OutrageouslyBad200/SMAT_ablations
0
1[init] variant=baseline seed=0 params=63.38M
2[init] use_S=False use_G=False
3[baseline s0] step 0/4000 loss=10.8790 gn=15.69 lr=3.00e-06 tps=8974 mem=5.5G lam=None Gm=None
4[baseline s0] step 20/4000 loss=9.8580 gn=1.93 lr=6.30e-05 tps=18098 mem=6.0G lam=None Gm=None
5[baseline s0] step 40/4000 loss=9.1089 gn=1.75 lr=1.23e-04 tps=18558 mem=6.0G lam=None Gm=None
6[baseline s0] step 60/4000 loss=8.1093 gn=1.57 lr=1.83e-04 tps=18726 mem=6.0G lam=None Gm=None
7[baseline s0] step 80/4000 loss=7.4265 gn=0.73 lr=2.43e-04 tps=18779 mem=6.0G lam=None Gm=None
8[baseline s0] step 100/4000 loss=7.2358 gn=0.90 lr=3.03e-04 tps=18805 mem=6.0G lam=None Gm=None
9[baseline s0] step 120/4000 loss=7.0669 gn=0.67 lr=3.63e-04 tps=18838 mem=6.0G lam=None Gm=None
10[baseline s0] step 140/4000 loss=6.8653 gn=0.92 lr=4.23e-04 tps=18849 mem=6.0G lam=None Gm=None
11[baseline s0] step 160/4000 loss=6.9054 gn=0.56 lr=4.83e-04 tps=18856 mem=6.0G lam=None Gm=None
12[baseline s0] step 180/4000 loss=6.8181 gn=0.80 lr=5.43e-04 tps=18835 mem=6.0G lam=None Gm=None
13[baseline s0] step 200/4000 loss=6.6996 gn=0.67 lr=6.00e-04 tps=18841 mem=6.0G lam=None Gm=None
14[baseline s0] EVAL step 200: val_loss=6.7583 val_ppl=861.17
15[baseline s0] step 220/4000 loss=6.5854 gn=0.53 lr=6.00e-04 tps=18244 mem=6.0G lam=None Gm=None
16[baseline s0] step 240/4000 loss=6.6331 gn=0.57 lr=6.00e-04 tps=18291 mem=6.0G lam=None Gm=None
17[baseline s0] step 260/4000 loss=6.6194 gn=0.57 lr=6.00e-04 tps=18333 mem=6.0G lam=None Gm=None
18[baseline s0] step 280/4000 loss=6.5077 gn=0.54 lr=5.99e-04 tps=18370 mem=6.0G lam=None Gm=None
19[baseline s0] step 300/4000 loss=6.3495 gn=0.50 lr=5.99e-04 tps=18407 mem=6.0G lam=None Gm=None
20[baseline s0] step 320/4000 loss=6.4110 gn=0.53 lr=5.99e-04 tps=18431 mem=6.0G lam=None Gm=None
21[baseline s0] step 340/4000 loss=6.3930 gn=0.50 lr=5.98e-04 tps=18462 mem=6.0G lam=None Gm=None
22[baseline s0] step 360/4000 loss=6.2602 gn=0.54 lr=5.98e-04 tps=18487 mem=6.0G lam=None Gm=None
23[baseline s0] step 380/4000 loss=6.2289 gn=0.55 lr=5.97e-04 tps=18512 mem=6.0G lam=None Gm=None
24[baseline s0] step 400/4000 loss=6.2561 gn=0.53 lr=5.96e-04 tps=18533 mem=6.0G lam=None Gm=None
25[baseline s0] EVAL step 400: val_loss=6.4037 val_ppl=604.09
26[baseline s0] step 420/4000 loss=6.4232 gn=0.80 lr=5.96e-04 tps=18241 mem=6.0G lam=None Gm=None
27[baseline s0] step 440/4000 loss=6.3506 gn=0.59 lr=5.95e-04 tps=18273 mem=6.0G lam=None Gm=None
28[baseline s0] step 460/4000 loss=6.3034 gn=0.55 lr=5.94e-04 tps=18295 mem=6.0G lam=None Gm=None
29[baseline s0] step 480/4000 loss=6.3906 gn=0.58 lr=5.93e-04 tps=18308 mem=6.0G lam=None Gm=None
30[baseline s0] step 500/4000 loss=6.3182 gn=0.56 lr=5.92e-04 tps=18328 mem=6.0G lam=None Gm=None
31[baseline s0] step 520/4000 loss=6.1685 gn=0.55 lr=5.91e-04 tps=18349 mem=6.0G lam=None Gm=None
32[baseline s0] step 540/4000 loss=6.1850 gn=0.49 lr=5.89e-04 tps=18368 mem=6.0G lam=None Gm=None
33[baseline s0] step 560/4000 loss=6.0446 gn=0.56 lr=5.88e-04 tps=18388 mem=6.0G lam=None Gm=None
34[baseline s0] step 580/4000 loss=6.2096 gn=0.51 lr=5.87e-04 tps=18407 mem=6.0G lam=None Gm=None
35[baseline s0] step 600/4000 loss=5.9779 gn=0.55 lr=5.85e-04 tps=18425 mem=6.0G lam=None Gm=None
36[baseline s0] EVAL step 600: val_loss=6.1749 val_ppl=480.52
37[baseline s0] step 620/4000 loss=6.0322 gn=0.50 lr=5.84e-04 tps=18235 mem=6.0G lam=None Gm=None
38[baseline s0] step 640/4000 loss=6.1268 gn=0.65 lr=5.82e-04 tps=18254 mem=6.0G lam=None Gm=None
39[baseline s0] step 660/4000 loss=5.9413 gn=0.54 lr=5.81e-04 tps=18275 mem=6.0G lam=None Gm=None
40[baseline s0] step 680/4000 loss=5.9597 gn=0.56 lr=5.79e-04 tps=18276 mem=6.0G lam=None Gm=None
41[baseline s0] step 700/4000 loss=6.0550 gn=0.57 lr=5.77e-04 tps=18288 mem=6.0G lam=None Gm=None
42[baseline s0] step 720/4000 loss=6.0272 gn=0.54 lr=5.75e-04 tps=18306 mem=6.0G lam=None Gm=None
43[baseline s0] step 740/4000 loss=6.0446 gn=0.52 lr=5.74e-04 tps=18320 mem=6.0G lam=None Gm=None
44[baseline s0] step 760/4000 loss=5.9250 gn=0.57 lr=5.72e-04 tps=18335 mem=6.0G lam=None Gm=None
45[baseline s0] step 780/4000 loss=5.9373 gn=0.54 lr=5.70e-04 tps=18348 mem=6.0G lam=None Gm=None
46[baseline s0] step 800/4000 loss=5.9007 gn=0.55 lr=5.67e-04 tps=18362 mem=6.0G lam=None Gm=None
47[baseline s0] EVAL step 800: val_loss=5.9891 val_ppl=399.06
48[baseline s0] step 820/4000 loss=5.9677 gn=0.51 lr=5.65e-04 tps=18212 mem=6.0G lam=None Gm=None
49[baseline s0] step 840/4000 loss=6.1449 gn=0.53 lr=5.63e-04 tps=18229 mem=6.0G lam=None Gm=None
50[baseline s0] step 860/4000 loss=5.9614 gn=0.54 lr=5.61e-04 tps=18243 mem=6.0G lam=None Gm=None
51[baseline s0] step 880/4000 loss=5.9419 gn=0.64 lr=5.58e-04 tps=18255 mem=6.0G lam=None Gm=None
52[baseline s0] step 900/4000 loss=5.8523 gn=0.52 lr=5.56e-04 tps=18269 mem=6.0G lam=None Gm=None
53[baseline s0] step 920/4000 loss=5.7571 gn=0.52 lr=5.54e-04 tps=18283 mem=6.0G lam=None Gm=None
54[baseline s0] step 940/4000 loss=5.8122 gn=0.55 lr=5.51e-04 tps=18294 mem=6.0G lam=None Gm=None
55[baseline s0] step 960/4000 loss=5.7968 gn=0.58 lr=5.48e-04 tps=18305 mem=6.0G lam=None Gm=None
56[baseline s0] step 980/4000 loss=5.7111 gn=0.51 lr=5.46e-04 tps=18318 mem=6.0G lam=None Gm=None
57[baseline s0] step 1000/4000 loss=5.8198 gn=0.57 lr=5.43e-04 tps=18330 mem=6.0G lam=None Gm=None
58[baseline s0] EVAL step 1000: val_loss=5.8357 val_ppl=342.32
59[baseline s0] step 1020/4000 loss=5.5483 gn=0.50 lr=5.40e-04 tps=18217 mem=6.0G lam=None Gm=None
60[baseline s0] step 1040/4000 loss=5.8019 gn=0.51 lr=5.37e-04 tps=18230 mem=6.0G lam=None Gm=None
61[baseline s0] step 1060/4000 loss=5.5165 gn=0.56 lr=5.35e-04 tps=18240 mem=6.0G lam=None Gm=None
62[baseline s0] step 1080/4000 loss=5.7172 gn=0.57 lr=5.32e-04 tps=18253 mem=6.0G lam=None Gm=None
63[baseline s0] step 1100/4000 loss=5.6952 gn=0.56 lr=5.29e-04 tps=18265 mem=6.0G lam=None Gm=None
64[baseline s0] step 1120/4000 loss=5.6321 gn=0.59 lr=5.26e-04 tps=18275 mem=6.0G lam=None Gm=None
65[baseline s0] step 1140/4000 loss=5.8674 gn=0.51 lr=5.22e-04 tps=18282 mem=6.0G lam=None Gm=None
66[baseline s0] step 1160/4000 loss=5.6313 gn=0.54 lr=5.19e-04 tps=18287 mem=6.0G lam=None Gm=None
67[baseline s0] step 1180/4000 loss=5.6920 gn=0.53 lr=5.16e-04 tps=18296 mem=6.0G lam=None Gm=None
68[baseline s0] step 1200/4000 loss=5.5678 gn=0.53 lr=5.13e-04 tps=18304 mem=6.0G lam=None Gm=None
69[baseline s0] EVAL step 1200: val_loss=5.6887 val_ppl=295.50
70[baseline s0] step 1220/4000 loss=5.5960 gn=0.61 lr=5.10e-04 tps=18204 mem=6.0G lam=None Gm=None
71[baseline s0] step 1240/4000 loss=5.4634 gn=0.54 lr=5.06e-04 tps=18208 mem=6.0G lam=None Gm=None
72[baseline s0] step 1260/4000 loss=5.4685 gn=0.56 lr=5.03e-04 tps=18216 mem=6.0G lam=None Gm=None
73[baseline s0] step 1280/4000 loss=5.5946 gn=0.58 lr=4.99e-04 tps=18223 mem=6.0G lam=None Gm=None
74[baseline s0] step 1300/4000 loss=5.6908 gn=0.59 lr=4.96e-04 tps=18232 mem=6.0G lam=None Gm=None
75[baseline s0] step 1320/4000 loss=5.6382 gn=0.59 lr=4.92e-04 tps=18238 mem=6.0G lam=None Gm=None
76[baseline s0] step 1340/4000 loss=5.7493 gn=0.54 lr=4.89e-04 tps=18242 mem=6.0G lam=None Gm=None
77[baseline s0] step 1360/4000 loss=5.5626 gn=0.56 lr=4.85e-04 tps=18249 mem=6.0G lam=None Gm=None
78[baseline s0] step 1380/4000 loss=5.4316 gn=0.56 lr=4.81e-04 tps=18252 mem=6.0G lam=None Gm=None
79[baseline s0] step 1400/4000 loss=5.4411 gn=0.59 lr=4.78e-04 tps=18257 mem=6.0G lam=None Gm=None
80[baseline s0] EVAL step 1400: val_loss=5.5751 val_ppl=263.76
81[baseline s0] step 1420/4000 loss=5.4237 gn=0.60 lr=4.74e-04 tps=18171 mem=6.0G lam=None Gm=None
82[baseline s0] step 1440/4000 loss=5.4160 gn=0.62 lr=4.70e-04 tps=18177 mem=6.0G lam=None Gm=None
83[baseline s0] step 1460/4000 loss=5.5336 gn=0.60 lr=4.66e-04 tps=18184 mem=6.0G lam=None Gm=None
84[baseline s0] step 1480/4000 loss=5.5815 gn=0.57 lr=4.62e-04 tps=18191 mem=6.0G lam=None Gm=None
85[baseline s0] step 1500/4000 loss=5.4963 gn=0.61 lr=4.59e-04 tps=18197 mem=6.0G lam=None Gm=None
86[baseline s0] step 1520/4000 loss=5.4520 gn=0.57 lr=4.55e-04 tps=18206 mem=6.0G lam=None Gm=None
87[baseline s0] step 1540/4000 loss=5.3651 gn=0.55 lr=4.51e-04 tps=18212 mem=6.0G lam=None Gm=None
88[baseline s0] step 1560/4000 loss=5.3625 gn=0.58 lr=4.47e-04 tps=18218 mem=6.0G lam=None Gm=None
89[baseline s0] step 1580/4000 loss=5.4415 gn=0.61 lr=4.43e-04 tps=18225 mem=6.0G lam=None Gm=None
90[baseline s0] step 1600/4000 loss=5.4264 gn=0.57 lr=4.38e-04 tps=18233 mem=6.0G lam=None Gm=None
91[baseline s0] EVAL step 1600: val_loss=5.4921 val_ppl=242.78
92[baseline s0] step 1620/4000 loss=5.4024 gn=0.56 lr=4.34e-04 tps=18161 mem=6.0G lam=None Gm=None
93[baseline s0] step 1640/4000 loss=5.2968 gn=0.58 lr=4.30e-04 tps=18168 mem=6.0G lam=None Gm=None
94[baseline s0] step 1660/4000 loss=5.4722 gn=0.60 lr=4.26e-04 tps=18176 mem=6.0G lam=None Gm=None
95[baseline s0] step 1680/4000 loss=5.6134 gn=0.56 lr=4.22e-04 tps=18183 mem=6.0G lam=None Gm=None
96[baseline s0] step 1700/4000 loss=5.4681 gn=0.59 lr=4.18e-04 tps=18190 mem=6.0G lam=None Gm=None
97[baseline s0] step 1720/4000 loss=5.2368 gn=0.60 lr=4.13e-04 tps=18195 mem=6.0G lam=None Gm=None
98[baseline s0] step 1740/4000 loss=5.2863 gn=0.59 lr=4.09e-04 tps=18195 mem=6.0G lam=None Gm=None
99[baseline s0] step 1760/4000 loss=5.3700 gn=0.57 lr=4.05e-04 tps=18198 mem=6.0G lam=None Gm=None
100[baseline s0] step 1780/4000 loss=5.2317 gn=0.60 lr=4.01e-04 tps=18204 mem=6.0G lam=None Gm=None
101[baseline s0] step 1800/4000 loss=5.3057 gn=0.59 lr=3.96e-04 tps=18209 mem=6.0G lam=None Gm=None
102[baseline s0] EVAL step 1800: val_loss=5.4254 val_ppl=227.11
103[baseline s0] step 1820/4000 loss=5.3907 gn=0.58 lr=3.92e-04 tps=18146 mem=6.0G lam=None Gm=None
104[baseline s0] step 1840/4000 loss=5.2270 gn=0.61 lr=3.88e-04 tps=18154 mem=6.0G lam=None Gm=None
105[baseline s0] step 1860/4000 loss=5.2737 gn=0.59 lr=3.83e-04 tps=18159 mem=6.0G lam=None Gm=None
106[baseline s0] step 1880/4000 loss=5.3796 gn=0.70 lr=3.79e-04 tps=18166 mem=6.0G lam=None Gm=None
107[baseline s0] step 1900/4000 loss=5.3310 gn=0.66 lr=3.74e-04 tps=18173 mem=6.0G lam=None Gm=None
108[baseline s0] step 1920/4000 loss=5.2796 gn=0.59 lr=3.70e-04 tps=18179 mem=6.0G lam=None Gm=None
109[baseline s0] step 1940/4000 loss=5.3340 gn=0.61 lr=3.66e-04 tps=18186 mem=6.0G lam=None Gm=None
110[baseline s0] step 1960/4000 loss=5.5327 gn=0.63 lr=3.61e-04 tps=18193 mem=6.0G lam=None Gm=None
111[baseline s0] step 1980/4000 loss=5.4405 gn=0.61 lr=3.57e-04 tps=18199 mem=6.0G lam=None Gm=None
112[baseline s0] step 2000/4000 loss=5.2762 gn=0.61 lr=3.52e-04 tps=18205 mem=6.0G lam=None Gm=None
113[baseline s0] EVAL step 2000: val_loss=5.3791 val_ppl=216.82
114[baseline s0] step 2020/4000 loss=5.2284 gn=0.58 lr=3.48e-04 tps=18147 mem=6.0G lam=None Gm=None
115[baseline s0] step 2040/4000 loss=5.3605 gn=0.67 lr=3.43e-04 tps=18154 mem=6.0G lam=None Gm=None
116[baseline s0] step 2060/4000 loss=5.1268 gn=0.72 lr=3.39e-04 tps=18159 mem=6.0G lam=None Gm=None
117[baseline s0] step 2080/4000 loss=5.2638 gn=0.63 lr=3.34e-04 tps=18165 mem=6.0G lam=None Gm=None
118[baseline s0] step 2100/4000 loss=5.2701 gn=0.58 lr=3.30e-04 tps=18171 mem=6.0G lam=None Gm=None
119[baseline s0] step 2120/4000 loss=5.1915 gn=0.64 lr=3.26e-04 tps=18176 mem=6.0G lam=None Gm=None
120[baseline s0] step 2140/4000 loss=5.0964 gn=0.62 lr=3.21e-04 tps=18183 mem=6.0G lam=None Gm=None
121[baseline s0] step 2160/4000 loss=5.1444 gn=0.60 lr=3.17e-04 tps=18187 mem=6.0G lam=None Gm=None
122[baseline s0] step 2180/4000 loss=5.1955 gn=0.73 lr=3.12e-04 tps=18192 mem=6.0G lam=None Gm=None
123[baseline s0] step 2200/4000 loss=5.3442 gn=0.62 lr=3.08e-04 tps=18198 mem=6.0G lam=None Gm=None
124[baseline s0] EVAL step 2200: val_loss=5.2990 val_ppl=200.13
125[baseline s0] step 2220/4000 loss=5.2295 gn=0.63 lr=3.03e-04 tps=18147 mem=6.0G lam=None Gm=None
126[baseline s0] step 2240/4000 loss=5.1707 gn=0.62 lr=2.99e-04 tps=18152 mem=6.0G lam=None Gm=None
127[baseline s0] step 2260/4000 loss=5.1685 gn=0.60 lr=2.94e-04 tps=18158 mem=6.0G lam=None Gm=None
128[baseline s0] step 2280/4000 loss=5.3584 gn=0.66 lr=2.90e-04 tps=18163 mem=6.0G lam=None Gm=None
129[baseline s0] step 2300/4000 loss=5.1364 gn=0.66 lr=2.86e-04 tps=18169 mem=6.0G lam=None Gm=None
130[baseline s0] step 2320/4000 loss=5.2594 gn=0.67 lr=2.81e-04 tps=18174 mem=6.0G lam=None Gm=None
131[baseline s0] step 2340/4000 loss=5.1838 gn=0.63 lr=2.77e-04 tps=18179 mem=6.0G lam=None Gm=None
132[baseline s0] step 2360/4000 loss=5.2727 gn=0.66 lr=2.72e-04 tps=18185 mem=6.0G lam=None Gm=None
133[baseline s0] step 2380/4000 loss=5.1474 gn=0.68 lr=2.68e-04 tps=18190 mem=6.0G lam=None Gm=None
134[baseline s0] step 2400/4000 loss=5.0503 gn=0.68 lr=2.64e-04 tps=18195 mem=6.0G lam=None Gm=None
135[baseline s0] EVAL step 2400: val_loss=5.2233 val_ppl=185.54
136[baseline s0] step 2420/4000 loss=5.0304 gn=0.62 lr=2.59e-04 tps=18148 mem=6.0G lam=None Gm=None
137[baseline s0] step 2440/4000 loss=5.1615 gn=0.66 lr=2.55e-04 tps=18154 mem=6.0G lam=None Gm=None
138[baseline s0] step 2460/4000 loss=5.2470 gn=0.63 lr=2.51e-04 tps=18159 mem=6.0G lam=None Gm=None
139[baseline s0] step 2480/4000 loss=5.1272 gn=0.61 lr=2.47e-04 tps=18164 mem=6.0G lam=None Gm=None
140[baseline s0] step 2500/4000 loss=4.9969 gn=0.66 lr=2.42e-04 tps=18169 mem=6.0G lam=None Gm=None
141[baseline s0] step 2520/4000 loss=5.0420 gn=0.62 lr=2.38e-04 tps=18174 mem=6.0G lam=None Gm=None
142[baseline s0] step 2540/4000 loss=5.1197 gn=0.65 lr=2.34e-04 tps=18179 mem=6.0G lam=None Gm=None
143[baseline s0] step 2560/4000 loss=4.9050 gn=0.67 lr=2.30e-04 tps=18185 mem=6.0G lam=None Gm=None
144[baseline s0] step 2580/4000 loss=5.2086 gn=0.63 lr=2.26e-04 tps=18190 mem=6.0G lam=None Gm=None
145[baseline s0] step 2600/4000 loss=4.9868 gn=0.63 lr=2.22e-04 tps=18194 mem=6.0G lam=None Gm=None
146[baseline s0] EVAL step 2600: val_loss=5.1919 val_ppl=179.82
147[baseline s0] step 2620/4000 loss=5.1561 gn=0.69 lr=2.17e-04 tps=18151 mem=6.0G lam=None Gm=None
148[baseline s0] step 2640/4000 loss=5.1533 gn=0.64 lr=2.13e-04 tps=18156 mem=6.0G lam=None Gm=None
149[baseline s0] step 2660/4000 loss=5.0194 gn=0.64 lr=2.09e-04 tps=18160 mem=6.0G lam=None Gm=None
150[baseline s0] step 2680/4000 loss=4.8527 gn=0.79 lr=2.05e-04 tps=18165 mem=6.0G lam=None Gm=None
151[baseline s0] step 2700/4000 loss=5.1246 gn=0.73 lr=2.01e-04 tps=18170 mem=6.0G lam=None Gm=None
152[baseline s0] step 2720/4000 loss=5.1081 gn=0.66 lr=1.98e-04 tps=18175 mem=6.0G lam=None Gm=None
153[baseline s0] step 2740/4000 loss=5.0874 gn=0.63 lr=1.94e-04 tps=18179 mem=6.0G lam=None Gm=None
154[baseline s0] step 2760/4000 loss=4.9413 gn=0.66 lr=1.90e-04 tps=18184 mem=6.0G lam=None Gm=None
155[baseline s0] step 2780/4000 loss=5.0563 gn=0.64 lr=1.86e-04 tps=18188 mem=6.0G lam=None Gm=None
156[baseline s0] step 2800/4000 loss=5.0011 gn=0.69 lr=1.82e-04 tps=18193 mem=6.0G lam=None Gm=None
157[baseline s0] EVAL step 2800: val_loss=5.1333 val_ppl=169.58
158[baseline s0] step 2820/4000 loss=5.0585 gn=0.67 lr=1.79e-04 tps=18152 mem=6.0G lam=None Gm=None
159[baseline s0] step 2840/4000 loss=4.8624 gn=0.73 lr=1.75e-04 tps=18157 mem=6.0G lam=None Gm=None
160[baseline s0] step 2860/4000 loss=5.0362 gn=0.65 lr=1.71e-04 tps=18162 mem=6.0G lam=None Gm=None
161[baseline s0] step 2880/4000 loss=5.0276 gn=0.68 lr=1.68e-04 tps=18166 mem=6.0G lam=None Gm=None
162[baseline s0] step 2900/4000 loss=4.8816 gn=0.68 lr=1.64e-04 tps=18171 mem=6.0G lam=None Gm=None
163[baseline s0] step 2920/4000 loss=5.0774 gn=0.70 lr=1.61e-04 tps=18175 mem=6.0G lam=None Gm=None
164[baseline s0] step 2940/4000 loss=4.9940 gn=0.64 lr=1.57e-04 tps=18179 mem=6.0G lam=None Gm=None
165[baseline s0] step 2960/4000 loss=4.9860 gn=0.67 lr=1.54e-04 tps=18184 mem=6.0G lam=None Gm=None
166[baseline s0] step 2980/4000 loss=4.9343 gn=0.64 lr=1.50e-04 tps=18188 mem=6.0G lam=None Gm=None
167[baseline s0] step 3000/4000 loss=4.8842 gn=0.68 lr=1.47e-04 tps=18192 mem=6.0G lam=None Gm=None
168[baseline s0] EVAL step 3000: val_loss=5.1456 val_ppl=171.68
169[baseline s0] step 3020/4000 loss=4.9521 gn=0.66 lr=1.44e-04 tps=18155 mem=6.0G lam=None Gm=None
170[baseline s0] step 3040/4000 loss=5.0338 gn=0.72 lr=1.41e-04 tps=18159 mem=6.0G lam=None Gm=None
171[baseline s0] step 3060/4000 loss=4.9991 gn=0.70 lr=1.38e-04 tps=18163 mem=6.0G lam=None Gm=None
172[baseline s0] step 3080/4000 loss=4.8222 gn=0.65 lr=1.34e-04 tps=18167 mem=6.0G lam=None Gm=None
173[baseline s0] step 3100/4000 loss=5.0607 gn=0.70 lr=1.31e-04 tps=18171 mem=6.0G lam=None Gm=None
174[baseline s0] step 3120/4000 loss=5.0802 gn=0.81 lr=1.28e-04 tps=18173 mem=6.0G lam=None Gm=None
175[baseline s0] step 3140/4000 loss=4.9207 gn=0.68 lr=1.25e-04 tps=18173 mem=6.0G lam=None Gm=None
176[baseline s0] step 3160/4000 loss=5.0691 gn=0.69 lr=1.23e-04 tps=18177 mem=6.0G lam=None Gm=None
177[baseline s0] step 3180/4000 loss=4.9020 gn=0.67 lr=1.20e-04 tps=18181 mem=6.0G lam=None Gm=None
178[baseline s0] step 3200/4000 loss=4.9458 gn=0.73 lr=1.17e-04 tps=18185 mem=6.0G lam=None Gm=None
179[baseline s0] EVAL step 3200: val_loss=5.0536 val_ppl=156.58
180[baseline s0] step 3220/4000 loss=5.0413 gn=0.75 lr=1.14e-04 tps=18150 mem=6.0G lam=None Gm=None
181[baseline s0] step 3240/4000 loss=5.0531 gn=0.70 lr=1.12e-04 tps=18154 mem=6.0G lam=None Gm=None
182[baseline s0] step 3260/4000 loss=4.8129 gn=0.67 lr=1.09e-04 tps=18158 mem=6.0G lam=None Gm=None
183[baseline s0] step 3280/4000 loss=5.0799 gn=0.77 lr=1.06e-04 tps=18162 mem=6.0G lam=None Gm=None
184[baseline s0] step 3300/4000 loss=4.8907 gn=0.70 lr=1.04e-04 tps=18166 mem=6.0G lam=None Gm=None
185[baseline s0] step 3320/4000 loss=5.0456 gn=0.69 lr=1.02e-04 tps=18170 mem=6.0G lam=None Gm=None
186[baseline s0] step 3340/4000 loss=4.8715 gn=0.67 lr=9.92e-05 tps=18173 mem=6.0G lam=None Gm=None
187[baseline s0] step 3360/4000 loss=5.0376 gn=0.74 lr=9.69e-05 tps=18177 mem=6.0G lam=None Gm=None
188[baseline s0] step 3380/4000 loss=5.0292 gn=0.73 lr=9.47e-05 tps=18181 mem=6.0G lam=None Gm=None
189[baseline s0] step 3400/4000 loss=4.8974 gn=0.68 lr=9.25e-05 tps=18185 mem=6.0G lam=None Gm=None
190[baseline s0] EVAL step 3400: val_loss=5.0562 val_ppl=156.99
191[baseline s0] step 3420/4000 loss=4.9393 gn=0.71 lr=9.04e-05 tps=18152 mem=6.0G lam=None Gm=None
192[baseline s0] step 3440/4000 loss=4.9874 gn=0.75 lr=8.84e-05 tps=18156 mem=6.0G lam=None Gm=None
193[baseline s0] step 3460/4000 loss=4.8650 gn=0.76 lr=8.65e-05 tps=18160 mem=6.0G lam=None Gm=None
194[baseline s0] step 3480/4000 loss=4.9717 gn=0.70 lr=8.46e-05 tps=18164 mem=6.0G lam=None Gm=None
195[baseline s0] step 3500/4000 loss=4.8648 gn=0.70 lr=8.27e-05 tps=18168 mem=6.0G lam=None Gm=None
196[baseline s0] step 3520/4000 loss=4.7724 gn=0.78 lr=8.10e-05 tps=18171 mem=6.0G lam=None Gm=None
197[baseline s0] step 3540/4000 loss=4.7628 gn=0.68 lr=7.93e-05 tps=18176 mem=6.0G lam=None Gm=None
198[baseline s0] step 3560/4000 loss=5.0116 gn=0.68 lr=7.77e-05 tps=18179 mem=6.0G lam=None Gm=None
199[baseline s0] step 3580/4000 loss=4.8404 gn=0.66 lr=7.61e-05 tps=18183 mem=6.0G lam=None Gm=None
200[baseline s0] step 3600/4000 loss=4.8622 gn=0.77 lr=7.46e-05 tps=18187 mem=6.0G lam=None Gm=None
201[baseline s0] EVAL step 3600: val_loss=5.0255 val_ppl=152.25
202[baseline s0] step 3620/4000 loss=4.9256 gn=0.73 lr=7.32e-05 tps=18156 mem=6.0G lam=None Gm=None
203[baseline s0] step 3640/4000 loss=4.9979 gn=0.69 lr=7.19e-05 tps=18160 mem=6.0G lam=None Gm=None
204[baseline s0] step 3660/4000 loss=5.0407 gn=0.78 lr=7.06e-05 tps=18164 mem=6.0G lam=None Gm=None
205[baseline s0] step 3680/4000 loss=4.9707 gn=0.70 lr=6.94e-05 tps=18168 mem=6.0G lam=None Gm=None
206[baseline s0] step 3700/4000 loss=5.0190 gn=0.73 lr=6.83e-05 tps=18171 mem=6.0G lam=None Gm=None
207[baseline s0] step 3720/4000 loss=4.7931 gn=0.69 lr=6.72e-05 tps=18175 mem=6.0G lam=None Gm=None
208[baseline s0] step 3740/4000 loss=4.9578 gn=0.75 lr=6.62e-05 tps=18179 mem=6.0G lam=None Gm=None
209[baseline s0] step 3760/4000 loss=4.8983 gn=0.72 lr=6.53e-05 tps=18182 mem=6.0G lam=None Gm=None
210[baseline s0] step 3780/4000 loss=4.8004 gn=0.76 lr=6.45e-05 tps=18186 mem=6.0G lam=None Gm=None
211[baseline s0] step 3800/4000 loss=4.9087 gn=0.75 lr=6.37e-05 tps=18190 mem=6.0G lam=None Gm=None
212[baseline s0] EVAL step 3800: val_loss=5.0057 val_ppl=149.26
213[baseline s0] step 3820/4000 loss=4.9865 gn=0.74 lr=6.30e-05 tps=18160 mem=6.0G lam=None Gm=None
214[baseline s0] step 3840/4000 loss=4.9053 gn=0.76 lr=6.24e-05 tps=18164 mem=6.0G lam=None Gm=None
215[baseline s0] step 3860/4000 loss=4.9029 gn=0.68 lr=6.18e-05 tps=18167 mem=6.0G lam=None Gm=None
216[baseline s0] step 3880/4000 loss=4.9808 gn=0.79 lr=6.13e-05 tps=18171 mem=6.0G lam=None Gm=None
217[baseline s0] step 3900/4000 loss=4.8809 gn=0.78 lr=6.09e-05 tps=18174 mem=6.0G lam=None Gm=None
218[baseline s0] step 3920/4000 loss=4.8140 gn=0.76 lr=6.06e-05 tps=18178 mem=6.0G lam=None Gm=None
219[baseline s0] step 3940/4000 loss=4.9906 gn=0.77 lr=6.03e-05 tps=18182 mem=6.0G lam=None Gm=None
220[baseline s0] step 3960/4000 loss=4.9362 gn=0.75 lr=6.01e-05 tps=18185 mem=6.0G lam=None Gm=None
221[baseline s0] step 3980/4000 loss=4.8850 gn=0.76 lr=6.00e-05 tps=18189 mem=6.0G lam=None Gm=None
222[baseline s0] EVAL step 3999: val_loss=5.0002 val_ppl=148.44
223[init] variant=s_only seed=0 params=63.97M
224[init] use_S=True use_G=False
225[s_only s0] step 0/4000 loss=10.9196 gn=16.24 lr=3.00e-06 tps=7866 mem=5.6G lam=0.12692801654338837 Gm=None
226[s_only s0] step 20/4000 loss=9.8201 gn=1.89 lr=6.30e-05 tps=15859 mem=6.2G lam=0.1269285697489977 Gm=None
227[s_only s0] step 40/4000 loss=9.0759 gn=1.82 lr=1.23e-04 tps=16244 mem=6.2G lam=0.126861909404397 Gm=None
228[s_only s0] step 60/4000 loss=8.0906 gn=1.51 lr=1.83e-04 tps=16370 mem=6.2G lam=0.1267800386995077 Gm=None
229[s_only s0] step 80/4000 loss=7.4281 gn=0.74 lr=2.43e-04 tps=16458 mem=6.2G lam=0.12668094101051489 Gm=None
230[s_only s0] step 100/4000 loss=7.2174 gn=0.68 lr=3.03e-04 tps=16511 mem=6.2G lam=0.1265610239158074 Gm=None
231[s_only s0] step 120/4000 loss=7.0768 gn=0.80 lr=3.63e-04 tps=16546 mem=6.2G lam=0.12644990781943002 Gm=None
232[s_only s0] step 140/4000 loss=6.8256 gn=0.54 lr=4.23e-04 tps=16558 mem=6.2G lam=0.12640769282976785 Gm=None
233[s_only s0] step 160/4000 loss=6.9099 gn=0.79 lr=4.83e-04 tps=16572 mem=6.2G lam=0.12637964511911073 Gm=None
234[s_only s0] step 180/4000 loss=6.8124 gn=0.77 lr=5.43e-04 tps=16584 mem=6.2G lam=0.12634567295511565 Gm=None
235[s_only s0] step 200/4000 loss=6.6985 gn=0.77 lr=6.00e-04 tps=16593 mem=6.2G lam=0.12624020067354044 Gm=None
236[s_only s0] EVAL step 200: val_loss=6.7505 val_ppl=854.45
237[s_only s0] step 220/4000 loss=6.5877 gn=0.59 lr=6.00e-04 tps=16051 mem=6.2G lam=0.12625185896952948 Gm=None
238[s_only s0] step 240/4000 loss=6.6281 gn=0.54 lr=6.00e-04 tps=16097 mem=6.2G lam=0.1262657322610418 Gm=None
239[s_only s0] step 260/4000 loss=6.6135 gn=0.58 lr=6.00e-04 tps=16137 mem=6.2G lam=0.12628101371228695 Gm=None
240[s_only s0] step 280/4000 loss=6.5128 gn=0.54 lr=5.99e-04 tps=16173 mem=6.2G lam=0.12627421288440624 Gm=None
241[s_only s0] step 300/4000 loss=6.3416 gn=0.54 lr=5.99e-04 tps=16203 mem=6.2G lam=0.12628612201660872 Gm=None
242[s_only s0] step 320/4000 loss=6.4067 gn=0.48 lr=5.99e-04 tps=16231 mem=6.2G lam=0.1263035830731193 Gm=None
243[s_only s0] step 340/4000 loss=6.3989 gn=0.53 lr=5.98e-04 tps=16252 mem=6.2G lam=0.126261239250501 Gm=None
244[s_only s0] step 360/4000 loss=6.2496 gn=0.53 lr=5.98e-04 tps=16270 mem=6.2G lam=0.12632379370431104 Gm=None
245[s_only s0] step 380/4000 loss=6.2258 gn=0.55 lr=5.97e-04 tps=16290 mem=6.2G lam=0.12632035743445158 Gm=None
246[s_only s0] step 400/4000 loss=6.2672 gn=0.56 lr=5.96e-04 tps=16306 mem=6.2G lam=0.12638370475421348 Gm=None
247[s_only s0] EVAL step 400: val_loss=6.3981 val_ppl=600.70
248[s_only s0] step 420/4000 loss=6.3893 gn=0.77 lr=5.96e-04 tps=16042 mem=6.2G lam=0.1264245609442393 Gm=None
249[s_only s0] step 440/4000 loss=6.3583 gn=0.66 lr=5.95e-04 tps=16068 mem=6.2G lam=0.12652567743013302 Gm=None
250[s_only s0] step 460/4000 loss=6.2952 gn=0.53 lr=5.94e-04 tps=16091 mem=6.2G lam=0.1265407477815946 Gm=None
251[s_only s0] step 480/4000 loss=6.3867 gn=0.52 lr=5.93e-04 tps=16112 mem=6.2G lam=0.12667018050948778 Gm=None
252[s_only s0] step 500/4000 loss=6.3163 gn=0.54 lr=5.92e-04 tps=16132 mem=6.2G lam=0.1267355838790536 Gm=None
253[s_only s0] step 520/4000 loss=6.1707 gn=0.56 lr=5.91e-04 tps=16152 mem=6.2G lam=0.12688766761372486 Gm=None
254[s_only s0] step 540/4000 loss=6.1835 gn=0.53 lr=5.89e-04 tps=16170 mem=6.2G lam=0.12706252994636694 Gm=None
255[s_only s0] step 560/4000 loss=6.0407 gn=0.56 lr=5.88e-04 tps=16189 mem=6.2G lam=0.12717174366116524 Gm=None
256[s_only s0] step 580/4000 loss=6.2094 gn=0.59 lr=5.87e-04 tps=16205 mem=6.2G lam=0.12731845180193582 Gm=None
257[s_only s0] step 600/4000 loss=5.9602 gn=0.56 lr=5.85e-04 tps=16220 mem=6.2G lam=0.12746825131277242 Gm=None
258[s_only s0] EVAL step 600: val_loss=6.1718 val_ppl=479.03
259[s_only s0] step 620/4000 loss=6.0255 gn=0.50 lr=5.84e-04 tps=16046 mem=6.2G lam=0.1276232535019517 Gm=None
260[s_only s0] step 640/4000 loss=6.1234 gn=0.64 lr=5.82e-04 tps=16064 mem=6.2G lam=0.12782887276262045 Gm=None
261[s_only s0] step 660/4000 loss=5.9412 gn=0.59 lr=5.81e-04 tps=16079 mem=6.2G lam=0.12805543839931488 Gm=None
262[s_only s0] step 680/4000 loss=5.9505 gn=0.52 lr=5.79e-04 tps=16096 mem=6.2G lam=0.1282186039413015 Gm=None
263[s_only s0] step 700/4000 loss=6.0496 gn=0.58 lr=5.77e-04 tps=16110 mem=6.2G lam=0.12838227239747843 Gm=None
264[s_only s0] step 720/4000 loss=6.0181 gn=0.53 lr=5.75e-04 tps=16125 mem=6.2G lam=0.12860822770744562 Gm=None
265[s_only s0] step 740/4000 loss=6.0501 gn=0.58 lr=5.74e-04 tps=16138 mem=6.2G lam=0.1289047651613752 Gm=None
266[s_only s0] step 760/4000 loss=5.9148 gn=0.52 lr=5.72e-04 tps=16151 mem=6.2G lam=0.12920729536563158 Gm=None
267[s_only s0] step 780/4000 loss=5.9283 gn=0.52 lr=5.70e-04 tps=16163 mem=6.2G lam=0.12947439402341843 Gm=None
268[s_only s0] step 800/4000 loss=5.8881 gn=0.54 lr=5.67e-04 tps=16175 mem=6.2G lam=0.12977179295072952 Gm=None
269[s_only s0] EVAL step 800: val_loss=5.9815 val_ppl=396.01
270[s_only s0] step 820/4000 loss=5.9666 gn=0.53 lr=5.65e-04 tps=16042 mem=6.2G lam=0.13008342559138933 Gm=None
271[s_only s0] step 840/4000 loss=6.1374 gn=0.54 lr=5.63e-04 tps=16056 mem=6.2G lam=0.1303543041770657 Gm=None
272[s_only s0] step 860/4000 loss=5.9518 gn=0.56 lr=5.61e-04 tps=16070 mem=6.2G lam=0.13066580053418875 Gm=None
273[s_only s0] step 880/4000 loss=5.9320 gn=0.62 lr=5.58e-04 tps=16083 mem=6.2G lam=0.13096548213313022 Gm=None
274[s_only s0] step 900/4000 loss=5.8557 gn=0.52 lr=5.56e-04 tps=16094 mem=6.2G lam=0.1312484402830402 Gm=None
275[s_only s0] step 920/4000 loss=5.7370 gn=0.54 lr=5.54e-04 tps=16104 mem=6.2G lam=0.1315185483545065 Gm=None
276[s_only s0] step 940/4000 loss=5.7989 gn=0.55 lr=5.51e-04 tps=16115 mem=6.2G lam=0.13187734379122654 Gm=None
277[s_only s0] step 960/4000 loss=5.7840 gn=0.58 lr=5.48e-04 tps=16124 mem=6.2G lam=0.13223367060224214 Gm=None
278[s_only s0] step 980/4000 loss=5.7109 gn=0.57 lr=5.46e-04 tps=16134 mem=6.2G lam=0.13256636603424946 Gm=None
279[s_only s0] step 1000/4000 loss=5.8045 gn=0.58 lr=5.43e-04 tps=16143 mem=6.2G lam=0.13293919060379267 Gm=None
280[s_only s0] EVAL step 1000: val_loss=5.8303 val_ppl=340.47
281[s_only s0] step 1020/4000 loss=5.5516 gn=0.52 lr=5.40e-04 tps=16037 mem=6.2G lam=0.13332742483665547 Gm=None
282[s_only s0] step 1040/4000 loss=5.8040 gn=0.50 lr=5.37e-04 tps=16049 mem=6.2G lam=0.1336649345854918 Gm=None
283[s_only s0] step 1060/4000 loss=5.5133 gn=0.54 lr=5.35e-04 tps=16060 mem=6.2G lam=0.13397249206900597 Gm=None
284[s_only s0] step 1080/4000 loss=5.7270 gn=0.65 lr=5.32e-04 tps=16071 mem=6.2G lam=0.13436876454701027 Gm=None
285[s_only s0] step 1100/4000 loss=5.6774 gn=0.58 lr=5.29e-04 tps=16082 mem=6.2G lam=0.1346825072541833 Gm=None
286[s_only s0] step 1120/4000 loss=5.6185 gn=0.57 lr=5.26e-04 tps=16092 mem=6.2G lam=0.13510261848568916 Gm=None
287[s_only s0] step 1140/4000 loss=5.8675 gn=0.56 lr=5.22e-04 tps=16100 mem=6.2G lam=0.1354632107540965 Gm=None
288[s_only s0] step 1160/4000 loss=5.6197 gn=0.54 lr=5.19e-04 tps=16109 mem=6.2G lam=0.1358909821137786 Gm=None
289[s_only s0] step 1180/4000 loss=5.6770 gn=0.56 lr=5.16e-04 tps=16117 mem=6.2G lam=0.1362527391562859 Gm=None
290[s_only s0] step 1200/4000 loss=5.5601 gn=0.52 lr=5.13e-04 tps=16124 mem=6.2G lam=0.13655376713722944 Gm=None
291[s_only s0] EVAL step 1200: val_loss=5.6810 val_ppl=293.24
292[s_only s0] step 1220/4000 loss=5.5830 gn=0.60 lr=5.10e-04 tps=16036 mem=6.2G lam=0.13686958079536757 Gm=None
293[s_only s0] step 1240/4000 loss=5.4571 gn=0.53 lr=5.06e-04 tps=16045 mem=6.2G lam=0.13723452016711235 Gm=None
294[s_only s0] step 1260/4000 loss=5.4601 gn=0.56 lr=5.03e-04 tps=16053 mem=6.2G lam=0.1376482384900252 Gm=None
295[s_only s0] step 1280/4000 loss=5.5864 gn=0.55 lr=4.99e-04 tps=16062 mem=6.2G lam=0.13803729942689338 Gm=None
296[s_only s0] step 1300/4000 loss=5.6888 gn=0.61 lr=4.96e-04 tps=16070 mem=6.2G lam=0.1383867928137382 Gm=None
297[s_only s0] step 1320/4000 loss=5.6395 gn=0.58 lr=4.92e-04 tps=16079 mem=6.2G lam=0.1387265526379148 Gm=None
298[s_only s0] step 1340/4000 loss=5.7577 gn=0.57 lr=4.89e-04 tps=16087 mem=6.2G lam=0.13911849415550628 Gm=None
299[s_only s0] step 1360/4000 loss=5.5489 gn=0.58 lr=4.85e-04 tps=16095 mem=6.2G lam=0.13947931056221327 Gm=None
300[s_only s0] step 1380/4000 loss=5.4243 gn=0.57 lr=4.81e-04 tps=16101 mem=6.2G lam=0.13979739509522915 Gm=None
301[s_only s0] step 1400/4000 loss=5.4404 gn=0.58 lr=4.78e-04 tps=16106 mem=6.2G lam=0.14012548172225556 Gm=None
302[s_only s0] EVAL step 1400: val_loss=5.5705 val_ppl=262.55
303[s_only s0] step 1420/4000 loss=5.4244 gn=0.63 lr=4.74e-04 tps=16030 mem=6.2G lam=0.1404859513665239 Gm=None
304[s_only s0] step 1440/4000 loss=5.4084 gn=0.62 lr=4.70e-04 tps=16038 mem=6.2G lam=0.140826303511858 Gm=None
305[s_only s0] step 1460/4000 loss=5.5294 gn=0.63 lr=4.66e-04 tps=16046 mem=6.2G lam=0.14116969301054874 Gm=None
306[s_only s0] step 1480/4000 loss=5.5820 gn=0.57 lr=4.62e-04 tps=16053 mem=6.2G lam=0.14155861859520277 Gm=None
307[s_only s0] step 1500/4000 loss=5.4778 gn=0.63 lr=4.59e-04 tps=16061 mem=6.2G lam=0.1419260473921895 Gm=None
308[s_only s0] step 1520/4000 loss=5.4672 gn=0.58 lr=4.55e-04 tps=16067 mem=6.2G lam=0.14222985816498598 Gm=None
309[s_only s0] step 1540/4000 loss=5.3713 gn=0.61 lr=4.51e-04 tps=16074 mem=6.2G lam=0.14253480080515146 Gm=None
310[s_only s0] step 1560/4000 loss=5.3767 gn=0.57 lr=4.47e-04 tps=16081 mem=6.2G lam=0.14288722444325686 Gm=None
311[s_only s0] step 1580/4000 loss=5.4439 gn=0.65 lr=4.43e-04 tps=16088 mem=6.2G lam=0.14320783875882626 Gm=None
312[s_only s0] step 1600/4000 loss=5.4300 gn=0.57 lr=4.38e-04 tps=16093 mem=6.2G lam=0.14355705057581267 Gm=None
313[s_only s0] EVAL step 1600: val_loss=5.4941 val_ppl=243.26
314[s_only s0] step 1620/4000 loss=5.4115 gn=0.64 lr=4.34e-04 tps=16027 mem=6.2G lam=0.14388109495242438 Gm=None
315[s_only s0] step 1640/4000 loss=5.2882 gn=0.59 lr=4.30e-04 tps=16034 mem=6.2G lam=0.1442484213039279 Gm=None
316[s_only s0] step 1660/4000 loss=5.4577 gn=0.59 lr=4.26e-04 tps=16041 mem=6.2G lam=0.14454306599994501 Gm=None
317[s_only s0] step 1680/4000 loss=5.6048 gn=0.59 lr=4.22e-04 tps=16047 mem=6.2G lam=0.14485854376107454 Gm=None
318[s_only s0] step 1700/4000 loss=5.4706 gn=0.58 lr=4.18e-04 tps=16053 mem=6.2G lam=0.14520258642733097 Gm=None
319[s_only s0] step 1720/4000 loss=5.2282 gn=0.60 lr=4.13e-04 tps=16059 mem=6.2G lam=0.14549836268027624 Gm=None
320[s_only s0] step 1740/4000 loss=5.2964 gn=0.60 lr=4.09e-04 tps=16065 mem=6.2G lam=0.1458452930673957 Gm=None
321[s_only s0] step 1760/4000 loss=5.3622 gn=0.61 lr=4.05e-04 tps=16072 mem=6.2G lam=0.14619173482060432 Gm=None
322[s_only s0] step 1780/4000 loss=5.2387 gn=0.63 lr=4.01e-04 tps=16077 mem=6.2G lam=0.14650953530023494 Gm=None
323[s_only s0] step 1800/4000 loss=5.2973 gn=0.61 lr=3.96e-04 tps=16083 mem=6.2G lam=0.14679855667054653 Gm=None
324[s_only s0] EVAL step 1800: val_loss=5.4237 val_ppl=226.71
325[s_only s0] step 1820/4000 loss=5.3838 gn=0.56 lr=3.92e-04 tps=16025 mem=6.2G lam=0.14706969012816748 Gm=None
326[s_only s0] step 1840/4000 loss=5.2330 gn=0.63 lr=3.88e-04 tps=16030 mem=6.2G lam=0.1473601134493947 Gm=None
327[s_only s0] step 1860/4000 loss=5.2746 gn=0.59 lr=3.83e-04 tps=16036 mem=6.2G lam=0.14768042291204134 Gm=None
328[s_only s0] step 1880/4000 loss=5.3684 gn=0.65 lr=3.79e-04 tps=16042 mem=6.2G lam=0.14795536641031504 Gm=None
329[s_only s0] step 1900/4000 loss=5.3387 gn=0.66 lr=3.74e-04 tps=16047 mem=6.2G lam=0.14826269975552955 Gm=None
330[s_only s0] step 1920/4000 loss=5.2729 gn=0.60 lr=3.70e-04 tps=16053 mem=6.2G lam=0.1485254509995381 Gm=None
331[s_only s0] step 1940/4000 loss=5.3370 gn=0.62 lr=3.66e-04 tps=16058 mem=6.2G lam=0.1487580065925916 Gm=None
332[s_only s0] step 1960/4000 loss=5.5228 gn=0.66 lr=3.61e-04 tps=16063 mem=6.2G lam=0.1491000996902585 Gm=None
333[s_only s0] step 1980/4000 loss=5.4326 gn=0.62 lr=3.57e-04 tps=16068 mem=6.2G lam=0.1493905217697223 Gm=None
334[s_only s0] step 2000/4000 loss=5.2669 gn=0.60 lr=3.52e-04 tps=16074 mem=6.2G lam=0.14968158894528946 Gm=None
335[s_only s0] EVAL step 2000: val_loss=5.3785 val_ppl=216.70
336[s_only s0] step 2020/4000 loss=5.2263 gn=0.59 lr=3.48e-04 tps=16022 mem=6.2G lam=0.14995385675380626 Gm=None
337[s_only s0] step 2040/4000 loss=5.3614 gn=0.67 lr=3.43e-04 tps=16027 mem=6.2G lam=0.15023998698840538 Gm=None
338[s_only s0] step 2060/4000 loss=5.1188 gn=0.67 lr=3.39e-04 tps=16033 mem=6.2G lam=0.15052441755930582 Gm=None
339[s_only s0] step 2080/4000 loss=5.2630 gn=0.63 lr=3.34e-04 tps=16038 mem=6.2G lam=0.15079285173366466 Gm=None
340[s_only s0] step 2100/4000 loss=5.2620 gn=0.60 lr=3.30e-04 tps=16043 mem=6.2G lam=0.15108702021340528 Gm=None
341[s_only s0] step 2120/4000 loss=5.1780 gn=0.61 lr=3.26e-04 tps=16047 mem=6.2G lam=0.15135678462684155 Gm=None
342[s_only s0] step 2140/4000 loss=5.0995 gn=0.63 lr=3.21e-04 tps=16052 mem=6.2G lam=0.15164329887678227 Gm=None
343[s_only s0] step 2160/4000 loss=5.1503 gn=0.62 lr=3.17e-04 tps=16057 mem=6.2G lam=0.15192982678612074 Gm=None
344[s_only s0] step 2180/4000 loss=5.1897 gn=0.68 lr=3.12e-04 tps=16061 mem=6.2G lam=0.152193579201897 Gm=None
345[s_only s0] step 2200/4000 loss=5.3355 gn=0.63 lr=3.08e-04 tps=16066 mem=6.2G lam=0.15246522768090168 Gm=None
346[s_only s0] EVAL step 2200: val_loss=5.2928 val_ppl=198.90
347[s_only s0] step 2220/4000 loss=5.2258 gn=0.65 lr=3.03e-04 tps=16018 mem=6.2G lam=0.1527368761599064 Gm=None
348[s_only s0] step 2240/4000 loss=5.1550 gn=0.60 lr=2.99e-04 tps=16023 mem=6.2G lam=0.15297844167798758 Gm=None
349[s_only s0] step 2260/4000 loss=5.1668 gn=0.59 lr=2.94e-04 tps=16028 mem=6.2G lam=0.15325090506424507 Gm=None
350[s_only s0] step 2280/4000 loss=5.3703 gn=0.73 lr=2.90e-04 tps=16032 mem=6.2G lam=0.15346690950294337 Gm=None
351[s_only s0] step 2300/4000 loss=5.1344 gn=0.68 lr=2.86e-04 tps=16037 mem=6.2G lam=0.15374540848036608 Gm=None
352[s_only s0] step 2320/4000 loss=5.2560 gn=0.68 lr=2.81e-04 tps=16041 mem=6.2G lam=0.15396491096665463 Gm=None
353[s_only s0] step 2340/4000 loss=5.1864 gn=0.67 lr=2.77e-04 tps=16046 mem=6.2G lam=0.1541658298422893 Gm=None
354[s_only s0] step 2360/4000 loss=5.2709 gn=0.64 lr=2.72e-04 tps=16050 mem=6.2G lam=0.15444632650663456 Gm=None
355[s_only s0] step 2380/4000 loss=5.1435 gn=0.71 lr=2.68e-04 tps=16054 mem=6.2G lam=0.15467509596298137 Gm=None
356[s_only s0] step 2400/4000 loss=5.0452 gn=0.66 lr=2.64e-04 tps=16059 mem=6.2G lam=0.1549065876752138 Gm=None
357[s_only s0] EVAL step 2400: val_loss=5.2244 val_ppl=185.75
358[s_only s0] step 2420/4000 loss=5.0378 gn=0.65 lr=2.59e-04 tps=16015 mem=6.2G lam=0.1551545119533936 Gm=None
359[s_only s0] step 2440/4000 loss=5.1605 gn=0.67 lr=2.55e-04 tps=16020 mem=6.2G lam=0.15540442988276482 Gm=None
360[s_only s0] step 2460/4000 loss=5.2340 gn=0.67 lr=2.51e-04 tps=16024 mem=6.2G lam=0.15560725517570972 Gm=None
361[s_only s0] step 2480/4000 loss=5.1356 gn=0.63 lr=2.47e-04 tps=16029 mem=6.2G lam=0.15582442035277685 Gm=None
362[s_only s0] step 2500/4000 loss=4.9740 gn=0.65 lr=2.42e-04 tps=16033 mem=6.2G lam=0.15603429606805244 Gm=None
363[s_only s0] step 2520/4000 loss=5.0311 gn=0.63 lr=2.38e-04 tps=16038 mem=6.2G lam=0.15623840938011804 Gm=None
364[s_only s0] step 2540/4000 loss=5.1130 gn=0.64 lr=2.34e-04 tps=16042 mem=6.2G lam=0.15647442545741796 Gm=None
365[s_only s0] step 2560/4000 loss=4.9190 gn=0.71 lr=2.30e-04 tps=16047 mem=6.2G lam=0.15665741016467413 Gm=None
366[s_only s0] step 2580/4000 loss=5.1944 gn=0.71 lr=2.26e-04 tps=16051 mem=6.2G lam=0.15685578901320696 Gm=None
367[s_only s0] step 2600/4000 loss=4.9774 gn=0.65 lr=2.22e-04 tps=16055 mem=6.2G lam=0.1570213104908665 Gm=None
368[s_only s0] EVAL step 2600: val_loss=5.1874 val_ppl=179.00
369[s_only s0] step 2620/4000 loss=5.1547 gn=0.72 lr=2.17e-04 tps=16014 mem=6.2G lam=0.15720977572103342 Gm=None
370[s_only s0] step 2640/4000 loss=5.1468 gn=0.77 lr=2.13e-04 tps=16018 mem=6.2G lam=0.1574209677055478 Gm=None
371[s_only s0] step 2660/4000 loss=5.0151 gn=0.65 lr=2.09e-04 tps=16023 mem=6.2G lam=0.15761489141732454 Gm=None
372[s_only s0] step 2680/4000 loss=4.8517 gn=0.86 lr=2.05e-04 tps=16027 mem=6.2G lam=0.15779203393807015 Gm=None
373[s_only s0] step 2700/4000 loss=5.1029 gn=0.68 lr=2.01e-04 tps=16031 mem=6.2G lam=0.1579585187137127 Gm=None
374[s_only s0] step 2720/4000 loss=5.0902 gn=0.68 lr=1.98e-04 tps=16035 mem=6.2G lam=0.15815662002811828 Gm=None
375[s_only s0] step 2740/4000 loss=5.0854 gn=0.65 lr=1.94e-04 tps=16039 mem=6.2G lam=0.15834523302813372 Gm=None
376[s_only s0] step 2760/4000 loss=4.9313 gn=0.66 lr=1.90e-04 tps=16043 mem=6.2G lam=0.15851814641306797 Gm=None
377[s_only s0] step 2780/4000 loss=5.0645 gn=0.64 lr=1.86e-04 tps=16046 mem=6.2G lam=0.15868712433924279 Gm=None
378[s_only s0] step 2800/4000 loss=5.0114 gn=0.72 lr=1.82e-04 tps=16050 mem=6.2G lam=0.15886648620168367 Gm=None
379[s_only s0] EVAL step 2800: val_loss=5.1255 val_ppl=168.25
380[s_only s0] step 2820/4000 loss=5.0534 gn=0.65 lr=1.79e-04 tps=16013 mem=6.2G lam=0.1590363153566917 Gm=None
381[s_only s0] step 2840/4000 loss=4.8508 gn=0.69 lr=1.75e-04 tps=16017 mem=6.2G lam=0.1591800656169653 Gm=None
382[s_only s0] step 2860/4000 loss=5.0366 gn=0.64 lr=1.71e-04 tps=16020 mem=6.2G lam=0.15933304311086735 Gm=None
383[s_only s0] step 2880/4000 loss=5.0254 gn=0.66 lr=1.68e-04 tps=16024 mem=6.2G lam=0.15949062475313744 Gm=None
384[s_only s0] step 2900/4000 loss=4.8737 gn=0.72 lr=1.64e-04 tps=16028 mem=6.2G lam=0.15964605690290531 Gm=None
385[s_only s0] step 2920/4000 loss=5.0762 gn=0.74 lr=1.61e-04 tps=16032 mem=6.2G lam=0.1597970174625516 Gm=None
386[s_only s0] step 2940/4000 loss=4.9907 gn=0.68 lr=1.57e-04 tps=16036 mem=6.2G lam=0.15993541137625775 Gm=None
387[s_only s0] step 2960/4000 loss=4.9784 gn=0.68 lr=1.54e-04 tps=16040 mem=6.2G lam=0.16006658226251602 Gm=None
388[s_only s0] step 2980/4000 loss=4.9393 gn=0.64 lr=1.50e-04 tps=16044 mem=6.2G lam=0.16020564020921788 Gm=None
389[s_only s0] step 3000/4000 loss=4.8949 gn=0.69 lr=1.47e-04 tps=16048 mem=6.2G lam=0.1603458309546113 Gm=None
390[s_only s0] EVAL step 3000: val_loss=5.1437 val_ppl=171.34
391[s_only s0] step 3020/4000 loss=4.9444 gn=0.68 lr=1.44e-04 tps=16013 mem=6.2G lam=0.1604727047185103 Gm=None
392[s_only s0] step 3040/4000 loss=5.0316 gn=0.69 lr=1.41e-04 tps=16016 mem=6.2G lam=0.16060379706323147 Gm=None
393[s_only s0] step 3060/4000 loss=4.9870 gn=0.74 lr=1.38e-04 tps=16020 mem=6.2G lam=0.1607404612004757 Gm=None
394[s_only s0] step 3080/4000 loss=4.8295 gn=0.66 lr=1.34e-04 tps=16023 mem=6.2G lam=0.16087450087070465 Gm=None
395[s_only s0] step 3100/4000 loss=5.0511 gn=0.69 lr=1.31e-04 tps=16027 mem=6.2G lam=0.16098162159323692 Gm=None
396[s_only s0] step 3120/4000 loss=5.0937 gn=1.61 lr=1.28e-04 tps=16030 mem=6.2G lam=0.1611141012981534 Gm=None
397[s_only s0] step 3140/4000 loss=4.9128 gn=0.69 lr=1.25e-04 tps=16034 mem=6.2G lam=0.16125171103825173 Gm=None
398[s_only s0] step 3160/4000 loss=5.0753 gn=0.70 lr=1.23e-04 tps=16037 mem=6.2G lam=0.16136677749454975 Gm=None
399[s_only s0] step 3180/4000 loss=4.8985 gn=0.65 lr=1.20e-04 tps=16041 mem=6.2G lam=0.16147833845267692 Gm=None
400[s_only s0] step 3200/4000 loss=4.9372 gn=0.73 lr=1.17e-04 tps=16044 mem=6.2G lam=0.16159397643059492 Gm=None
401[s_only s0] EVAL step 3200: val_loss=5.0508 val_ppl=156.14
402[s_only s0] step 3220/4000 loss=5.0421 gn=0.77 lr=1.14e-04 tps=16012 mem=6.2G lam=0.16170350648462772 Gm=None
403[s_only s0] step 3240/4000 loss=5.0556 gn=0.68 lr=1.12e-04 tps=16015 mem=6.2G lam=0.16180988680571318 Gm=None
404[s_only s0] step 3260/4000 loss=4.8043 gn=0.69 lr=1.09e-04 tps=16018 mem=6.2G lam=0.16191981422404447 Gm=None
405[s_only s0] step 3280/4000 loss=5.0738 gn=0.77 lr=1.06e-04 tps=16022 mem=6.2G lam=0.1620280851299564 Gm=None
406[s_only s0] step 3300/4000 loss=4.8867 gn=0.73 lr=1.04e-04 tps=16025 mem=6.2G lam=0.16213232837617397 Gm=None
407[s_only s0] step 3320/4000 loss=5.0384 gn=0.73 lr=1.02e-04 tps=16028 mem=6.2G lam=0.16222489128510156 Gm=None
408[s_only s0] step 3340/4000 loss=4.8808 gn=0.68 lr=9.92e-05 tps=16031 mem=6.2G lam=0.16232431990404925 Gm=None
409[s_only s0] step 3360/4000 loss=5.0324 gn=0.69 lr=9.69e-05 tps=16034 mem=6.2G lam=0.162418513558805 Gm=None
410[s_only s0] step 3380/4000 loss=5.0188 gn=0.73 lr=9.47e-05 tps=16038 mem=6.2G lam=0.16249898045013347 Gm=None
411[s_only s0] step 3400/4000 loss=4.8903 gn=0.68 lr=9.25e-05 tps=16041 mem=6.2G lam=0.1626018676906824 Gm=None
412[s_only s0] EVAL step 3400: val_loss=5.0540 val_ppl=156.65
413[s_only s0] step 3420/4000 loss=4.9338 gn=0.71 lr=9.04e-05 tps=16010 mem=6.2G lam=0.1626981602360805 Gm=None
414[s_only s0] step 3440/4000 loss=4.9736 gn=0.79 lr=8.84e-05 tps=16014 mem=6.2G lam=0.16279214837898812 Gm=None
415[s_only s0] step 3460/4000 loss=4.8672 gn=0.74 lr=8.65e-05 tps=16017 mem=6.2G lam=0.1628744974732399 Gm=None
416[s_only s0] step 3480/4000 loss=4.9579 gn=0.74 lr=8.46e-05 tps=16020 mem=6.2G lam=0.16295635203520456 Gm=None
417[s_only s0] step 3500/4000 loss=4.8617 gn=0.71 lr=8.27e-05 tps=16023 mem=6.2G lam=0.16303630731999874 Gm=None
418[s_only s0] step 3520/4000 loss=4.7774 gn=0.74 lr=8.10e-05 tps=16027 mem=6.2G lam=0.16311296044538418 Gm=None
419[s_only s0] step 3540/4000 loss=4.7582 gn=0.70 lr=7.93e-05 tps=16030 mem=6.2G lam=0.16320041629175344 Gm=None
420[s_only s0] step 3560/4000 loss=5.0014 gn=0.70 lr=7.77e-05 tps=16033 mem=6.2G lam=0.16328275483101606 Gm=None
421[s_only s0] step 3580/4000 loss=4.8396 gn=0.69 lr=7.61e-05 tps=16036 mem=6.2G lam=0.163363772444427 Gm=None
422[s_only s0] step 3600/4000 loss=4.8693 gn=0.80 lr=7.46e-05 tps=16039 mem=6.2G lam=0.16343969789644083 Gm=None
423[s_only s0] EVAL step 3600: val_loss=5.0216 val_ppl=151.66
424[s_only s0] step 3620/4000 loss=4.9188 gn=0.75 lr=7.32e-05 tps=16009 mem=6.2G lam=0.16351322084665298 Gm=None
425[s_only s0] step 3640/4000 loss=4.9910 gn=0.72 lr=7.19e-05 tps=16012 mem=6.2G lam=0.16358474300553402 Gm=None
426[s_only s0] step 3660/4000 loss=5.0379 gn=0.80 lr=7.06e-05 tps=16016 mem=6.2G lam=0.16365407779812813 Gm=None
427[s_only s0] step 3680/4000 loss=4.9527 gn=0.71 lr=6.94e-05 tps=16019 mem=6.2G lam=0.16372491512447596 Gm=None
428[s_only s0] step 3700/4000 loss=5.0138 gn=0.69 lr=6.83e-05 tps=16021 mem=6.2G lam=0.16379747508714596 Gm=None
429[s_only s0] step 3720/4000 loss=4.7727 gn=0.70 lr=6.72e-05 tps=16024 mem=6.2G lam=0.16386881253371635 Gm=None
430[s_only s0] step 3740/4000 loss=4.9555 gn=0.71 lr=6.62e-05 tps=16027 mem=6.2G lam=0.1639383789151907 Gm=None
431[s_only s0] step 3760/4000 loss=4.9015 gn=0.71 lr=6.53e-05 tps=16030 mem=6.2G lam=0.16399934142827988 Gm=None
432[s_only s0] step 3780/4000 loss=4.8116 gn=0.73 lr=6.45e-05 tps=16033 mem=6.2G lam=0.16407388728111982 Gm=None
433[s_only s0] step 3800/4000 loss=4.9232 gn=0.77 lr=6.37e-05 tps=16036 mem=6.2G lam=0.16412982872376838 Gm=None
434[s_only s0] EVAL step 3800: val_loss=5.0013 val_ppl=148.60
435[s_only s0] step 3820/4000 loss=4.9799 gn=0.76 lr=6.30e-05 tps=16006 mem=6.2G lam=0.1641976566364368 Gm=None
436[s_only s0] step 3840/4000 loss=4.8974 gn=0.75 lr=6.24e-05 tps=16008 mem=6.2G lam=0.16426111416270336 Gm=None
437[s_only s0] step 3860/4000 loss=4.9083 gn=0.70 lr=6.18e-05 tps=16009 mem=6.2G lam=0.1643272421012322 Gm=None
438[s_only s0] step 3880/4000 loss=4.9749 gn=0.78 lr=6.13e-05 tps=16011 mem=6.2G lam=0.1643884942556421 Gm=None
439[s_only s0] step 3900/4000 loss=4.8772 gn=0.77 lr=6.09e-05 tps=16012 mem=6.2G lam=0.164447876624763 Gm=None
440[s_only s0] step 3920/4000 loss=4.8052 gn=0.78 lr=6.06e-05 tps=16013 mem=6.2G lam=0.16450831635544697 Gm=None
441[s_only s0] step 3940/4000 loss=4.9839 gn=0.77 lr=6.03e-05 tps=16014 mem=6.2G lam=0.1645667472233375 Gm=None
442[s_only s0] step 3960/4000 loss=4.9248 gn=0.77 lr=6.01e-05 tps=16014 mem=6.2G lam=0.16462658531963825 Gm=None
443[s_only s0] step 3980/4000 loss=4.8933 gn=0.82 lr=6.00e-05 tps=16014 mem=6.2G lam=0.16468850150704384 Gm=None
444[s_only s0] EVAL step 3999: val_loss=4.9965 val_ppl=147.89
445[init] variant=g_only seed=0 params=63.39M
446[init] use_S=False use_G=True
447[g_only s0] step 0/4000 loss=10.9116 gn=10.42 lr=3.00e-06 tps=7314 mem=5.8G lam=None Gm=0.4969075520833333
448[g_only s0] step 20/4000 loss=9.8400 gn=2.09 lr=6.30e-05 tps=16617 mem=6.3G lam=None Gm=0.4551595052083333
449[g_only s0] step 40/4000 loss=9.0910 gn=1.78 lr=1.23e-04 tps=17205 mem=6.3G lam=None Gm=0.4195963541666667
450[g_only s0] step 60/4000 loss=8.0935 gn=1.55 lr=1.83e-04 tps=17418 mem=6.3G lam=None Gm=0.4182942708333333
451[g_only s0] step 80/4000 loss=7.4238 gn=0.78 lr=2.43e-04 tps=17527 mem=6.3G lam=None Gm=0.4287923177083333
452[g_only s0] step 100/4000 loss=7.2057 gn=0.72 lr=3.03e-04 tps=17595 mem=6.3G lam=None Gm=0.4186197916666667
453[g_only s0] step 120/4000 loss=7.0529 gn=0.57 lr=3.63e-04 tps=17643 mem=6.3G lam=None Gm=0.4054361979166667
454[g_only s0] step 140/4000 loss=6.8452 gn=0.84 lr=4.23e-04 tps=17671 mem=6.3G lam=None Gm=0.4053548177083333
455[g_only s0] step 160/4000 loss=6.9112 gn=0.60 lr=4.83e-04 tps=17691 mem=6.3G lam=None Gm=0.4039713541666667
456[g_only s0] step 180/4000 loss=6.7947 gn=0.49 lr=5.43e-04 tps=17707 mem=6.3G lam=None Gm=0.3907063802083333
457[g_only s0] step 200/4000 loss=6.6789 gn=0.54 lr=6.00e-04 tps=17716 mem=6.3G lam=None Gm=0.4029947916666667
458[g_only s0] EVAL step 200: val_loss=6.7469 val_ppl=851.43
459[g_only s0] step 220/4000 loss=6.5800 gn=0.53 lr=6.00e-04 tps=17149 mem=6.3G lam=None Gm=0.39599609375
460[g_only s0] step 240/4000 loss=6.6272 gn=0.50 lr=6.00e-04 tps=17204 mem=6.3G lam=None Gm=0.3841145833333333
461[g_only s0] step 260/4000 loss=6.5962 gn=0.51 lr=6.00e-04 tps=17252 mem=6.3G lam=None Gm=0.3935546875
462[g_only s0] step 280/4000 loss=6.5161 gn=0.47 lr=5.99e-04 tps=17292 mem=6.3G lam=None Gm=0.3889973958333333
463[g_only s0] step 300/4000 loss=6.3437 gn=0.49 lr=5.99e-04 tps=17325 mem=6.3G lam=None Gm=0.4014485677083333
464[g_only s0] step 320/4000 loss=6.4121 gn=0.48 lr=5.99e-04 tps=17354 mem=6.3G lam=None Gm=0.4002278645833333
465[g_only s0] step 340/4000 loss=6.3941 gn=0.46 lr=5.98e-04 tps=17380 mem=6.3G lam=None Gm=0.3993326822916667
466[g_only s0] step 360/4000 loss=6.2629 gn=0.47 lr=5.98e-04 tps=17403 mem=6.3G lam=None Gm=0.3941243489583333
467[g_only s0] step 380/4000 loss=6.2178 gn=0.50 lr=5.97e-04 tps=17425 mem=6.3G lam=None Gm=0.4176432291666667
468[g_only s0] step 400/4000 loss=6.2702 gn=0.50 lr=5.96e-04 tps=17444 mem=6.3G lam=None Gm=0.41162109375
469[g_only s0] EVAL step 400: val_loss=6.4036 val_ppl=604.04
470[g_only s0] step 420/4000 loss=6.4023 gn=0.70 lr=5.96e-04 tps=17163 mem=6.3G lam=None Gm=0.4402669270833333
471[g_only s0] step 440/4000 loss=6.3540 gn=0.56 lr=5.95e-04 tps=17192 mem=6.3G lam=None Gm=0.4210611979166667
472[g_only s0] step 460/4000 loss=6.3021 gn=0.47 lr=5.94e-04 tps=17216 mem=6.3G lam=None Gm=0.416015625
473[g_only s0] step 480/4000 loss=6.3700 gn=0.49 lr=5.93e-04 tps=17240 mem=6.3G lam=None Gm=0.4447428385416667
474[g_only s0] step 500/4000 loss=6.3026 gn=0.49 lr=5.92e-04 tps=17262 mem=6.3G lam=None Gm=0.44580078125
475[g_only s0] step 520/4000 loss=6.1557 gn=0.47 lr=5.91e-04 tps=17283 mem=6.3G lam=None Gm=0.4806315104166667
476[g_only s0] step 540/4000 loss=6.1738 gn=0.46 lr=5.89e-04 tps=17302 mem=6.3G lam=None Gm=0.4768880208333333
477[g_only s0] step 560/4000 loss=6.0412 gn=0.56 lr=5.88e-04 tps=17321 mem=6.3G lam=None Gm=0.4951985677083333
478[g_only s0] step 580/4000 loss=6.2063 gn=0.54 lr=5.87e-04 tps=17338 mem=6.3G lam=None Gm=0.5145670572916666
479[g_only s0] step 600/4000 loss=5.9606 gn=0.48 lr=5.85e-04 tps=17353 mem=6.3G lam=None Gm=0.526123046875
480[g_only s0] EVAL step 600: val_loss=6.1682 val_ppl=477.31
481[g_only s0] step 620/4000 loss=6.0180 gn=0.48 lr=5.84e-04 tps=17168 mem=6.3G lam=None Gm=0.5377604166666666
482[g_only s0] step 640/4000 loss=6.1040 gn=0.61 lr=5.82e-04 tps=17189 mem=6.3G lam=None Gm=0.5563151041666666
483[g_only s0] step 660/4000 loss=5.9488 gn=0.48 lr=5.81e-04 tps=17208 mem=6.3G lam=None Gm=0.5437825520833334
484[g_only s0] step 680/4000 loss=5.9541 gn=0.48 lr=5.79e-04 tps=17225 mem=6.3G lam=None Gm=0.5411783854166666
485[g_only s0] step 700/4000 loss=6.0356 gn=0.49 lr=5.77e-04 tps=17242 mem=6.3G lam=None Gm=0.581298828125
486[g_only s0] step 720/4000 loss=6.0191 gn=0.51 lr=5.75e-04 tps=17258 mem=6.3G lam=None Gm=0.5633138020833334
487[g_only s0] step 740/4000 loss=6.0332 gn=0.50 lr=5.74e-04 tps=17273 mem=6.3G lam=None Gm=0.5607096354166666
488[g_only s0] step 760/4000 loss=5.9036 gn=0.51 lr=5.72e-04 tps=17287 mem=6.3G lam=None Gm=0.5413411458333334
489[g_only s0] step 780/4000 loss=5.9232 gn=0.51 lr=5.70e-04 tps=17301 mem=6.3G lam=None Gm=0.5846354166666666
490[g_only s0] step 800/4000 loss=5.8875 gn=0.56 lr=5.67e-04 tps=17314 mem=6.3G lam=None Gm=0.596923828125
491[g_only s0] EVAL step 800: val_loss=5.9858 val_ppl=397.74
492[g_only s0] step 820/4000 loss=5.9567 gn=0.45 lr=5.65e-04 tps=17172 mem=6.3G lam=None Gm=0.5618489583333334
493[g_only s0] step 840/4000 loss=6.1340 gn=0.50 lr=5.63e-04 tps=17187 mem=6.3G lam=None Gm=0.5763346354166666
494[g_only s0] step 860/4000 loss=5.9471 gn=0.53 lr=5.61e-04 tps=17202 mem=6.3G lam=None Gm=0.5804036458333334
495[g_only s0] step 880/4000 loss=5.9260 gn=0.53 lr=5.58e-04 tps=17214 mem=6.3G lam=None Gm=0.5691731770833334
496[g_only s0] step 900/4000 loss=5.8427 gn=0.48 lr=5.56e-04 tps=17225 mem=6.3G lam=None Gm=0.5870768229166666
497[g_only s0] step 920/4000 loss=5.7483 gn=0.53 lr=5.54e-04 tps=17235 mem=6.3G lam=None Gm=0.591552734375
498[g_only s0] step 940/4000 loss=5.7881 gn=0.49 lr=5.51e-04 tps=17243 mem=6.3G lam=None Gm=0.5843912760416666
499[g_only s0] step 960/4000 loss=5.7854 gn=0.57 lr=5.48e-04 tps=17250 mem=6.3G lam=None Gm=0.602783203125
500[g_only s0] step 980/4000 loss=5.7003 gn=0.50 lr=5.46e-04 tps=17255 mem=6.3G lam=None Gm=0.5874837239583334
501[g_only s0] step 1000/4000 loss=5.7974 gn=0.52 lr=5.43e-04 tps=17263 mem=6.3G lam=None Gm=0.5813802083333334
502[g_only s0] EVAL step 1000: val_loss=5.8254 val_ppl=338.79
503[g_only s0] step 1020/4000 loss=5.5420 gn=0.50 lr=5.40e-04 tps=17146 mem=6.3G lam=None Gm=0.577880859375
504[g_only s0] step 1040/4000 loss=5.7812 gn=0.48 lr=5.37e-04 tps=17156 mem=6.3G lam=None Gm=0.5787353515625
505[g_only s0] step 1060/4000 loss=5.4968 gn=0.50 lr=5.35e-04 tps=17165 mem=6.3G lam=None Gm=0.5699869791666666
506[g_only s0] step 1080/4000 loss=5.7063 gn=0.53 lr=5.32e-04 tps=17175 mem=6.3G lam=None Gm=0.5565592447916666
507[g_only s0] step 1100/4000 loss=5.6721 gn=0.49 lr=5.29e-04 tps=17183 mem=6.3G lam=None Gm=0.5721435546875
508[g_only s0] step 1120/4000 loss=5.6122 gn=0.53 lr=5.26e-04 tps=17192 mem=6.3G lam=None Gm=0.5660807291666666
509[g_only s0] step 1140/4000 loss=5.8697 gn=0.54 lr=5.22e-04 tps=17200 mem=6.3G lam=None Gm=0.5578206380208334
510[g_only s0] step 1160/4000 loss=5.6152 gn=0.52 lr=5.19e-04 tps=17208 mem=6.3G lam=None Gm=0.5635579427083334
511[g_only s0] step 1180/4000 loss=5.6635 gn=0.51 lr=5.16e-04 tps=17216 mem=6.3G lam=None Gm=0.5602213541666666
512[g_only s0] step 1200/4000 loss=5.5461 gn=0.51 lr=5.13e-04 tps=17223 mem=6.3G lam=None Gm=0.5563557942708334
513[g_only s0] EVAL step 1200: val_loss=5.6726 val_ppl=290.78
514[g_only s0] step 1220/4000 loss=5.5768 gn=0.53 lr=5.10e-04 tps=17129 mem=6.3G lam=None Gm=0.5413818359375
515[g_only s0] step 1240/4000 loss=5.4647 gn=0.51 lr=5.06e-04 tps=17137 mem=6.3G lam=None Gm=0.5687255859375
516[g_only s0] step 1260/4000 loss=5.4552 gn=0.52 lr=5.03e-04 tps=17146 mem=6.3G lam=None Gm=0.5423177083333334
517[g_only s0] step 1280/4000 loss=5.5719 gn=0.58 lr=4.99e-04 tps=17154 mem=6.3G lam=None Gm=0.5581461588541666
518[g_only s0] step 1300/4000 loss=5.6728 gn=0.56 lr=4.96e-04 tps=17162 mem=6.3G lam=None Gm=0.5564371744791666
519[g_only s0] step 1320/4000 loss=5.6312 gn=0.57 lr=4.92e-04 tps=17169 mem=6.3G lam=None Gm=0.5655517578125
520[g_only s0] step 1340/4000 loss=5.7253 gn=0.54 lr=4.89e-04 tps=17176 mem=6.3G lam=None Gm=0.5574544270833334
521[g_only s0] step 1360/4000 loss=5.5400 gn=0.56 lr=4.85e-04 tps=17184 mem=6.3G lam=None Gm=0.5557861328125
522[g_only s0] step 1380/4000 loss=5.4131 gn=0.54 lr=4.81e-04 tps=17192 mem=6.3G lam=None Gm=0.5404052734375
523[g_only s0] step 1400/4000 loss=5.4202 gn=0.61 lr=4.78e-04 tps=17198 mem=6.3G lam=None Gm=0.5691731770833334
524[g_only s0] EVAL step 1400: val_loss=5.5586 val_ppl=259.45
525[g_only s0] step 1420/4000 loss=5.4152 gn=0.60 lr=4.74e-04 tps=17118 mem=6.3G lam=None Gm=0.5450846354166666
526[g_only s0] step 1440/4000 loss=5.3993 gn=0.56 lr=4.70e-04 tps=17127 mem=6.3G lam=None Gm=0.5409342447916666
527[g_only s0] step 1460/4000 loss=5.5153 gn=0.56 lr=4.66e-04 tps=17134 mem=6.3G lam=None Gm=0.5529378255208334
528[g_only s0] step 1480/4000 loss=5.5728 gn=0.57 lr=4.62e-04 tps=17141 mem=6.3G lam=None Gm=0.5250244140625
529[g_only s0] step 1500/4000 loss=5.4834 gn=0.60 lr=4.59e-04 tps=17149 mem=6.3G lam=None Gm=0.5370279947916666
530[g_only s0] step 1520/4000 loss=5.4469 gn=0.53 lr=4.55e-04 tps=17157 mem=6.3G lam=None Gm=0.5386555989583334
531[g_only s0] step 1540/4000 loss=5.3628 gn=0.52 lr=4.51e-04 tps=17164 mem=6.3G lam=None Gm=0.5360514322916666
532[g_only s0] step 1560/4000 loss=5.3563 gn=0.52 lr=4.47e-04 tps=17171 mem=6.3G lam=None Gm=0.5361328125
533[g_only s0] step 1580/4000 loss=5.4303 gn=0.58 lr=4.43e-04 tps=17178 mem=6.3G lam=None Gm=0.5266520182291666
534[g_only s0] step 1600/4000 loss=5.4288 gn=0.55 lr=4.38e-04 tps=17185 mem=6.3G lam=None Gm=0.5325520833333334
535[g_only s0] EVAL step 1600: val_loss=5.4864 val_ppl=241.38
536[g_only s0] step 1620/4000 loss=5.3893 gn=0.52 lr=4.34e-04 tps=17116 mem=6.3G lam=None Gm=0.54119873046875
537[g_only s0] step 1640/4000 loss=5.2717 gn=0.53 lr=4.30e-04 tps=17124 mem=6.3G lam=None Gm=0.52777099609375
538[g_only s0] step 1660/4000 loss=5.4406 gn=0.58 lr=4.26e-04 tps=17132 mem=6.3G lam=None Gm=0.5328776041666666
539[g_only s0] step 1680/4000 loss=5.5967 gn=0.55 lr=4.22e-04 tps=17140 mem=6.3G lam=None Gm=0.5220540364583334
540[g_only s0] step 1700/4000 loss=5.4539 gn=0.58 lr=4.18e-04 tps=17147 mem=6.3G lam=None Gm=0.5311482747395834
541[g_only s0] step 1720/4000 loss=5.2196 gn=0.56 lr=4.13e-04 tps=17154 mem=6.3G lam=None Gm=0.5096638997395834
542[g_only s0] step 1740/4000 loss=5.2746 gn=0.56 lr=4.09e-04 tps=17162 mem=6.3G lam=None Gm=0.5119222005208334
543[g_only s0] step 1760/4000 loss=5.3570 gn=0.54 lr=4.05e-04 tps=17169 mem=6.3G lam=None Gm=0.50897216796875
544[g_only s0] step 1780/4000 loss=5.2094 gn=0.54 lr=4.01e-04 tps=17175 mem=6.3G lam=None Gm=0.5138956705729166
545[g_only s0] step 1800/4000 loss=5.2770 gn=0.58 lr=3.96e-04 tps=17181 mem=6.3G lam=None Gm=0.5162760416666666
546[g_only s0] EVAL step 1800: val_loss=5.4124 val_ppl=224.17
547[g_only s0] step 1820/4000 loss=5.3734 gn=0.59 lr=3.92e-04 tps=17119 mem=6.3G lam=None Gm=0.5155843098958334
548[g_only s0] step 1840/4000 loss=5.2164 gn=0.59 lr=3.88e-04 tps=17126 mem=6.3G lam=None Gm=0.5003458658854166
549[g_only s0] step 1860/4000 loss=5.2742 gn=0.64 lr=3.83e-04 tps=17133 mem=6.3G lam=None Gm=0.5200398763020834
550[g_only s0] step 1880/4000 loss=5.3516 gn=0.59 lr=3.79e-04 tps=17139 mem=6.3G lam=None Gm=0.5229085286458334
551[g_only s0] step 1900/4000 loss=5.3081 gn=0.62 lr=3.74e-04 tps=17145 mem=6.3G lam=None Gm=0.5034586588541666
552[g_only s0] step 1920/4000 loss=5.2620 gn=0.57 lr=3.70e-04 tps=17151 mem=6.3G lam=None Gm=0.5080362955729166
553[g_only s0] step 1940/4000 loss=5.3117 gn=0.57 lr=3.66e-04 tps=17158 mem=6.3G lam=None Gm=0.5031331380208334
554[g_only s0] step 1960/4000 loss=5.5161 gn=0.62 lr=3.61e-04 tps=17163 mem=6.3G lam=None Gm=0.50909423828125
555[g_only s0] step 1980/4000 loss=5.4206 gn=0.57 lr=3.57e-04 tps=17170 mem=6.3G lam=None Gm=0.5130411783854166
556[g_only s0] step 2000/4000 loss=5.2534 gn=0.57 lr=3.52e-04 tps=17176 mem=6.3G lam=None Gm=0.5052286783854166
557[g_only s0] EVAL step 2000: val_loss=5.3724 val_ppl=215.37
558[g_only s0] step 2020/4000 loss=5.2267 gn=0.61 lr=3.48e-04 tps=17122 mem=6.3G lam=None Gm=0.5080159505208334
559[g_only s0] step 2040/4000 loss=5.3692 gn=0.66 lr=3.43e-04 tps=17129 mem=6.3G lam=None Gm=0.5000203450520834
560[g_only s0] step 2060/4000 loss=5.0994 gn=0.62 lr=3.39e-04 tps=17136 mem=6.3G lam=None Gm=0.5040283203125
561[g_only s0] step 2080/4000 loss=5.2446 gn=0.59 lr=3.34e-04 tps=17142 mem=6.3G lam=None Gm=0.5069580078125
562[g_only s0] step 2100/4000 loss=5.2432 gn=0.55 lr=3.30e-04 tps=17149 mem=6.3G lam=None Gm=0.49273681640625
563[g_only s0] step 2120/4000 loss=5.1525 gn=0.59 lr=3.26e-04 tps=17155 mem=6.3G lam=None Gm=0.4863688151041667
564[g_only s0] step 2140/4000 loss=5.0868 gn=0.60 lr=3.21e-04 tps=17161 mem=6.3G lam=None Gm=0.4986979166666667
565[g_only s0] step 2160/4000 loss=5.1276 gn=0.62 lr=3.17e-04 tps=17167 mem=6.3G lam=None Gm=0.4900105794270833
566[g_only s0] step 2180/4000 loss=5.1740 gn=0.65 lr=3.12e-04 tps=17173 mem=6.3G lam=None Gm=0.4993896484375
567[g_only s0] step 2200/4000 loss=5.3146 gn=0.60 lr=3.08e-04 tps=17179 mem=6.3G lam=None Gm=0.4974365234375
568[g_only s0] EVAL step 2200: val_loss=5.2782 val_ppl=196.02
569[g_only s0] step 2220/4000 loss=5.2071 gn=0.60 lr=3.03e-04 tps=17129 mem=6.3G lam=None Gm=0.4936726888020833
570[g_only s0] step 2240/4000 loss=5.1489 gn=0.57 lr=2.99e-04 tps=17134 mem=6.3G lam=None Gm=0.4821573893229167
571[g_only s0] step 2260/4000 loss=5.1579 gn=0.57 lr=2.94e-04 tps=17140 mem=6.3G lam=None Gm=0.4826456705729167
572[g_only s0] step 2280/4000 loss=5.3368 gn=0.64 lr=2.90e-04 tps=17146 mem=6.3G lam=None Gm=0.483154296875
573[g_only s0] step 2300/4000 loss=5.1231 gn=0.63 lr=2.86e-04 tps=17152 mem=6.3G lam=None Gm=0.4803059895833333
574[g_only s0] step 2320/4000 loss=5.2337 gn=0.63 lr=2.81e-04 tps=17157 mem=6.3G lam=None Gm=0.478759765625
575[g_only s0] step 2340/4000 loss=5.1670 gn=0.63 lr=2.77e-04 tps=17156 mem=6.3G lam=None Gm=0.483642578125
576[g_only s0] step 2360/4000 loss=5.2510 gn=0.61 lr=2.72e-04 tps=17150 mem=6.3G lam=None Gm=0.4729207356770833
577[g_only s0] step 2380/4000 loss=5.1356 gn=0.61 lr=2.68e-04 tps=17153 mem=6.3G lam=None Gm=0.48333740234375
578[g_only s0] step 2400/4000 loss=5.0308 gn=0.62 lr=2.64e-04 tps=17156 mem=6.3G lam=None Gm=0.4855143229166667
579[g_only s0] EVAL step 2400: val_loss=5.2118 val_ppl=183.43
580[g_only s0] step 2420/4000 loss=5.0347 gn=0.64 lr=2.59e-04 tps=17108 mem=6.3G lam=None Gm=0.47509765625
581[g_only s0] step 2440/4000 loss=5.1393 gn=0.62 lr=2.55e-04 tps=17111 mem=6.3G lam=None Gm=0.46600341796875
582[g_only s0] step 2460/4000 loss=5.2275 gn=0.62 lr=2.51e-04 tps=17115 mem=6.3G lam=None Gm=0.47979736328125
583[g_only s0] step 2480/4000 loss=5.1158 gn=0.61 lr=2.47e-04 tps=17119 mem=6.3G lam=None Gm=0.4676513671875
584[g_only s0] step 2500/4000 loss=4.9617 gn=0.61 lr=2.42e-04 tps=17123 mem=6.3G lam=None Gm=0.4730021158854167
585[g_only s0] step 2520/4000 loss=5.0374 gn=0.58 lr=2.38e-04 tps=17128 mem=6.3G lam=None Gm=0.46563720703125
586[g_only s0] step 2540/4000 loss=5.0929 gn=0.61 lr=2.34e-04 tps=17132 mem=6.3G lam=None Gm=0.4740397135416667
587[g_only s0] step 2560/4000 loss=4.8987 gn=0.66 lr=2.30e-04 tps=17136 mem=6.3G lam=None Gm=0.4639485677083333
588[g_only s0] step 2580/4000 loss=5.1882 gn=0.65 lr=2.26e-04 tps=17141 mem=6.3G lam=None Gm=0.4627888997395833
589[g_only s0] step 2600/4000 loss=4.9632 gn=0.61 lr=2.22e-04 tps=17146 mem=6.3G lam=None Gm=0.4612630208333333
590[g_only s0] EVAL step 2600: val_loss=5.1738 val_ppl=176.59
591[g_only s0] step 2620/4000 loss=5.1408 gn=0.67 lr=2.17e-04 tps=17103 mem=6.3G lam=None Gm=0.4739176432291667
592[g_only s0] step 2640/4000 loss=5.1478 gn=0.69 lr=2.13e-04 tps=17108 mem=6.3G lam=None Gm=0.4818522135416667
593[g_only s0] step 2660/4000 loss=5.0090 gn=0.63 lr=2.09e-04 tps=17113 mem=6.3G lam=None Gm=0.4649454752604167
594[g_only s0] step 2680/4000 loss=4.8474 gn=0.78 lr=2.05e-04 tps=17118 mem=6.3G lam=None Gm=0.4655558268229167
595[g_only s0] step 2700/4000 loss=5.0901 gn=0.69 lr=2.01e-04 tps=17122 mem=6.3G lam=None Gm=0.4690958658854167
596[g_only s0] step 2720/4000 loss=5.0884 gn=0.69 lr=1.98e-04 tps=17127 mem=6.3G lam=None Gm=0.4544677734375
597[g_only s0] step 2740/4000 loss=5.0685 gn=0.64 lr=1.94e-04 tps=17131 mem=6.3G lam=None Gm=0.4671630859375
598[g_only s0] step 2760/4000 loss=4.9182 gn=0.66 lr=1.90e-04 tps=17136 mem=6.3G lam=None Gm=0.4716796875
599[g_only s0] step 2780/4000 loss=5.0491 gn=0.64 lr=1.86e-04 tps=17140 mem=6.3G lam=None Gm=0.4642130533854167
600[g_only s0] step 2800/4000 loss=4.9962 gn=0.67 lr=1.82e-04 tps=17144 mem=6.3G lam=None Gm=0.4649658203125
601[g_only s0] EVAL step 2800: val_loss=5.1065 val_ppl=165.09
602[g_only s0] step 2820/4000 loss=5.0312 gn=0.65 lr=1.79e-04 tps=17104 mem=6.3G lam=None Gm=0.46160888671875
603[g_only s0] step 2840/4000 loss=4.8502 gn=0.68 lr=1.75e-04 tps=17108 mem=6.3G lam=None Gm=0.449462890625
604[g_only s0] step 2860/4000 loss=5.0163 gn=0.63 lr=1.71e-04 tps=17112 mem=6.3G lam=None Gm=0.4520263671875
605[g_only s0] step 2880/4000 loss=5.0066 gn=0.76 lr=1.68e-04 tps=17117 mem=6.3G lam=None Gm=0.4512125651041667
606[g_only s0] step 2900/4000 loss=4.8510 gn=0.65 lr=1.64e-04 tps=17121 mem=6.3G lam=None Gm=0.4589640299479167
607[g_only s0] step 2920/4000 loss=5.0601 gn=0.66 lr=1.61e-04 tps=17125 mem=6.3G lam=None Gm=0.4770711263020833
608[g_only s0] step 2940/4000 loss=4.9675 gn=0.62 lr=1.57e-04 tps=17129 mem=6.3G lam=None Gm=0.4576619466145833
609[g_only s0] step 2960/4000 loss=4.9508 gn=0.66 lr=1.54e-04 tps=17133 mem=6.3G lam=None Gm=0.4636637369791667
610[g_only s0] step 2980/4000 loss=4.9112 gn=0.64 lr=1.50e-04 tps=17137 mem=6.3G lam=None Gm=0.4538981119791667
611[g_only s0] step 3000/4000 loss=4.8637 gn=0.68 lr=1.47e-04 tps=17141 mem=6.3G lam=None Gm=0.4453328450520833
612[g_only s0] EVAL step 3000: val_loss=5.1212 val_ppl=167.53
613[g_only s0] step 3020/4000 loss=4.9321 gn=0.62 lr=1.44e-04 tps=17104 mem=6.3G lam=None Gm=0.4552815755208333
614[g_only s0] step 3040/4000 loss=5.0088 gn=0.71 lr=1.41e-04 tps=17109 mem=6.3G lam=None Gm=0.4568888346354167
615[g_only s0] step 3060/4000 loss=4.9715 gn=0.69 lr=1.38e-04 tps=17113 mem=6.3G lam=None Gm=0.4580485026041667
616[g_only s0] step 3080/4000 loss=4.8162 gn=0.64 lr=1.34e-04 tps=17118 mem=6.3G lam=None Gm=0.4505411783854167
617[g_only s0] step 3100/4000 loss=5.0214 gn=0.69 lr=1.31e-04 tps=17121 mem=6.3G lam=None Gm=0.4586181640625
618[g_only s0] step 3120/4000 loss=5.0558 gn=0.71 lr=1.28e-04 tps=17125 mem=6.3G lam=None Gm=0.4617106119791667
619[g_only s0] step 3140/4000 loss=4.9006 gn=0.65 lr=1.25e-04 tps=17130 mem=6.3G lam=None Gm=0.456298828125
620[g_only s0] step 3160/4000 loss=5.0574 gn=0.80 lr=1.23e-04 tps=17134 mem=6.3G lam=None Gm=0.462158203125
621[g_only s0] step 3180/4000 loss=4.8870 gn=0.65 lr=1.20e-04 tps=17138 mem=6.3G lam=None Gm=0.4506429036458333
622[g_only s0] step 3200/4000 loss=4.9075 gn=0.69 lr=1.17e-04 tps=17142 mem=6.3G lam=None Gm=0.45135498046875
623[g_only s0] EVAL step 3200: val_loss=5.0290 val_ppl=152.78
624[g_only s0] step 3220/4000 loss=5.0219 gn=0.74 lr=1.14e-04 tps=17108 mem=6.3G lam=None Gm=0.4473470052083333
625[g_only s0] step 3240/4000 loss=5.0297 gn=0.67 lr=1.12e-04 tps=17112 mem=6.3G lam=None Gm=0.4608561197916667
626[g_only s0] step 3260/4000 loss=4.7921 gn=0.67 lr=1.09e-04 tps=17116 mem=6.3G lam=None Gm=0.45697021484375
627[g_only s0] step 3280/4000 loss=5.0383 gn=0.70 lr=1.06e-04 tps=17120 mem=6.3G lam=None Gm=0.4521077473958333
628[g_only s0] step 3300/4000 loss=4.8601 gn=0.68 lr=1.04e-04 tps=17124 mem=6.3G lam=None Gm=0.4398193359375
629[g_only s0] step 3320/4000 loss=5.0289 gn=0.71 lr=1.02e-04 tps=17128 mem=6.3G lam=None Gm=0.4434407552083333
630[g_only s0] step 3340/4000 loss=4.8558 gn=0.67 lr=9.92e-05 tps=17131 mem=6.3G lam=None Gm=0.4430745442708333
631[g_only s0] step 3360/4000 loss=5.0094 gn=0.72 lr=9.69e-05 tps=17135 mem=6.3G lam=None Gm=0.445068359375
632[g_only s0] step 3380/4000 loss=4.9999 gn=0.69 lr=9.47e-05 tps=17139 mem=6.3G lam=None Gm=0.4417724609375
633[g_only s0] step 3400/4000 loss=4.8891 gn=0.68 lr=9.25e-05 tps=17142 mem=6.3G lam=None Gm=0.4418131510416667
634[g_only s0] EVAL step 3400: val_loss=5.0328 val_ppl=153.36
635[g_only s0] step 3420/4000 loss=4.9196 gn=0.69 lr=9.04e-05 tps=17110 mem=6.3G lam=None Gm=0.4532674153645833
636[g_only s0] step 3440/4000 loss=4.9692 gn=0.73 lr=8.84e-05 tps=17114 mem=6.3G lam=None Gm=0.4556070963541667
637[g_only s0] step 3460/4000 loss=4.8335 gn=0.70 lr=8.65e-05 tps=17118 mem=6.3G lam=None Gm=0.4396565755208333
638[g_only s0] step 3480/4000 loss=4.9484 gn=0.69 lr=8.46e-05 tps=17122 mem=6.3G lam=None Gm=0.4453125
639[g_only s0] step 3500/4000 loss=4.8527 gn=0.68 lr=8.27e-05 tps=17126 mem=6.3G lam=None Gm=0.4424031575520833
640[g_only s0] step 3520/4000 loss=4.7420 gn=0.74 lr=8.10e-05 tps=17130 mem=6.3G lam=None Gm=0.4512532552083333
641[g_only s0] step 3540/4000 loss=4.7381 gn=0.70 lr=7.93e-05 tps=17133 mem=6.3G lam=None Gm=0.4396565755208333
642[g_only s0] step 3560/4000 loss=4.9895 gn=0.68 lr=7.77e-05 tps=17137 mem=6.3G lam=None Gm=0.4422607421875
643[g_only s0] step 3580/4000 loss=4.8193 gn=0.66 lr=7.61e-05 tps=17141 mem=6.3G lam=None Gm=0.43359375
644[g_only s0] step 3600/4000 loss=4.8507 gn=0.76 lr=7.46e-05 tps=17145 mem=6.3G lam=None Gm=0.4462483723958333
645[g_only s0] EVAL step 3600: val_loss=4.9974 val_ppl=148.03
646[g_only s0] step 3620/4000 loss=4.9051 gn=0.72 lr=7.32e-05 tps=17115 mem=6.3G lam=None Gm=0.435791015625
647[g_only s0] step 3640/4000 loss=4.9773 gn=0.70 lr=7.19e-05 tps=17119 mem=6.3G lam=None Gm=0.4398600260416667
648[g_only s0] step 3660/4000 loss=5.0183 gn=0.80 lr=7.06e-05 tps=17123 mem=6.3G lam=None Gm=0.4494222005208333
649[g_only s0] step 3680/4000 loss=4.9392 gn=0.69 lr=6.94e-05 tps=17127 mem=6.3G lam=None Gm=0.435791015625
650[g_only s0] step 3700/4000 loss=4.9855 gn=0.68 lr=6.83e-05 tps=17130 mem=6.3G lam=None Gm=0.4390665690104167
651[g_only s0] step 3720/4000 loss=4.7607 gn=0.68 lr=6.72e-05 tps=17134 mem=6.3G lam=None Gm=0.43572998046875
652[g_only s0] step 3740/4000 loss=4.9413 gn=0.75 lr=6.62e-05 tps=17138 mem=6.3G lam=None Gm=0.43365478515625
653[g_only s0] step 3760/4000 loss=4.8730 gn=0.70 lr=6.53e-05 tps=17141 mem=6.3G lam=None Gm=0.4378662109375
654[g_only s0] step 3780/4000 loss=4.7835 gn=0.71 lr=6.45e-05 tps=17144 mem=6.3G lam=None Gm=0.4366455078125
655[g_only s0] step 3800/4000 loss=4.9004 gn=0.73 lr=6.37e-05 tps=17148 mem=6.3G lam=None Gm=0.4463297526041667
656[g_only s0] EVAL step 3800: val_loss=4.9756 val_ppl=144.84
657[g_only s0] step 3820/4000 loss=4.9450 gn=0.68 lr=6.30e-05 tps=17119 mem=6.3G lam=None Gm=0.4392903645833333
658[g_only s0] step 3840/4000 loss=4.8772 gn=0.69 lr=6.24e-05 tps=17123 mem=6.3G lam=None Gm=0.4384765625
659[g_only s0] step 3860/4000 loss=4.8904 gn=0.73 lr=6.18e-05 tps=17126 mem=6.3G lam=None Gm=0.4364217122395833
660[g_only s0] step 3880/4000 loss=4.9577 gn=0.74 lr=6.13e-05 tps=17129 mem=6.3G lam=None Gm=0.445556640625
661[g_only s0] step 3900/4000 loss=4.8484 gn=0.71 lr=6.09e-05 tps=17133 mem=6.3G lam=None Gm=0.4344685872395833
662[g_only s0] step 3920/4000 loss=4.7851 gn=0.75 lr=6.06e-05 tps=17136 mem=6.3G lam=None Gm=0.433837890625
663[g_only s0] step 3940/4000 loss=4.9613 gn=0.74 lr=6.03e-05 tps=17140 mem=6.3G lam=None Gm=0.4343668619791667
664[g_only s0] step 3960/4000 loss=4.9078 gn=0.72 lr=6.01e-05 tps=17143 mem=6.3G lam=None Gm=0.4307047526041667
665[g_only s0] step 3980/4000 loss=4.8536 gn=0.76 lr=6.00e-05 tps=17147 mem=6.3G lam=None Gm=0.4358113606770833
666[g_only s0] EVAL step 3999: val_loss=4.9765 val_ppl=144.97
667[init] variant=full seed=0 params=63.98M
668[init] use_S=True use_G=True
669[full s0] step 0/4000 loss=10.9002 gn=11.22 lr=3.00e-06 tps=7177 mem=5.9G lam=0.12692801654338837 Gm=0.5060585426787535
670[full s0] step 20/4000 loss=9.8395 gn=2.04 lr=6.30e-05 tps=14665 mem=6.4G lam=0.12691388465464115 Gm=0.46947087223331135
671[full s0] step 40/4000 loss=9.1119 gn=1.75 lr=1.23e-04 tps=15000 mem=6.4G lam=0.12681447404126325 Gm=0.4522867053747177
672[full s0] step 60/4000 loss=8.1301 gn=1.59 lr=1.83e-04 tps=15154 mem=6.4G lam=0.12670463137328625 Gm=0.44430506974458694
673[full s0] step 80/4000 loss=7.4261 gn=0.77 lr=2.43e-04 tps=15223 mem=6.4G lam=0.12660598816970983 Gm=0.4503638831277688
674[full s0] step 100/4000 loss=7.2137 gn=0.97 lr=3.03e-04 tps=15267 mem=6.4G lam=0.126494025811553 Gm=0.4433469759921233
675[full s0] step 120/4000 loss=7.0544 gn=0.63 lr=3.63e-04 tps=15301 mem=6.4G lam=0.12652451234559217 Gm=0.4225224206844966
676[full s0] step 140/4000 loss=6.8386 gn=0.58 lr=4.23e-04 tps=15322 mem=6.4G lam=0.12646274889508882 Gm=0.4161691988507907
677[full s0] step 160/4000 loss=6.9098 gn=0.79 lr=4.83e-04 tps=15327 mem=6.4G lam=0.12636082619428635 Gm=0.41686703140536946
678[full s0] step 180/4000 loss=6.7997 gn=0.61 lr=5.43e-04 tps=15329 mem=6.4G lam=0.12636513697604337 Gm=0.40593364586432773
679[full s0] step 200/4000 loss=6.6765 gn=0.48 lr=6.00e-04 tps=15341 mem=6.4G lam=0.12632009697457156 Gm=0.4146435211102168
680[full s0] EVAL step 200: val_loss=6.7406 val_ppl=846.05
681[full s0] step 220/4000 loss=6.5711 gn=0.49 lr=6.00e-04 tps=14849 mem=6.4G lam=0.12637140632917485 Gm=0.4111607000231743
682[full s0] step 240/4000 loss=6.6276 gn=0.55 lr=6.00e-04 tps=14896 mem=6.4G lam=0.12644176309307417 Gm=0.4052786926428477
683[full s0] step 260/4000 loss=6.5910 gn=0.50 lr=6.00e-04 tps=14935 mem=6.4G lam=0.12642955655852953 Gm=0.40804433325926465
684[full s0] step 280/4000 loss=6.5056 gn=0.49 lr=5.99e-04 tps=14969 mem=6.4G lam=0.12644678788880506 Gm=0.4035586975514889
685[full s0] step 300/4000 loss=6.3354 gn=0.47 lr=5.99e-04 tps=14999 mem=6.4G lam=0.1264594110349814 Gm=0.41425324728091556
686[full s0] step 320/4000 loss=6.4140 gn=0.53 lr=5.99e-04 tps=15024 mem=6.4G lam=0.12646780970195928 Gm=0.4137684901555379
687[full s0] step 340/4000 loss=6.3941 gn=0.53 lr=5.98e-04 tps=15047 mem=6.4G lam=0.12651423954715332 Gm=0.41517286996046704
688[full s0] step 360/4000 loss=6.2520 gn=0.48 lr=5.98e-04 tps=15071 mem=6.4G lam=0.1266111023724079 Gm=0.40464745213588077
689[full s0] step 380/4000 loss=6.2105 gn=0.51 lr=5.97e-04 tps=15091 mem=6.4G lam=0.12667184695601463 Gm=0.41207972665627796
690[full s0] step 400/4000 loss=6.2497 gn=0.48 lr=5.96e-04 tps=15109 mem=6.4G lam=0.12674504642685255 Gm=0.41418295974532765
691[full s0] EVAL step 400: val_loss=6.3913 val_ppl=596.65
692[full s0] step 420/4000 loss=6.3861 gn=0.66 lr=5.96e-04 tps=14860 mem=6.4G lam=0.1268323635061582 Gm=0.43083703393737477
693[full s0] step 440/4000 loss=6.3404 gn=0.53 lr=5.95e-04 tps=14884 mem=6.4G lam=0.1269418727606535 Gm=0.4135419428348541
694[full s0] step 460/4000 loss=6.2730 gn=0.50 lr=5.94e-04 tps=14907 mem=6.4G lam=0.12701714721818766 Gm=0.42971306045850116
695[full s0] step 480/4000 loss=6.3662 gn=0.48 lr=5.93e-04 tps=14927 mem=6.4G lam=0.12716670148074627 Gm=0.43850669637322426
696[full s0] step 500/4000 loss=6.2924 gn=0.50 lr=5.92e-04 tps=14946 mem=6.4G lam=0.12731456570327282 Gm=0.4456465107699235
697[full s0] step 520/4000 loss=6.1516 gn=0.52 lr=5.91e-04 tps=14963 mem=6.4G lam=0.12751984472076097 Gm=0.4579197404285272
698[full s0] step 540/4000 loss=6.1714 gn=0.48 lr=5.89e-04 tps=14981 mem=6.4G lam=0.12762735784053802 Gm=0.46552255749702454
699[full s0] step 560/4000 loss=6.0217 gn=0.51 lr=5.88e-04 tps=14998 mem=6.4G lam=0.127838846296072 Gm=0.4710191413760185
700[full s0] step 580/4000 loss=6.1941 gn=0.53 lr=5.87e-04 tps=15012 mem=6.4G lam=0.1280172864596049 Gm=0.4879833509524663
701[full s0] step 600/4000 loss=5.9497 gn=0.48 lr=5.85e-04 tps=15023 mem=6.4G lam=0.1282313127691547 Gm=0.5293142658968767
702[full s0] EVAL step 600: val_loss=6.1594 val_ppl=473.16
703[full s0] step 620/4000 loss=6.0152 gn=0.47 lr=5.84e-04 tps=14857 mem=6.4G lam=0.1284641269594431 Gm=0.5108194835484028
704[full s0] step 640/4000 loss=6.0979 gn=0.60 lr=5.82e-04 tps=14872 mem=6.4G lam=0.12870528859396776 Gm=0.5306381694972515
705[full s0] step 660/4000 loss=5.9328 gn=0.52 lr=5.81e-04 tps=14885 mem=6.4G lam=0.12897684735556444 Gm=0.569448359310627
706[full s0] step 680/4000 loss=5.9431 gn=0.50 lr=5.79e-04 tps=14900 mem=6.4G lam=0.12917988343785206 Gm=0.5352130594352881
707[full s0] step 700/4000 loss=6.0199 gn=0.45 lr=5.77e-04 tps=14917 mem=6.4G lam=0.12941528173784414 Gm=0.5478734436134497
708[full s0] step 720/4000 loss=6.0052 gn=0.50 lr=5.75e-04 tps=14928 mem=6.4G lam=0.12969180072347322 Gm=0.5475347439448038
709[full s0] step 740/4000 loss=6.0267 gn=0.49 lr=5.74e-04 tps=14940 mem=6.4G lam=0.13002433255314827 Gm=0.5399082625905672
710[full s0] step 760/4000 loss=5.8964 gn=0.54 lr=5.72e-04 tps=14954 mem=6.4G lam=0.13041084135572115 Gm=0.5374414511024952
711[full s0] step 780/4000 loss=5.9099 gn=0.47 lr=5.70e-04 tps=14966 mem=6.4G lam=0.13078376495589814 Gm=0.5692521668970585
712[full s0] step 800/4000 loss=5.8641 gn=0.48 lr=5.67e-04 tps=14977 mem=6.4G lam=0.13108982487271229 Gm=0.5646844009558359
713[full s0] EVAL step 800: val_loss=5.9699 val_ppl=391.45
714[full s0] step 820/4000 loss=5.9465 gn=0.50 lr=5.65e-04 tps=14853 mem=6.4G lam=0.13149288886537155 Gm=0.5422952758769194
715[full s0] step 840/4000 loss=6.1234 gn=0.49 lr=5.63e-04 tps=14864 mem=6.4G lam=0.13181905820965767 Gm=0.5669704626003901
716[full s0] step 860/4000 loss=5.9374 gn=0.54 lr=5.61e-04 tps=14873 mem=6.4G lam=0.1321754902601242 Gm=0.5561091440419356
717[full s0] step 880/4000 loss=5.9279 gn=0.59 lr=5.58e-04 tps=14883 mem=6.4G lam=0.13254042839010557 Gm=0.563983965665102
718[full s0] step 900/4000 loss=5.8279 gn=0.48 lr=5.56e-04 tps=14895 mem=6.4G lam=0.13288607758780321 Gm=0.565286461263895
719[full s0] step 920/4000 loss=5.7348 gn=0.49 lr=5.54e-04 tps=14906 mem=6.4G lam=0.13325560869028172 Gm=0.5610882913072904
720[full s0] step 940/4000 loss=5.7798 gn=0.52 lr=5.51e-04 tps=14916 mem=6.4G lam=0.13368192439277968 Gm=0.5797739451130232
721[full s0] step 960/4000 loss=5.7783 gn=0.58 lr=5.48e-04 tps=14926 mem=6.4G lam=0.1340254961202542 Gm=0.5871407128870487
722[full s0] step 980/4000 loss=5.6761 gn=0.49 lr=5.46e-04 tps=14935 mem=6.4G lam=0.13437930028885603 Gm=0.563348725438118
723[full s0] step 1000/4000 loss=5.7929 gn=0.52 lr=5.43e-04 tps=14946 mem=6.4G lam=0.13478029984980822 Gm=0.5567546946307024
724[full s0] EVAL step 1000: val_loss=5.8160 val_ppl=335.64
725[full s0] step 1020/4000 loss=5.5333 gn=0.50 lr=5.40e-04 tps=14848 mem=6.4G lam=0.13519225362688303 Gm=0.5625596257547537
726[full s0] step 1040/4000 loss=5.7766 gn=0.52 lr=5.37e-04 tps=14856 mem=6.4G lam=0.13549139443784952 Gm=0.57581627368927
727[full s0] step 1060/4000 loss=5.4969 gn=0.51 lr=5.35e-04 tps=14866 mem=6.4G lam=0.13587642088532448 Gm=0.532783313964804
728[full s0] step 1080/4000 loss=5.6966 gn=0.53 lr=5.32e-04 tps=14874 mem=6.4G lam=0.13626651465892792 Gm=0.5369457434862852
729[full s0] step 1100/4000 loss=5.6652 gn=0.60 lr=5.29e-04 tps=14883 mem=6.4G lam=0.13660530528674522 Gm=0.5775514940420786
730[full s0] step 1120/4000 loss=5.5989 gn=0.53 lr=5.26e-04 tps=14892 mem=6.4G lam=0.13700281300892433 Gm=0.5345892341186603
731[full s0] step 1140/4000 loss=5.8584 gn=0.53 lr=5.22e-04 tps=14899 mem=6.4G lam=0.1373259505877892 Gm=0.5394855334113041
732[full s0] step 1160/4000 loss=5.5913 gn=0.49 lr=5.19e-04 tps=14907 mem=6.4G lam=0.1377501521880428 Gm=0.54696986079216
733[full s0] step 1180/4000 loss=5.6599 gn=0.51 lr=5.16e-04 tps=14916 mem=6.4G lam=0.13811772099385658 Gm=0.5348725412040949
734[full s0] step 1200/4000 loss=5.5536 gn=0.50 lr=5.13e-04 tps=14924 mem=6.4G lam=0.13851364267369112 Gm=0.5457661300897598
735[full s0] EVAL step 1200: val_loss=5.6667 val_ppl=289.07
736[full s0] step 1220/4000 loss=5.5612 gn=0.54 lr=5.10e-04 tps=14842 mem=6.4G lam=0.13890980711827675 Gm=0.5422924067825079
737[full s0] step 1240/4000 loss=5.4509 gn=0.52 lr=5.06e-04 tps=14851 mem=6.4G lam=0.13927926247318587 Gm=0.5455869063735008
738[full s0] step 1260/4000 loss=5.4327 gn=0.49 lr=5.03e-04 tps=14858 mem=6.4G lam=0.13965522777289152 Gm=0.5358441267162561
739[full s0] step 1280/4000 loss=5.5771 gn=0.57 lr=4.99e-04 tps=14866 mem=6.4G lam=0.140042661378781 Gm=0.5416806563735008
740[full s0] step 1300/4000 loss=5.6711 gn=0.57 lr=4.96e-04 tps=14874 mem=6.4G lam=0.14043778398384651 Gm=0.539463714386026
741[full s0] step 1320/4000 loss=5.6225 gn=0.56 lr=4.92e-04 tps=14881 mem=6.4G lam=0.14081103230516115 Gm=0.534868306790789
742[full s0] step 1340/4000 loss=5.7384 gn=0.55 lr=4.89e-04 tps=14889 mem=6.4G lam=0.1412075301632285 Gm=0.537356911227107
743[full s0] step 1360/4000 loss=5.5310 gn=0.57 lr=4.85e-04 tps=14897 mem=6.4G lam=0.14158294132600227 Gm=0.5233028481403986
744[full s0] step 1380/4000 loss=5.4097 gn=0.53 lr=4.81e-04 tps=14904 mem=6.4G lam=0.14195760743071636 Gm=0.5073543048153321
745[full s0] step 1400/4000 loss=5.4238 gn=0.57 lr=4.78e-04 tps=14910 mem=6.4G lam=0.14229492781062922 Gm=0.524269043157498
746[full s0] EVAL step 1400: val_loss=5.5600 val_ppl=259.81
747[full s0] step 1420/4000 loss=5.3996 gn=0.56 lr=4.74e-04 tps=14840 mem=6.4G lam=0.14264872639129558 Gm=0.5051169519623121
748[full s0] step 1440/4000 loss=5.3948 gn=0.57 lr=4.70e-04 tps=14847 mem=6.4G lam=0.14301192946732044 Gm=0.5213417652994394
749[full s0] step 1460/4000 loss=5.5144 gn=0.59 lr=4.66e-04 tps=14853 mem=6.4G lam=0.1434003453080853 Gm=0.5063190547128519
750[full s0] step 1480/4000 loss=5.5660 gn=0.56 lr=4.62e-04 tps=14860 mem=6.4G lam=0.1437656677638491 Gm=0.5025668038676182
751[full s0] step 1500/4000 loss=5.4600 gn=0.57 lr=4.59e-04 tps=14866 mem=6.4G lam=0.14413854635010162 Gm=0.4969680073360602
752[full s0] step 1520/4000 loss=5.4459 gn=0.54 lr=4.55e-04 tps=14873 mem=6.4G lam=0.14453936771800122 Gm=0.5143021817008654
753[full s0] step 1540/4000 loss=5.3414 gn=0.51 lr=4.51e-04 tps=14880 mem=6.4G lam=0.1449366994202137 Gm=0.5127804862956206
754[full s0] step 1560/4000 loss=5.3644 gn=0.56 lr=4.47e-04 tps=14886 mem=6.4G lam=0.14528835751116276 Gm=0.4950882631043593
755[full s0] step 1580/4000 loss=5.4323 gn=0.62 lr=4.43e-04 tps=14892 mem=6.4G lam=0.14564090967178345 Gm=0.49802837893366814
756[full s0] step 1600/4000 loss=5.4197 gn=0.55 lr=4.38e-04 tps=14898 mem=6.4G lam=0.14599672239273787 Gm=0.5047087588657936
757[full s0] EVAL step 1600: val_loss=5.4789 val_ppl=239.59
758[full s0] step 1620/4000 loss=5.3902 gn=0.54 lr=4.34e-04 tps=14836 mem=6.4G lam=0.14638349320739508 Gm=0.5047999446590742
759[full s0] step 1640/4000 loss=5.2625 gn=0.53 lr=4.30e-04 tps=14843 mem=6.4G lam=0.14677999913692474 Gm=0.5010045052816471
760[full s0] step 1660/4000 loss=5.4421 gn=0.54 lr=4.26e-04 tps=14850 mem=6.4G lam=0.14713385639091334 Gm=0.501681626153489
761[full s0] step 1680/4000 loss=5.5906 gn=0.65 lr=4.22e-04 tps=14856 mem=6.4G lam=0.1474885701512297 Gm=0.5060627665370703
762[full s0] step 1700/4000 loss=5.4464 gn=0.54 lr=4.18e-04 tps=14862 mem=6.4G lam=0.14782526809722185 Gm=0.4984721277529995
763[full s0] step 1720/4000 loss=5.2113 gn=0.56 lr=4.13e-04 tps=14867 mem=6.4G lam=0.14819134771823883 Gm=0.4783928568164508
764[full s0] step 1740/4000 loss=5.2665 gn=0.59 lr=4.09e-04 tps=14873 mem=6.4G lam=0.1485797017812729 Gm=0.4868131143351396
765[full s0] step 1760/4000 loss=5.3345 gn=0.53 lr=4.05e-04 tps=14878 mem=6.4G lam=0.14889246287445226 Gm=0.49005475516120595
766[full s0] step 1780/4000 loss=5.2198 gn=0.64 lr=4.01e-04 tps=14883 mem=6.4G lam=0.14913748173664013 Gm=0.484699172899127
767[full s0] step 1800/4000 loss=5.2915 gn=0.62 lr=3.96e-04 tps=14888 mem=6.4G lam=0.14946761603156725 Gm=0.48724480501065653
768[full s0] EVAL step 1800: val_loss=5.4097 val_ppl=223.56
769[full s0] step 1820/4000 loss=5.3663 gn=0.58 lr=3.92e-04 tps=14834 mem=6.4G lam=0.14975205467393002 Gm=0.48330537664393586
770[full s0] step 1840/4000 loss=5.2046 gn=0.57 lr=3.88e-04 tps=14840 mem=6.4G lam=0.15008515336861214 Gm=0.47825769148766994
771[full s0] step 1860/4000 loss=5.2649 gn=0.60 lr=3.83e-04 tps=14846 mem=6.4G lam=0.15041279854873815 Gm=0.4925144451359908
772[full s0] step 1880/4000 loss=5.3589 gn=0.74 lr=3.79e-04 tps=14851 mem=6.4G lam=0.15075742484380802 Gm=0.5020403222491344
773[full s0] step 1900/4000 loss=5.3196 gn=0.62 lr=3.74e-04 tps=14858 mem=6.4G lam=0.15108068535725275 Gm=0.474533099681139
774[full s0] step 1920/4000 loss=5.2498 gn=0.61 lr=3.70e-04 tps=14862 mem=6.4G lam=0.15135481270651022 Gm=0.4734061478326718
775[full s0] step 1940/4000 loss=5.3131 gn=0.58 lr=3.66e-04 tps=14866 mem=6.4G lam=0.1516401826714476 Gm=0.4780598379050692
776[full s0] step 1960/4000 loss=5.5175 gn=0.66 lr=3.61e-04 tps=14871 mem=6.4G lam=0.151948948080341 Gm=0.478161768677334
777[full s0] step 1980/4000 loss=5.4044 gn=0.58 lr=3.57e-04 tps=14875 mem=6.4G lam=0.1522312369197607 Gm=0.4881879823903243
778[full s0] step 2000/4000 loss=5.2543 gn=0.58 lr=3.52e-04 tps=14880 mem=6.4G lam=0.15254930996646485 Gm=0.4732829745238026
779[full s0] EVAL step 2000: val_loss=5.3565 val_ppl=211.97
780[full s0] step 2020/4000 loss=5.2164 gn=0.60 lr=3.48e-04 tps=14831 mem=6.4G lam=0.1528791462381681 Gm=0.4744356938948234
781[full s0] step 2040/4000 loss=5.3486 gn=0.64 lr=3.43e-04 tps=14836 mem=6.4G lam=0.15319231462975344 Gm=0.4752217202136914
782[full s0] step 2060/4000 loss=5.0926 gn=0.66 lr=3.39e-04 tps=14841 mem=6.4G lam=0.15346764431645474 Gm=0.4769606015955408
783[full s0] step 2080/4000 loss=5.2508 gn=0.62 lr=3.34e-04 tps=14847 mem=6.4G lam=0.15376037017752728 Gm=0.47535418905317783
784[full s0] step 2100/4000 loss=5.2524 gn=0.59 lr=3.30e-04 tps=14851 mem=6.4G lam=0.1540603842586279 Gm=0.4680200172588229
785[full s0] step 2120/4000 loss=5.1454 gn=0.60 lr=3.26e-04 tps=14856 mem=6.4G lam=0.15434814927478632 Gm=0.46248059688756865
786[full s0] step 2140/4000 loss=5.0802 gn=0.57 lr=3.21e-04 tps=14860 mem=6.4G lam=0.15463932634641728 Gm=0.47031607913474244
787[full s0] step 2160/4000 loss=5.1326 gn=0.64 lr=3.17e-04 tps=14865 mem=6.4G lam=0.15491167145470777 Gm=0.46407177827010554
788[full s0] step 2180/4000 loss=5.1673 gn=0.65 lr=3.12e-04 tps=14870 mem=6.4G lam=0.15516483535369238 Gm=0.4632302640626828
789[full s0] step 2200/4000 loss=5.3162 gn=0.62 lr=3.08e-04 tps=14875 mem=6.4G lam=0.15542515770842633 Gm=0.47018811851739883
790[full s0] EVAL step 2200: val_loss=5.2720 val_ppl=194.80
791[full s0] step 2220/4000 loss=5.1938 gn=0.60 lr=3.03e-04 tps=14830 mem=6.4G lam=0.15566586796194315 Gm=0.4673734738801916
792[full s0] step 2240/4000 loss=5.1447 gn=0.58 lr=2.99e-04 tps=14835 mem=6.4G lam=0.1559303660566608 Gm=0.46014175936579704
793[full s0] step 2260/4000 loss=5.1438 gn=0.58 lr=2.94e-04 tps=14840 mem=6.4G lam=0.15621465692917505 Gm=0.4625430420661966
794[full s0] step 2280/4000 loss=5.3504 gn=0.63 lr=2.90e-04 tps=14844 mem=6.4G lam=0.15644852196176848 Gm=0.45796986234684783
795[full s0] step 2300/4000 loss=5.1145 gn=0.60 lr=2.86e-04 tps=14849 mem=6.4G lam=0.15674412685136 Gm=0.4564528313154976
796[full s0] step 2320/4000 loss=5.2212 gn=0.62 lr=2.81e-04 tps=14853 mem=6.4G lam=0.1569712401057283 Gm=0.45699896446118754
797[full s0] step 2340/4000 loss=5.1594 gn=0.66 lr=2.77e-04 tps=14857 mem=6.4G lam=0.15721193235367537 Gm=0.4585708274195592
798[full s0] step 2360/4000 loss=5.2329 gn=0.59 lr=2.72e-04 tps=14862 mem=6.4G lam=0.1574559115494291 Gm=0.4535028813406825
799[full s0] step 2380/4000 loss=5.1232 gn=0.66 lr=2.68e-04 tps=14865 mem=6.4G lam=0.1576677868142724 Gm=0.4569100613395373
800[full s0] step 2400/4000 loss=5.0122 gn=0.69 lr=2.64e-04 tps=14868 mem=6.4G lam=0.15789554237077633 Gm=0.46349408601721126
801[full s0] EVAL step 2400: val_loss=5.1964 val_ppl=180.62
802[full s0] step 2420/4000 loss=5.0019 gn=0.65 lr=2.59e-04 tps=14826 mem=6.4G lam=0.15812122200926146 Gm=0.4450850347056985
803[full s0] step 2440/4000 loss=5.1299 gn=0.64 lr=2.55e-04 tps=14831 mem=6.4G lam=0.15837650404622158 Gm=0.44899142626672983
804[full s0] step 2460/4000 loss=5.2215 gn=0.64 lr=2.51e-04 tps=14835 mem=6.4G lam=0.15859906499584517 Gm=0.4574682153761387
805[full s0] step 2480/4000 loss=5.1100 gn=0.63 lr=2.47e-04 tps=14839 mem=6.4G lam=0.15882964866856733 Gm=0.447203541174531
806[full s0] step 2500/4000 loss=4.9479 gn=0.66 lr=2.42e-04 tps=14843 mem=6.4G lam=0.15902110282331705 Gm=0.4507615032295386
807[full s0] step 2520/4000 loss=5.0244 gn=0.60 lr=2.38e-04 tps=14848 mem=6.4G lam=0.15924381371587515 Gm=0.4480586371695002
808[full s0] step 2540/4000 loss=5.0845 gn=0.64 lr=2.34e-04 tps=14852 mem=6.4G lam=0.15946097454677025 Gm=0.4542156746610999
809[full s0] step 2560/4000 loss=4.8791 gn=0.63 lr=2.30e-04 tps=14856 mem=6.4G lam=0.15967680917431912 Gm=0.4438647028679649
810[full s0] step 2580/4000 loss=5.1637 gn=0.69 lr=2.26e-04 tps=14860 mem=6.4G lam=0.15990279925366244 Gm=0.4438582655663292
811[full s0] step 2600/4000 loss=4.9393 gn=0.66 lr=2.22e-04 tps=14863 mem=6.4G lam=0.1600928260013461 Gm=0.4401321733991305
812[full s0] EVAL step 2600: val_loss=5.1592 val_ppl=174.03
813[full s0] step 2620/4000 loss=5.1345 gn=0.70 lr=2.17e-04 tps=14825 mem=6.4G lam=0.16027720520893732 Gm=0.44931645939747494
814[full s0] step 2640/4000 loss=5.1138 gn=0.69 lr=2.13e-04 tps=14829 mem=6.4G lam=0.16046080024292073 Gm=0.4500795149554809
815[full s0] step 2660/4000 loss=4.9892 gn=0.66 lr=2.09e-04 tps=14833 mem=6.4G lam=0.1606252252434691 Gm=0.44443822465837
816[full s0] step 2680/4000 loss=4.8321 gn=0.81 lr=2.05e-04 tps=14837 mem=6.4G lam=0.1608069073408842 Gm=0.4425751141582926
817[full s0] step 2700/4000 loss=5.0882 gn=0.68 lr=2.01e-04 tps=14840 mem=6.4G lam=0.16100376565009356 Gm=0.4432884758959214
818[full s0] step 2720/4000 loss=5.0655 gn=0.69 lr=1.98e-04 tps=14844 mem=6.4G lam=0.16119438720246157 Gm=0.4416095061848561
819[full s0] step 2740/4000 loss=5.0501 gn=0.65 lr=1.94e-04 tps=14848 mem=6.4G lam=0.16136763679484525 Gm=0.44492132123559713
820[full s0] step 2760/4000 loss=4.8957 gn=0.66 lr=1.90e-04 tps=14851 mem=6.4G lam=0.1615272182971239 Gm=0.44701481858889264
821[full s0] step 2780/4000 loss=5.0218 gn=0.66 lr=1.86e-04 tps=14855 mem=6.4G lam=0.16169719553242126 Gm=0.4440632878492276
822[full s0] step 2800/4000 loss=4.9772 gn=0.75 lr=1.82e-04 tps=14858 mem=6.4G lam=0.1618487909436226 Gm=0.4365478853384654
823[full s0] EVAL step 2800: val_loss=5.0906 val_ppl=162.49
824[full s0] step 2820/4000 loss=5.0181 gn=0.67 lr=1.79e-04 tps=14824 mem=6.4G lam=0.16200900978098312 Gm=0.4376520048826933
825[full s0] step 2840/4000 loss=4.8275 gn=0.67 lr=1.75e-04 tps=14827 mem=6.4G lam=0.16216511682917675 Gm=0.43590452677259844
826[full s0] step 2860/4000 loss=4.9932 gn=0.63 lr=1.71e-04 tps=14831 mem=6.4G lam=0.16232836867372194 Gm=0.4350059659530719
827[full s0] step 2880/4000 loss=4.9926 gn=0.67 lr=1.68e-04 tps=14834 mem=6.4G lam=0.16248036300142607 Gm=0.4299806446457903
828[full s0] step 2900/4000 loss=4.8354 gn=0.72 lr=1.64e-04 tps=14838 mem=6.4G lam=0.16262648720294237 Gm=0.4372340639432271
829[full s0] step 2920/4000 loss=5.0496 gn=0.68 lr=1.61e-04 tps=14842 mem=6.4G lam=0.16277169901877642 Gm=0.44400854377696913
830[full s0] step 2940/4000 loss=4.9499 gn=0.64 lr=1.57e-04 tps=14845 mem=6.4G lam=0.16289791526893774 Gm=0.43647429657479125
831[full s0] step 2960/4000 loss=4.9569 gn=0.70 lr=1.54e-04 tps=14849 mem=6.4G lam=0.16302600068350634 Gm=0.43787237629294395
832[full s0] step 2980/4000 loss=4.9083 gn=0.64 lr=1.50e-04 tps=14852 mem=6.4G lam=0.16315330937504768 Gm=0.43096340727061033
833[full s0] step 3000/4000 loss=4.8441 gn=0.66 lr=1.47e-04 tps=14855 mem=6.4G lam=0.16329338184247413 Gm=0.4300241045032938
834[full s0] EVAL step 3000: val_loss=5.1066 val_ppl=165.11
835[full s0] step 3020/4000 loss=4.9192 gn=0.64 lr=1.44e-04 tps=14822 mem=6.4G lam=0.16343002735326687 Gm=0.4366761473938823
836[full s0] step 3040/4000 loss=4.9838 gn=0.68 lr=1.41e-04 tps=14826 mem=6.4G lam=0.1635501366108656 Gm=0.4348375468204419
837[full s0] step 3060/4000 loss=4.9574 gn=0.69 lr=1.38e-04 tps=14829 mem=6.4G lam=0.163682596758008 Gm=0.43888544384390116
838[full s0] step 3080/4000 loss=4.8064 gn=0.72 lr=1.34e-04 tps=14833 mem=6.4G lam=0.16378284866611162 Gm=0.42993523366749287
839[full s0] step 3100/4000 loss=5.0109 gn=0.75 lr=1.31e-04 tps=14836 mem=6.4G lam=0.1638804872830709 Gm=0.4370205185065667
840[full s0] step 3120/4000 loss=5.0369 gn=0.79 lr=1.28e-04 tps=14839 mem=6.4G lam=0.1640077323342363 Gm=0.43461609942217666
841[full s0] step 3140/4000 loss=4.8797 gn=0.67 lr=1.25e-04 tps=14843 mem=6.4G lam=0.16411774698644876 Gm=0.43366697058081627
842[full s0] step 3160/4000 loss=5.0279 gn=0.77 lr=1.23e-04 tps=14846 mem=6.4G lam=0.1642256739238898 Gm=0.4341176695500811
843[full s0] step 3180/4000 loss=4.8814 gn=0.68 lr=1.20e-04 tps=14850 mem=6.4G lam=0.16433581709861755 Gm=0.4314587600529194
844[full s0] step 3200/4000 loss=4.8908 gn=0.74 lr=1.17e-04 tps=14852 mem=6.4G lam=0.1644469192251563 Gm=0.43080346534649533
845[full s0] EVAL step 3200: val_loss=5.0128 val_ppl=150.33
846[full s0] step 3220/4000 loss=5.0098 gn=0.75 lr=1.14e-04 tps=14822 mem=6.4G lam=0.16455033235251904 Gm=0.423968060563008
847[full s0] step 3240/4000 loss=5.0121 gn=0.68 lr=1.12e-04 tps=14825 mem=6.4G lam=0.164661834637324 Gm=0.4364341401184599
848[full s0] step 3260/4000 loss=4.7679 gn=0.68 lr=1.09e-04 tps=14827 mem=6.4G lam=0.16475263237953186 Gm=0.42899767123162746
849[full s0] step 3280/4000 loss=5.0351 gn=0.75 lr=1.06e-04 tps=14830 mem=6.4G lam=0.16484248265624046 Gm=0.42665910472472507
850[full s0] step 3300/4000 loss=4.8429 gn=0.69 lr=1.04e-04 tps=14834 mem=6.4G lam=0.1649355192979177 Gm=0.4216356985270977
851[full s0] step 3320/4000 loss=5.0145 gn=0.71 lr=1.02e-04 tps=14837 mem=6.4G lam=0.16502051676313081 Gm=0.4199893881256382
852[full s0] step 3340/4000 loss=4.8423 gn=0.67 lr=9.92e-05 tps=14840 mem=6.4G lam=0.16509454355885586 Gm=0.4251132579520345
853[full s0] step 3360/4000 loss=5.0007 gn=0.71 lr=9.69e-05 tps=14843 mem=6.4G lam=0.16518759479125342 Gm=0.42819516733288765
854[full s0] step 3380/4000 loss=4.9616 gn=0.71 lr=9.47e-05 tps=14847 mem=6.4G lam=0.16527666803449392 Gm=0.42280324424306553
855[full s0] step 3400/4000 loss=4.8588 gn=0.69 lr=9.25e-05 tps=14850 mem=6.4G lam=0.16536484193056822 Gm=0.4210895566890637
856[full s0] EVAL step 3400: val_loss=5.0103 val_ppl=149.95
857[full s0] step 3420/4000 loss=4.8995 gn=0.70 lr=9.04e-05 tps=14820 mem=6.4G lam=0.16544855292886496 Gm=0.43159068406869966
858[full s0] step 3440/4000 loss=4.9295 gn=0.72 lr=8.84e-05 tps=14823 mem=6.4G lam=0.1655213125050068 Gm=0.432841124633948
859[full s0] step 3460/4000 loss=4.8067 gn=0.73 lr=8.65e-05 tps=14826 mem=6.4G lam=0.16559547061721483 Gm=0.42248809306571883
860[full s0] step 3480/4000 loss=4.9165 gn=0.71 lr=8.46e-05 tps=14830 mem=6.4G lam=0.16566639052083096 Gm=0.4233105207482974
861[full s0] step 3500/4000 loss=4.8292 gn=0.68 lr=8.27e-05 tps=14832 mem=6.4G lam=0.16573535868277153 Gm=0.424697866042455
862[full s0] step 3520/4000 loss=4.7182 gn=0.77 lr=8.10e-05 tps=14835 mem=6.4G lam=0.16580549317101637 Gm=0.42924326844513416
863[full s0] step 3540/4000 loss=4.7145 gn=0.73 lr=7.93e-05 tps=14838 mem=6.4G lam=0.1658791173249483 Gm=0.42203957090775174
864[full s0] step 3560/4000 loss=4.9728 gn=0.67 lr=7.77e-05 tps=14841 mem=6.4G lam=0.165962186952432 Gm=0.4248941869785388
865[full s0] step 3580/4000 loss=4.7990 gn=0.66 lr=7.61e-05 tps=14844 mem=6.4G lam=0.16603116939465204 Gm=0.4190280182907979
866[full s0] step 3600/4000 loss=4.8231 gn=0.75 lr=7.46e-05 tps=14847 mem=6.4G lam=0.16610508256902298 Gm=0.4256579965973894
867[full s0] EVAL step 3600: val_loss=4.9789 val_ppl=145.31
868[full s0] step 3620/4000 loss=4.8785 gn=0.75 lr=7.32e-05 tps=14819 mem=6.4G lam=0.16617669692883888 Gm=0.4205183299879233
869[full s0] step 3640/4000 loss=4.9570 gn=0.71 lr=7.19e-05 tps=14822 mem=6.4G lam=0.1662404810388883 Gm=0.4220012916872899
870[full s0] step 3660/4000 loss=4.9918 gn=0.80 lr=7.06e-05 tps=14825 mem=6.4G lam=0.16630153047541776 Gm=0.42589141490558785
871[full s0] step 3680/4000 loss=4.9207 gn=0.68 lr=6.94e-05 tps=14828 mem=6.4G lam=0.1663592647140225 Gm=0.416047852486372
872[full s0] step 3700/4000 loss=4.9807 gn=0.70 lr=6.83e-05 tps=14831 mem=6.4G lam=0.1664255866780877 Gm=0.4216521034638087
873[full s0] step 3720/4000 loss=4.7302 gn=0.69 lr=6.72e-05 tps=14833 mem=6.4G lam=0.16649126230428615 Gm=0.4164570265760024
874[full s0] step 3740/4000 loss=4.9217 gn=0.68 lr=6.62e-05 tps=14836 mem=6.4G lam=0.16654821671545506 Gm=0.4178890536228816
875[full s0] step 3760/4000 loss=4.8702 gn=0.73 lr=6.53e-05 tps=14839 mem=6.4G lam=0.16661168665935597 Gm=0.41870569561918575
876[full s0] step 3780/4000 loss=4.7588 gn=0.72 lr=6.45e-05 tps=14842 mem=6.4G lam=0.16666329310586056 Gm=0.4184085186570883
877[full s0] step 3800/4000 loss=4.8776 gn=0.77 lr=6.37e-05 tps=14843 mem=6.4G lam=0.16672033909708261 Gm=0.42720757083346445
878[full s0] EVAL step 3800: val_loss=4.9583 val_ppl=142.36
879[full s0] step 3820/4000 loss=4.9376 gn=0.74 lr=6.30e-05 tps=14816 mem=6.4G lam=0.16677486648162207 Gm=0.42343480822940666
880[full s0] step 3840/4000 loss=4.8648 gn=0.77 lr=6.24e-05 tps=14818 mem=6.4G lam=0.16683314150820175 Gm=0.41695012773076695
881[full s0] step 3860/4000 loss=4.8737 gn=0.70 lr=6.18e-05 tps=14820 mem=6.4G lam=0.1668974378456672 Gm=0.42252734551827115
882[full s0] step 3880/4000 loss=4.9272 gn=0.79 lr=6.13e-05 tps=14822 mem=6.4G lam=0.16694826819002628 Gm=0.428834938754638
883[full s0] step 3900/4000 loss=4.8355 gn=0.79 lr=6.09e-05 tps=14825 mem=6.4G lam=0.1669996650889516 Gm=0.41820415544013184
884[full s0] step 3920/4000 loss=4.7652 gn=0.78 lr=6.06e-05 tps=14827 mem=6.4G lam=0.16705392487347126 Gm=0.4173752376809716
885[full s0] step 3940/4000 loss=4.9381 gn=0.77 lr=6.03e-05 tps=14829 mem=6.4G lam=0.16710669671495756 Gm=0.4203516334916155
886[full s0] step 3960/4000 loss=4.8886 gn=0.73 lr=6.01e-05 tps=14832 mem=6.4G lam=0.16715015222628912 Gm=0.41584287708004314
887[full s0] step 3980/4000 loss=4.8316 gn=0.78 lr=6.00e-05 tps=14833 mem=6.4G lam=0.16720068734139204 Gm=0.4214151619623105
888[full s0] EVAL step 3999: val_loss=4.9528 val_ppl=141.57
889
890=== C:\Users\Bhorn\AppData\Local\Microsoft\WindowsApps\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\python.exe C:\Users\Bhorn\Desktop\SMATest\train.py --variant baseline --seed 0 --out C:\Users\Bhorn\Desktop\SMATest\runs\baseline_s0 --max_steps 4000 --block_size 256 --batch_size 16 --grad_accum 2 --eval_interval 200 --data_seed 1337 ===
891
892
893=== C:\Users\Bhorn\AppData\Local\Microsoft\WindowsApps\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\python.exe C:\Users\Bhorn\Desktop\SMATest\train.py --variant s_only --seed 0 --out C:\Users\Bhorn\Desktop\SMATest\runs\s_only_s0 --max_steps 4000 --block_size 256 --batch_size 16 --grad_accum 2 --eval_interval 200 --data_seed 1337 ===
894
895
896=== C:\Users\Bhorn\AppData\Local\Microsoft\WindowsApps\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\python.exe C:\Users\Bhorn\Desktop\SMATest\train.py --variant g_only --seed 0 --out C:\Users\Bhorn\Desktop\SMATest\runs\g_only_s0 --max_steps 4000 --block_size 256 --batch_size 16 --grad_accum 2 --eval_interval 200 --data_seed 1337 ===
897
898
899=== C:\Users\Bhorn\AppData\Local\Microsoft\WindowsApps\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\python.exe C:\Users\Bhorn\Desktop\SMATest\train.py --variant full --seed 0 --out C:\Users\Bhorn\Desktop\SMATest\runs\full_s0 --max_steps 4000 --block_size 256 --batch_size 16 --grad_accum 2 --eval_interval 200 --data_seed 1337 ===
900
901 