KOLM-Alpha / train_native_kolm.log
Dbrent's picture
add train_native_kolm.log
5967f1e verified
Raw
History Blame Contribute Delete
11.2 kB
kolm | 16,916,224 params | vocab 8192 | mps
step 1 | loss 9.0707 | gnorm 14.92 | 5s
step 25 | loss 8.2714 | gnorm 3.12 | 117s
step 50 | loss 7.9729 | gnorm 2.38 | 215s
step 75 | loss 7.5448 | gnorm 2.27 | 315s
step 100 | loss 7.1723 | gnorm 1.85 | 414s
step 125 | loss 6.5443 | gnorm 1.72 | 515s
step 150 | loss 6.0590 | gnorm 1.51 | 616s
step 175 | loss 5.4919 | gnorm 2.04 | 713s
step 200 | loss 5.2394 | gnorm 1.88 | 806s
step 225 | loss 5.1267 | gnorm 1.57 | 898s
step 250 | loss 4.7845 | gnorm 1.39 | 992s
val 4.8859 | ppl 132.41 | 1M tokens
step 275 | loss 4.8013 | gnorm 1.88 | 1113s
step 300 | loss 4.7008 | gnorm 1.89 | 1205s
step 325 | loss 4.4304 | gnorm 1.63 | 1298s
step 350 | loss 4.5788 | gnorm 1.77 | 1390s
step 375 | loss 4.1259 | gnorm 1.79 | 1483s
step 400 | loss 4.2311 | gnorm 1.64 | 1575s
step 425 | loss 4.2352 | gnorm 1.75 | 1668s
step 450 | loss 4.2986 | gnorm 1.79 | 1763s
step 475 | loss 4.1709 | gnorm 1.89 | 1855s
step 500 | loss 4.2580 | gnorm 1.78 | 1949s
val 4.1100 | ppl 60.94 | 1M tokens
step 525 | loss 3.9729 | gnorm 1.93 | 2070s
step 550 | loss 4.1246 | gnorm 1.88 | 2178s
step 575 | loss 3.9847 | gnorm 2.04 | 2297s
step 600 | loss 3.8311 | gnorm 1.77 | 2401s
step 625 | loss 3.7055 | gnorm 1.96 | 2505s
step 650 | loss 3.8843 | gnorm 2.09 | 2605s
step 675 | loss 3.7934 | gnorm 2.10 | 2709s
step 700 | loss 4.0944 | gnorm 2.24 | 2812s
step 725 | loss 3.9397 | gnorm 2.18 | 2920s
step 750 | loss 3.8257 | gnorm 2.13 | 3016s
val 3.7888 | ppl 44.20 | 2M tokens
step 775 | loss 3.7182 | gnorm 2.08 | 3138s
step 800 | loss 3.7808 | gnorm 2.28 | 3232s
step 825 | loss 3.5318 | gnorm 2.44 | 3326s
step 850 | loss 3.7255 | gnorm 2.02 | 3419s
step 875 | loss 3.7104 | gnorm 2.21 | 3515s
step 900 | loss 3.5115 | gnorm 2.11 | 3608s
step 925 | loss 3.7161 | gnorm 2.22 | 3701s
step 950 | loss 3.5390 | gnorm 2.31 | 3794s
step 975 | loss 3.5380 | gnorm 2.32 | 3887s
step 1000 | loss 3.6917 | gnorm 2.13 | 3980s
val 3.5668 | ppl 35.40 | 2M tokens
step 1025 | loss 3.3968 | gnorm 2.10 | 4104s
step 1050 | loss 3.7517 | gnorm 2.33 | 4198s
step 1075 | loss 3.5378 | gnorm 2.49 | 4291s
step 1100 | loss 3.6530 | gnorm 2.24 | 4384s
step 1125 | loss 3.4710 | gnorm 2.36 | 4477s
step 1150 | loss 3.3378 | gnorm 2.55 | 4571s
step 1175 | loss 3.6750 | gnorm 2.40 | 4664s
step 1200 | loss 3.4526 | gnorm 2.60 | 4759s
step 1225 | loss 3.4160 | gnorm 2.10 | 4854s
step 1250 | loss 3.4874 | gnorm 2.24 | 4947s
val 3.4039 | ppl 30.08 | 3M tokens
step 1275 | loss 3.3618 | gnorm 2.46 | 5069s
step 1300 | loss 3.2796 | gnorm 2.38 | 5166s
step 1325 | loss 3.1655 | gnorm 2.30 | 5259s
step 1350 | loss 3.0861 | gnorm 2.31 | 5354s
step 1375 | loss 3.1603 | gnorm 2.40 | 5449s
step 1400 | loss 3.3567 | gnorm 2.51 | 5542s
step 1425 | loss 3.4282 | gnorm 2.39 | 5635s
step 1450 | loss 3.0577 | gnorm 2.39 | 5728s
step 1475 | loss 3.2619 | gnorm 2.62 | 5821s
step 1500 | loss 3.2249 | gnorm 2.43 | 5915s
val 3.2811 | ppl 26.61 | 3M tokens
step 1525 | loss 3.4037 | gnorm 2.31 | 6037s
step 1550 | loss 3.0126 | gnorm 2.32 | 6131s
step 1575 | loss 3.1633 | gnorm 2.59 | 6225s
step 1600 | loss 3.0170 | gnorm 2.43 | 6319s
step 1625 | loss 3.2520 | gnorm 2.37 | 6412s
step 1650 | loss 3.2816 | gnorm 2.40 | 6506s
step 1675 | loss 3.0603 | gnorm 2.63 | 6599s
step 1700 | loss 3.1222 | gnorm 2.66 | 6692s
step 1725 | loss 3.0850 | gnorm 2.44 | 6786s
step 1750 | loss 3.2087 | gnorm 2.44 | 6879s
val 3.1717 | ppl 23.85 | 4M tokens
step 1775 | loss 3.2430 | gnorm 2.31 | 7002s
step 1800 | loss 3.1905 | gnorm 2.68 | 7096s
step 1825 | loss 3.3060 | gnorm 2.72 | 7191s
step 1850 | loss 3.1687 | gnorm 2.46 | 7284s
step 1875 | loss 3.3862 | gnorm 2.50 | 7377s
step 1900 | loss 3.3346 | gnorm 2.36 | 7470s
step 1925 | loss 2.9384 | gnorm 2.60 | 7563s
step 1950 | loss 3.2170 | gnorm 2.59 | 7657s
step 1975 | loss 2.8825 | gnorm 2.63 | 7751s
step 2000 | loss 3.3292 | gnorm 2.88 | 7844s
val 3.1003 | ppl 22.20 | 4M tokens
step 2025 | loss 2.8098 | gnorm 2.31 | 7966s
step 2050 | loss 2.7519 | gnorm 2.46 | 8060s
step 2075 | loss 3.0944 | gnorm 2.75 | 8155s
step 2100 | loss 3.2591 | gnorm 2.77 | 8248s
step 2125 | loss 3.2165 | gnorm 2.56 | 8345s
step 2150 | loss 2.9171 | gnorm 2.70 | 8439s
step 2175 | loss 3.0677 | gnorm 2.56 | 8536s
step 2200 | loss 3.1405 | gnorm 2.64 | 8632s
step 2225 | loss 3.0420 | gnorm 2.69 | 8727s
step 2250 | loss 3.0561 | gnorm 2.64 | 8820s
val 3.0273 | ppl 20.64 | 5M tokens
step 2275 | loss 3.2381 | gnorm 2.66 | 8944s
step 2300 | loss 2.9071 | gnorm 2.61 | 9039s
step 2325 | loss 3.0277 | gnorm 2.59 | 9134s
step 2350 | loss 2.9954 | gnorm 2.65 | 9229s
step 2375 | loss 2.9762 | gnorm 2.77 | 9325s
step 2400 | loss 3.0373 | gnorm 2.75 | 9418s
step 2425 | loss 3.0645 | gnorm 2.61 | 9512s
step 2450 | loss 2.8629 | gnorm 2.58 | 9608s
step 2475 | loss 2.8987 | gnorm 2.56 | 9723s
step 2500 | loss 3.1771 | gnorm 2.67 | 9823s
val 2.9779 | ppl 19.65 | 5M tokens
step 2525 | loss 2.7999 | gnorm 2.63 | 9947s
step 2550 | loss 2.8363 | gnorm 2.58 | 10041s
step 2575 | loss 3.1341 | gnorm 2.98 | 10136s
step 2600 | loss 2.8335 | gnorm 2.90 | 10230s
step 2625 | loss 2.8166 | gnorm 2.70 | 10325s
step 2650 | loss 3.0086 | gnorm 2.80 | 10420s
step 2675 | loss 2.9273 | gnorm 2.79 | 10515s
step 2700 | loss 3.0408 | gnorm 2.83 | 10608s
step 2725 | loss 2.7906 | gnorm 2.91 | 10703s
step 2750 | loss 3.0174 | gnorm 2.93 | 10797s
val 2.9121 | ppl 18.40 | 6M tokens
step 2775 | loss 2.8560 | gnorm 2.52 | 10921s
step 2800 | loss 3.0436 | gnorm 2.99 | 11015s
step 2825 | loss 3.0029 | gnorm 2.78 | 11108s
step 2850 | loss 2.9033 | gnorm 2.99 | 11203s
step 2875 | loss 3.0680 | gnorm 2.73 | 11300s
step 2900 | loss 2.8864 | gnorm 2.78 | 11393s
step 2925 | loss 2.9173 | gnorm 2.81 | 11487s
step 2950 | loss 2.9114 | gnorm 2.73 | 11582s
step 2975 | loss 2.7408 | gnorm 2.83 | 11675s
step 3000 | loss 2.9040 | gnorm 2.85 | 11770s
val 2.8773 | ppl 17.77 | 6M tokens
step 3025 | loss 3.1609 | gnorm 3.00 | 11895s
step 3050 | loss 2.8313 | gnorm 2.93 | 11991s
step 3075 | loss 2.6176 | gnorm 3.05 | 12085s
step 3100 | loss 2.7470 | gnorm 3.09 | 12178s
step 3125 | loss 2.8237 | gnorm 3.06 | 12273s
step 3150 | loss 2.7972 | gnorm 2.97 | 12366s
step 3175 | loss 3.0566 | gnorm 3.05 | 12460s
step 3200 | loss 3.0810 | gnorm 3.08 | 12555s
step 3225 | loss 3.0267 | gnorm 2.95 | 12649s
step 3250 | loss 2.9154 | gnorm 3.09 | 12745s
val 2.8339 | ppl 17.01 | 7M tokens
step 3275 | loss 2.7173 | gnorm 3.13 | 12867s
step 3300 | loss 2.7352 | gnorm 2.87 | 12962s
step 3325 | loss 2.7853 | gnorm 3.23 | 13056s
step 3350 | loss 3.0113 | gnorm 3.17 | 13152s
step 3375 | loss 3.0887 | gnorm 3.13 | 13249s
step 3400 | loss 2.8065 | gnorm 3.02 | 13344s
step 3425 | loss 2.8143 | gnorm 2.97 | 13439s
step 3450 | loss 2.8024 | gnorm 3.06 | 13534s
step 3475 | loss 2.9455 | gnorm 3.09 | 13628s
step 3500 | loss 2.9508 | gnorm 3.17 | 13724s
val 2.8008 | ppl 16.46 | 7M tokens
step 3525 | loss 2.9241 | gnorm 3.22 | 13849s
step 3550 | loss 2.8766 | gnorm 3.07 | 13943s
step 3575 | loss 2.6811 | gnorm 3.14 | 14037s
step 3600 | loss 2.5598 | gnorm 2.88 | 14132s
step 3625 | loss 2.6413 | gnorm 3.08 | 14226s
step 3650 | loss 2.6940 | gnorm 3.05 | 14321s
step 3675 | loss 2.5285 | gnorm 3.18 | 14415s
step 3700 | loss 2.9373 | gnorm 3.06 | 14509s
step 3725 | loss 2.7980 | gnorm 3.18 | 14606s
step 3750 | loss 2.8113 | gnorm 3.06 | 14702s
val 2.7743 | ppl 16.03 | 8M tokens
step 3775 | loss 2.8005 | gnorm 3.21 | 14825s
step 3800 | loss 2.8123 | gnorm 3.23 | 14919s
step 3825 | loss 2.7244 | gnorm 2.99 | 15012s
step 3850 | loss 2.6036 | gnorm 3.19 | 15107s
step 3875 | loss 2.6206 | gnorm 3.01 | 15201s
step 3900 | loss 2.5713 | gnorm 3.03 | 15315s
step 3925 | loss 2.4534 | gnorm 3.24 | 15412s
step 3950 | loss 2.8260 | gnorm 3.17 | 15507s
step 3975 | loss 2.8093 | gnorm 3.11 | 15602s
step 4000 | loss 2.7496 | gnorm 3.18 | 15697s
val 2.7499 | ppl 15.64 | 8M tokens
step 4025 | loss 2.9053 | gnorm 3.24 | 15819s
step 4050 | loss 2.6918 | gnorm 3.02 | 15912s
step 4075 | loss 2.8410 | gnorm 3.26 | 16012s
step 4100 | loss 2.7221 | gnorm 2.95 | 16120s
step 4125 | loss 2.6604 | gnorm 3.27 | 16219s
step 4150 | loss 2.6568 | gnorm 3.22 | 16325s
step 4175 | loss 2.7244 | gnorm 3.41 | 16440s
step 4200 | loss 2.7475 | gnorm 3.21 | 16546s
step 4225 | loss 2.8774 | gnorm 3.32 | 16642s
step 4250 | loss 2.8362 | gnorm 3.35 | 16739s
val 2.7310 | ppl 15.35 | 9M tokens
step 4275 | loss 2.7458 | gnorm 3.41 | 16865s
step 4300 | loss 2.6054 | gnorm 3.30 | 16961s
step 4325 | loss 2.6604 | gnorm 3.29 | 17057s
step 4350 | loss 2.5408 | gnorm 3.12 | 17152s
step 4375 | loss 3.0791 | gnorm 3.80 | 17246s
step 4400 | loss 2.8963 | gnorm 3.23 | 17342s
step 4425 | loss 2.7916 | gnorm 3.23 | 17448s
step 4450 | loss 2.8221 | gnorm 3.23 | 17547s
step 4475 | loss 2.4218 | gnorm 3.16 | 17641s
step 4500 | loss 2.5457 | gnorm 3.34 | 17736s
val 2.7140 | ppl 15.09 | 9M tokens
step 4525 | loss 2.5867 | gnorm 3.32 | 17858s
step 4550 | loss 2.6694 | gnorm 3.33 | 17954s
step 4575 | loss 2.6873 | gnorm 3.40 | 18049s
step 4600 | loss 2.6940 | gnorm 3.30 | 18143s
step 4625 | loss 2.8509 | gnorm 3.33 | 18237s
step 4650 | loss 2.6937 | gnorm 3.37 | 18331s
step 4675 | loss 3.1286 | gnorm 3.44 | 18425s
step 4700 | loss 2.6899 | gnorm 3.41 | 18521s
step 4725 | loss 2.7882 | gnorm 3.41 | 18615s
step 4750 | loss 2.5785 | gnorm 3.22 | 18709s
val 2.7055 | ppl 14.96 | 10M tokens
step 4775 | loss 2.5895 | gnorm 3.22 | 18830s
step 4800 | loss 2.8947 | gnorm 3.45 | 18922s
step 4825 | loss 2.5815 | gnorm 3.48 | 19017s
step 4850 | loss 2.7833 | gnorm 3.45 | 19115s
step 4875 | loss 2.9061 | gnorm 3.32 | 19219s
step 4900 | loss 2.4818 | gnorm 3.57 | 19314s
step 4925 | loss 3.0002 | gnorm 3.48 | 19408s
step 4950 | loss 2.6376 | gnorm 3.23 | 19520s
step 4975 | loss 2.6142 | gnorm 3.37 | 19618s
step 5000 | loss 2.8309 | gnorm 3.54 | 19717s
val 2.6979 | ppl 14.85 | 10M tokens
step 5025 | loss 2.7186 | gnorm 3.29 | 19846s
step 5050 | loss 2.6430 | gnorm 3.65 | 24809s
step 5075 | loss 2.6572 | gnorm 3.23 | 25039s
step 5100 | loss 2.7021 | gnorm 3.35 | 25219s
step 5125 | loss 2.9417 | gnorm 3.29 | 25303s
step 5150 | loss 2.8187 | gnorm 3.22 | 25386s
step 5175 | loss 2.7659 | gnorm 3.67 | 25472s
step 5200 | loss 2.6885 | gnorm 3.25 | 25571s
step 5225 | loss 2.6109 | gnorm 3.32 | 25687s
step 5250 | loss 3.0765 | gnorm 3.43 | 25789s
val 2.6950 | ppl 14.81 | 11M tokens
step 5275 | loss 2.7892 | gnorm 3.44 | 25918s
step 5300 | loss 2.8057 | gnorm 3.15 | 26007s
step 5325 | loss 2.5539 | gnorm 3.16 | 26097s
step 5350 | loss 2.7243 | gnorm 3.31 | 26180s
step 5375 | loss 2.7055 | gnorm 3.33 | 26295s
step 5400 | loss 2.8173 | gnorm 3.31 | 26385s
step 5425 | loss 2.8325 | gnorm 3.47 | 26482s
step 5450 | loss 2.5946 | gnorm 3.36 | 26576s
step 5475 | loss 2.6821 | gnorm 3.44 | 26670s
step 5500 | loss 2.4097 | gnorm 3.38 | 26770s
val 2.6946 | ppl 14.80 | 11M tokens
saved native_kolm.pt