8.0053 | gnorm 2.37 | 36s
step 75 | loss 7.5894 | gnorm 2.29 | 51s
step 100 | loss 7.2227 | gnorm 1.95 | 66s
step 125 | loss 6.5297 | gnorm 1.79 | 83s
step 150 | loss 6.0340 | gnorm 1.76 | 99s
step 175 | loss 5.3586 | gnorm 1.44 | 117s
step 200 | loss 5.1159 | gnorm 1.48 | 133s
step 225 | loss 5.0118 | gnorm 1.45 | 148s
step 250 | loss 4.6414 | gnorm 1.30 | 163s
val 4.7656 | ppl 117.40 | 1M tokens
step 275 | loss 4.6810 | gnorm 1.50 | 182s
step 300 | loss 4.5977 | gnorm 1.46 | 197s
step 325 | loss 4.3595 | gnorm 1.41 | 212s
step 350 | loss 4.4616 | gnorm 1.52 | 228s
step 375 | loss 4.0337 | gnorm 1.43 | 244s
step 400 | loss 4.1460 | gnorm 1.46 | 259s
step 425 | loss 4.1583 | gnorm 1.43 | 275s
step 450 | loss 4.2407 | gnorm 1.49 | 290s
step 475 | loss 4.1234 | gnorm 1.59 | 305s
step 500 | loss 4.1939 | gnorm 1.50 | 320s
val 4.0422 | ppl 56.95 | 1M tokens
step 525 | loss 3.9165 | gnorm 1.65 | 338s
step 550 | loss 4.0645 | gnorm 1.59 | 353s
step 575 | loss 3.9202 | gnorm 1.70 | 369s
step 600 | loss 3.7699 | gnorm 1.48 | 384s
step 625 | loss 3.6540 | gnorm 1.65 | 399s
step 650 | loss 3.8476 | gnorm 1.67 | 414s
step 675 | loss 3.7533 | gnorm 1.69 | 430s
step 700 | loss 4.0363 | gnorm 1.59 | 446s
step 725 | loss 3.8794 | gnorm 1.61 | 462s
step 750 | loss 3.7619 | gnorm 1.65 | 478s
val 3.7400 | ppl 42.10 | 2M tokens
step 775 | loss 3.6882 | gnorm 1.73 | 498s
step 800 | loss 3.7385 | gnorm 1.74 | 515s
step 825 | loss 3.4740 | gnorm 1.78 | 531s
step 850 | loss 3.6808 | gnorm 1.80 | 547s
step 875 | loss 3.6482 | gnorm 1.77 | 562s
step 900 | loss 3.4743 | gnorm 1.75 | 577s
step 925 | loss 3.6641 | gnorm 1.71 | 593s
step 950 | loss 3.4985 | gnorm 1.95 | 608s
step 975 | loss 3.4978 | gnorm 1.86 | 623s
step 1000 | loss 3.6681 | gnorm 1.93 | 638s
val 3.5336 | ppl 34.25 | 2M tokens
step 1025 | loss 3.3692 | gnorm 1.85 | 656s
step 1050 | loss 3.7098 | gnorm 1.80 | 674s
step 1075 | loss 3.4980 | gnorm 1.89 | 692s
step 1100 | loss 3.6340 | gnorm 1.88 | 708s
step 1125 | loss 3.4360 | gnorm 1.74 | 727s
step 1150 | loss 3.2896 | gnorm 1.88 | 744s
step 1175 | loss 3.6475 | gnorm 2.15 | 765s
step 1200 | loss 3.4324 | gnorm 2.00 | 784s
step 1225 | loss 3.3737 | gnorm 1.80 | 802s
step 1250 | loss 3.4515 | gnorm 1.92 | 818s
val 3.3742 | ppl 29.20 | 3M tokens
step 1275 | loss 3.3330 | gnorm 1.96 | 837s
step 1300 | loss 3.2365 | gnorm 1.89 | 853s
step 1325 | loss 3.1443 | gnorm 1.79 | 869s
step 1350 | loss 3.0539 | gnorm 1.84 | 885s
step 1375 | loss 3.1282 | gnorm 1.84 | 901s
step 1400 | loss 3.3304 | gnorm 1.85 | 917s
step 1425 | loss 3.4560 | gnorm 2.02 | 934s
step 1450 | loss 3.0201 | gnorm 1.82 | 953s
step 1475 | loss 3.2094 | gnorm 1.87 | 972s
step 1500 | loss 3.1902 | gnorm 2.00 | 992s
val 3.2581 | ppl 26.00 | 3M tokens
step 1525 | loss 3.4144 | gnorm 1.97 | 1012s
step 1550 | loss 2.9986 | gnorm 1.92 | 1027s
step 1575 | loss 3.1162 | gnorm 1.97 | 1045s
step 1600 | loss 2.9763 | gnorm 1.97 | 1061s
step 1625 | loss 3.2165 | gnorm 1.76 | 1076s
step 1650 | loss 3.2712 | gnorm 1.91 | 1090s
step 1675 | loss 3.0484 | gnorm 2.04 | 1107s
step 1700 | loss 3.1147 | gnorm 2.01 | 1122s
step 1725 | loss 3.0797 | gnorm 1.95 | 1140s
step 1750 | loss 3.1958 | gnorm 1.92 | 1155s
val 3.1551 | ppl 23.46 | 4M tokens
step 1775 | loss 3.2327 | gnorm 1.99 | 1176s
step 1800 | loss 3.1617 | gnorm 2.09 | 1190s
step 1825 | loss 3.2977 | gnorm 1.95 | 1207s
step 1850 | loss 3.1577 | gnorm 2.26 | 1224s
step 1875 | loss 3.3915 | gnorm 2.02 | 1239s
step 1900 | loss 3.3305 | gnorm 1.93 | 1254s
step 1925 | loss 2.9534 | gnorm 2.16 | 1269s
step 1950 | loss 3.1981 | gnorm 2.04 | 1284s
step 1975 | loss 2.8632 | gnorm 1.96 | 1299s
step 2000 | loss 3.3176 | gnorm 2.07 | 1314s
val 3.0857 | ppl 21.88 | 4M tokens
step 2025 | loss 2.8043 | gnorm 1.89 | 1332s
step 2050 | loss 2.7409 | gnorm 2.03 | 1347s
step 2075 | loss 3.0667 | gnorm 2.06 | 1363s
step 2100 | loss 3.2567 | gnorm 2.02 | 1377s
step 2125 | loss 3.1960 | gnorm 2.09 | 1392s
step 2150 | loss 2.9177 | gnorm 2.08 | 1407s
step 2175 | loss 3.0648 | gnorm 2.19 | 1422s
step 2200 | loss 3.1265 | gnorm 2.02 | 1438s
step 2225 | loss 3.0281 | gnorm 2.10 | 1453s
step 2250 | loss 3.0515 | gnorm 1.95 | 1468s
val 3.0140 | ppl 20.37 | 5M tokens
step 2275 | loss 3.2566 | gnorm 2.13 | 1486s
step 2300 | loss 2.8999 | gnorm 2.08 | 1505s
step 2325 | loss 3.0517 | gnorm 2.23 | 1524s
step 2350 | loss 2.9837 | gnorm 2.06 | 1539s
step 2375 | loss 2.9317 | gnorm 2.04 | 1555s
step 2400 | loss 3.0361 | gnorm 2.26 | 1571s
step 2425 | loss 3.0551 | gnorm 2.24 | 1586s
step 2450 | loss 2.8550 | gnorm 2.21 | 1601s
step 2475 | loss 2.9036 | gnorm 2.09 | 1617s
step 2500 | loss 3.1739 | gnorm 2.19 | 1633s
val 2.9613 | ppl 19.32 | 5M tokens
step 2525 | loss 2.7797 | gnorm 2.35 | 1652s
step 2550 | loss 2.8502 | gnorm 2.07 | 1668s
step 2575 | loss 3.1404 | gnorm 2.30 | 1683s
step 2600 | loss 2.8394 | gnorm 2.22 | 1700s
step 2625 | loss 2.8314 | gnorm 2.18 | 1717s
step 2650 | loss 3.0105 | gnorm 2.20 | 1733s
step 2675 | loss 2.9467 | gnorm 2.24 | 1748s
step 2700 | loss 3.0323 | gnorm 2.19 | 1764s
step 2725 | loss 2.7853 | gnorm 2.20 | 1779s
step 2750 | loss 3.0100 | gnorm 2.21 | 1794s
val 2.9078 | ppl 18.32 | 6M tokens
step 2775 | loss 2.8640 | gnorm 2.03 | 1812s
step 2800 | loss 3.0415 | gnorm 2.24 | 1827s
step 2825 | loss 3.0081 | gnorm 2.26 | 1843s
step 2850 | loss 2.9286 | gnorm 2.36 | 1858s
step 2875 | loss 3.0645 | gnorm 2.04 | 1873s
step 2900 | loss 2.8814 | gnorm 2.10 | 1888s
step 2925 | loss 2.9102 | gnorm 2.16 | 1903s
step 2950 | loss 2.9170 | gnorm 2.10 | 1918s
step 2975 | loss 2.7310 | gnorm 2.13 | 1933s
step 3000 | loss 2.8980 | gnorm 2.19 | 1948s
val 2.8735 | ppl 17.70 | 6M tokens
step 3025 | loss 3.1517 | gnorm 2.18 | 1966s
step 3050 | loss 2.8395 | gnorm 2.21 | 1982s
step 3075 | loss 2.6157 | gnorm 2.09 | 1997s
step 3100 | loss 2.7318 | gnorm 2.17 | 2012s
step 3125 | loss 2.7952 | gnorm 2.28 | 2027s
step 3150 | loss 2.8145 | gnorm 2.35 | 2042s
step 3175 | loss 3.0589 | gnorm 2.37 | 2057s
step 3200 | loss 3.0978 | gnorm 2.40 | 2072s
step 3225 | loss 3.0286 | gnorm 2.26 | 2087s
step 3250 | loss 2.9351 | gnorm 2.35 | 2102s
val 2.8314 | ppl 16.97 | 7M tokens
step 3275 | loss 2.7162 | gnorm 2.53 | 2121s
step 3300 | loss 2.7243 | gnorm 2.38 | 2136s
step 3325 | loss 2.7972 | gnorm 2.46 | 2151s
step 3350 | loss 3.0039 | gnorm 2.37 | 2166s
step 3375 | loss 3.0588 | gnorm 2.30 | 2181s
step 3400 | loss 2.8216 | gnorm 2.14 | 2196s
step 3425 | loss 2.8228 | gnorm 2.26 | 2211s
step 3450 | loss 2.8083 | gnorm 2.28 | 2227s
step 3475 | loss 2.9651 | gnorm 2.26 | 2242s
step 3500 | loss 2.9724 | gnorm 2.36 | 2257s
val 2.8072 | ppl 16.56 | 7M tokens
step 3525 | loss 2.9245 | gnorm 2.44 | 2276s
step 3550 | loss 2.8749 | gnorm 2.23 | 2291s
step 3575 | loss 2.6986 | gnorm 2.35 | 2306s
step 3600 | loss 2.5832 | gnorm 2.27 | 2321s
step 3625 | loss 2.6425 | gnorm 2.26 | 2336s
step 3650 | loss 2.7245 | gnorm 2.27 | 2351s
step 3675 | loss 2.5597 | gnorm 2.29 | 2366s
step 3700 | loss 2.9487 | gnorm 2.20 | 2381s
step 3725 | loss 2.8018 | gnorm 2.26 | 2396s
step 3750 | loss 2.8344 | gnorm 2.32 | 2411s
val 2.7824 | ppl 16.16 | 8M tokens
step 3775 | loss 2.7982 | gnorm 2.31 | 2429s
step 3800 | loss 2.8159 | gnorm 2.21 | 2444s
step 3825 | loss 2.7541 | gnorm 2.27 | 2459s
step 3850 | loss 2.6004 | gnorm 2.32 | 2475s
step 3875 | loss 2.6492 | gnorm 2.30 | 2490s
step 3900 | loss 2.5851 | gnorm 2.19 | 2505s
step 3925 | loss 2.4567 | gnorm 2.39 | 2520s
step 3950 | loss 2.8217 | gnorm 2.25 | 2535s
step 3975 | loss 2.8198 | gnorm 2.31 | 2550s
step 4000 | loss 2.7746 | gnorm 2.45 | 2566s
val 2.7633 | ppl 15.85 | 8M tokens
step 4025 | loss 2.9212 | gnorm 2.43 | 2584s
step 4050 | loss 2.6847 | gnorm 2.25 | 2599s
step 4075 | loss 2.8598 | gnorm 2.36 | 2615s
step 4100 | loss 2.7371 | gnorm 2.29 | 2630s
step 4125 | loss 2.7010 | gnorm 2.60 | 2645s
step 4150 | loss 2.6612 | gnorm 2.29 | 2660s
step 4175 | loss 2.7117 | gnorm 2.47 | 2676s
step 4200 | loss 2.7607 | gnorm 2.30 | 2691s
step 4225 | loss 2.8989 | gnorm 2.57 | 2706s
step 4250 | loss 2.8335 | gnorm 2.48 | 2721s
val 2.7466 | ppl 15.59 | 9M tokens
step 4275 | loss 2.7723 | gnorm 2.65 | 2740s
step 4300 | loss 2.6089 | gnorm 2.32 | 2755s
step 4325 | loss 2.6483 | gnorm 2.29 | 2770s
step 4350 | loss 2.5808 | gnorm 2.36 | 2785s
step 4375 | loss 3.1040 | gnorm 2.95 | 2800s
step 4400 | loss 2.9378 | gnorm 2.48 | 2815s
step 4425 | loss 2.8229 | gnorm 2.37 | 2830s
step 4450 | loss 2.8420 | gnorm 2.31 | 2846s
step 4475 | loss 2.4412 | gnorm 2.31 | 2861s
step 4500 | loss 2.5466 | gnorm 2.40 | 2877s
val 2.7340 | ppl 15.39 | 9M tokens
step 4525 | loss 2.6072 | gnorm 2.36 | 2895s
step 4550 | loss 2.6989 | gnorm 2.46 | 2911s
step 4575 | loss 2.7203 | gnorm 2.41 | 2927s
step 4600 | loss 2.7147 | gnorm 2.43 | 2944s
step 4625 | loss 2.8740 | gnorm 2.34 | 2960s
step 4650 | loss 2.7108 | gnorm 2.37 | 2977s
step 4675 | loss 3.1500 | gnorm 2.43 | 2993s
step 4700 | loss 2.7334 | gnorm 2.59 | 3008s
step 4725 | loss 2.8229 | gnorm 2.41 | 3023s
step 4750 | loss 2.6176 | gnorm 2.33 | 3038s
val 2.7261 | ppl 15.27 | 10M tokens
step 4775 | loss 2.6204 | gnorm 2.28 | 3056s
step 4800 | loss 2.9064 | gnorm 2.38 | 3071s
step 4825 | loss 2.6115 | gnorm 2.50 | 3086s
step 4850 | loss 2.8141 | gnorm 2.48 | 3101s
step 4875 | loss 2.9632 | gnorm 2.51 | 3116s
step 4900 | loss 2.5242 | gnorm 2.53 | 3131s
step 4925 | loss 3.0410 | gnorm 2.56 | 3146s
step 4950 | loss 2.6725 | gnorm 2.47 | 3161s
step 4975 | loss 2.6360 | gnorm 2.42 | 3176s
step 5000 | loss 2.8666 | gnorm 2.64 | 3191s
val 2.7214 | ppl 15.20 | 10M tokens
step 5025 | loss 2.7423 | gnorm 2.26 | 3209s
step 5050 | loss 2.6553 | gnorm 2.43 | 3224s
step 5075 | loss 2.6794 | gnorm 2.35 | 3239s
step 5100 | loss 2.7391 | gnorm 2.38 | 3254s
step 5125 | loss 2.9630 | gnorm 2.35 | 3270s
step 5150 | loss 2.8359 | gnorm 2.27 | 3285s
step 5175 | loss 2.7692 | gnorm 2.51 | 3301s
step 5200 | loss 2.7218 | gnorm 2.24 | 3316s
step 5225 | loss 2.6277 | gnorm 2.44 | 3332s
step 5250 | loss 3.1225 | gnorm 2.51 | 3347s
val 2.7194 | ppl 15.17 | 11M tokens
step 5275 | loss 2.8204 | gnorm 2.35 | 3367s
step 5300 | loss 2.8327 | gnorm 2.34 | 3384s
step 5325 | loss 2.5840 | gnorm 2.23 | 3399s
step 5350 | loss 2.7542 | gnorm 2.36 | 3414s
step 5375 | loss 2.7365 | gnorm 2.32 | 3428s
step 5400 | loss 2.8519 | gnorm 2.33 | 3443s
step 5425 | loss 2.8636 | gnorm 2.54 | 3457s
step 5450 | loss 2.6134 | gnorm 2.36 | 3471s
step 5475 | loss 2.7233 | gnorm 2.65 | 3485s
step 5500 | loss 2.4590 | gnorm 2.47 | 3499s
val 2.7188 | ppl 15.16 | 11M tokens
saved native_transformer.pt