update

logs_qkvo_pure/muon_lr_search/mode_muon_adam_lr_0.002_muon_lr_0.0005_seed_42.log

@@ -0,0 +1,708 @@
+step:0 validation loss:11.018949
+step:0 train loss:11.015424
+step:1 train loss:11.020226
+step:2 train loss:11.010934
+step:3 train loss:11.004910
+step:4 train loss:10.995027
+step:5 train loss:10.983676
+step:6 train loss:10.966785
+step:7 train loss:10.951075
+step:8 train loss:10.930555
+step:9 train loss:10.909342
+step:10 train loss:10.881525
+step:11 train loss:10.860868
+step:12 train loss:10.825106
+step:13 train loss:10.794963
+step:14 train loss:10.756892
+step:15 train loss:10.723540
+step:16 train loss:10.684589
+step:17 train loss:10.645847
+step:18 train loss:10.600722
+step:19 train loss:10.553440
+step:20 train loss:10.501952
+step:21 train loss:10.457741
+step:22 train loss:10.394848
+step:23 train loss:10.350867
+step:24 train loss:10.280701
+step:25 train loss:10.234509
+step:26 train loss:10.165276
+step:27 train loss:10.098972
+step:28 train loss:10.046017
+step:29 train loss:9.981583
+step:30 train loss:9.910318
+step:31 train loss:9.830559
+step:32 train loss:9.757634
+step:33 train loss:9.686543
+step:34 train loss:9.624190
+step:35 train loss:9.534348
+step:36 train loss:9.458261
+step:37 train loss:9.367187
+step:38 train loss:9.302447
+step:39 train loss:9.206266
+step:40 train loss:9.131829
+step:41 train loss:9.039160
+step:42 train loss:8.981525
+step:43 train loss:8.865849
+step:44 train loss:8.802135
+step:45 train loss:8.709974
+step:46 train loss:8.649731
+step:47 train loss:8.561390
+step:48 train loss:8.481024
+step:49 train loss:8.391659
+step:50 train loss:8.305132
+step:51 train loss:8.221050
+step:52 train loss:8.185582
+step:53 train loss:8.105663
+step:54 train loss:8.051863
+step:55 train loss:7.957572
+step:56 train loss:7.902859
+step:57 train loss:7.874125
+step:58 train loss:7.783765
+step:59 train loss:7.759455
+step:60 train loss:7.721583
+step:61 train loss:7.685786
+step:62 train loss:7.657363
+step:63 train loss:7.689013
+step:64 train loss:7.598711
+step:65 train loss:7.607277
+step:66 train loss:7.624282
+step:67 train loss:7.625703
+step:68 train loss:7.613712
+step:69 train loss:7.600225
+step:70 train loss:7.590165
+step:71 train loss:7.564875
+step:72 train loss:7.594479
+step:73 train loss:7.562237
+step:74 train loss:7.588921
+step:75 train loss:7.537924
+step:76 train loss:7.613895
+step:77 train loss:7.543286
+step:78 train loss:7.379977
+step:79 train loss:7.484556
+step:80 train loss:7.466608
+step:81 train loss:7.517383
+step:82 train loss:7.491320
+step:83 train loss:7.456035
+step:84 train loss:7.421113
+step:85 train loss:7.392701
+step:86 train loss:7.375344
+step:87 train loss:7.341681
+step:88 train loss:7.342148
+step:89 train loss:7.299115
+step:90 train loss:7.324209
+step:91 train loss:7.316027
+step:92 train loss:7.304208
+step:93 train loss:7.238531
+step:94 train loss:7.206048
+step:95 train loss:7.142555
+step:96 train loss:7.217937
+step:97 train loss:7.148865
+step:98 train loss:7.135708
+step:99 train loss:7.083437
+step:100 train loss:7.132834
+step:101 train loss:7.001304
+step:102 train loss:6.993102
+step:103 train loss:6.967243
+step:104 train loss:6.993345
+step:105 train loss:7.031693
+step:106 train loss:6.963967
+step:107 train loss:6.910224
+step:108 train loss:6.915035
+step:109 train loss:6.945443
+step:110 train loss:6.850996
+step:111 train loss:6.856386
+step:112 train loss:6.833060
+step:113 train loss:6.782409
+step:114 train loss:6.833003
+step:115 train loss:6.776278
+step:116 train loss:6.741609
+step:117 train loss:6.672617
+step:118 train loss:6.726232
+step:119 train loss:6.663210
+step:120 train loss:6.671381
+step:121 train loss:6.581036
+step:122 train loss:6.671529
+step:123 train loss:6.583089
+step:124 train loss:6.565653
+step:125 train loss:6.533828
+step:126 train loss:6.628871
+step:127 train loss:6.533030
+step:128 train loss:6.575610
+step:129 train loss:6.548450
+step:130 train loss:6.575078
+step:131 train loss:6.513773
+step:132 train loss:6.431153
+step:133 train loss:6.492492
+step:134 train loss:6.463880
+step:135 train loss:6.371145
+step:136 train loss:6.410591
+step:137 train loss:6.407073
+step:138 train loss:6.343710
+step:139 train loss:6.419179
+step:140 train loss:6.328504
+step:141 train loss:6.426642
+step:142 train loss:6.371757
+step:143 train loss:6.378293
+step:144 train loss:6.349747
+step:145 train loss:6.282640
+step:146 train loss:6.293992
+step:147 train loss:6.345430
+step:148 train loss:6.353338
+step:149 train loss:6.303926
+step:150 train loss:6.302433
+step:151 train loss:6.212167
+step:152 train loss:6.255791
+step:153 train loss:6.232176
+step:154 train loss:6.314730
+step:155 train loss:6.285755
+step:156 train loss:6.317771
+step:157 train loss:6.215672
+step:158 train loss:6.200233
+step:159 train loss:6.232330
+step:160 train loss:6.215403
+step:161 train loss:6.209176
+step:162 train loss:6.178005
+step:163 train loss:6.194960
+step:164 train loss:6.194765
+step:165 train loss:6.214000
+step:166 train loss:6.159634
+step:167 train loss:6.163892
+step:168 train loss:6.136242
+step:169 train loss:6.088985
+step:170 train loss:6.059000
+step:171 train loss:6.173427
+step:172 train loss:6.100713
+step:173 train loss:6.151384
+step:174 train loss:6.150017
+step:175 train loss:6.116456
+step:176 train loss:6.075884
+step:177 train loss:6.113425
+step:178 train loss:6.121142
+step:179 train loss:6.077321
+step:180 train loss:6.059571
+step:181 train loss:6.096775
+step:182 train loss:6.030660
+step:183 train loss:6.114416
+step:184 train loss:6.081723
+step:185 train loss:6.014050
+step:186 train loss:6.151138
+step:187 train loss:6.083933
+step:188 train loss:5.923741
+step:189 train loss:6.064409
+step:190 train loss:6.047544
+step:191 train loss:6.020611
+step:192 train loss:5.892870
+step:193 train loss:6.085207
+step:194 train loss:6.053609
+step:195 train loss:6.033236
+step:196 train loss:6.055465
+step:197 train loss:6.013649
+step:198 train loss:6.016072
+step:199 train loss:6.045198
+step:200 train loss:6.079931
+step:201 train loss:6.015906
+step:202 train loss:6.050323
+step:203 train loss:5.989286
+step:204 train loss:5.994330
+step:205 train loss:5.914724
+step:206 train loss:6.008636
+step:207 train loss:5.965271
+step:208 train loss:5.958157
+step:209 train loss:5.943935
+step:210 train loss:5.932757
+step:211 train loss:5.979126
+step:212 train loss:5.957274
+step:213 train loss:5.940977
+step:214 train loss:5.975878
+step:215 train loss:5.958874
+step:216 train loss:5.921475
+step:217 train loss:5.931257
+step:218 train loss:5.894432
+step:219 train loss:5.879438
+step:220 train loss:5.918484
+step:221 train loss:5.874071
+step:222 train loss:5.957614
+step:223 train loss:5.922194
+step:224 train loss:5.911490
+step:225 train loss:5.887461
+step:226 train loss:5.861408
+step:227 train loss:5.919147
+step:228 train loss:5.916242
+step:229 train loss:5.963889
+step:230 train loss:5.839691
+step:231 train loss:5.880941
+step:232 train loss:5.913732
+step:233 train loss:5.815922
+step:234 train loss:5.881902
+step:235 train loss:5.926662
+step:236 train loss:5.892923
+step:237 train loss:5.926398
+step:238 train loss:5.896809
+step:239 train loss:5.812529
+step:240 train loss:5.929081
+step:241 train loss:5.949173
+step:242 train loss:5.919721
+step:243 train loss:5.849203
+step:244 train loss:5.853334
+step:245 train loss:5.828572
+step:246 train loss:5.803466
+step:247 train loss:5.872789
+step:248 train loss:5.799171
+step:249 train loss:5.846122
+step:250 validation loss:5.837239
+step:250 train loss:5.816980
+step:251 train loss:5.867065
+step:252 train loss:5.802760
+step:253 train loss:5.809064
+step:254 train loss:5.793290
+step:255 train loss:5.837733
+step:256 train loss:5.812046
+step:257 train loss:5.885970
+step:258 train loss:5.763566
+step:259 train loss:5.792757
+step:260 train loss:5.785553
+step:261 train loss:5.788903
+step:262 train loss:5.795781
+step:263 train loss:5.845345
+step:264 train loss:5.759592
+step:265 train loss:5.797788
+step:266 train loss:5.744943
+step:267 train loss:5.810608
+step:268 train loss:5.760835
+step:269 train loss:5.788585
+step:270 train loss:5.814303
+step:271 train loss:5.759589
+step:272 train loss:5.797612
+step:273 train loss:5.807883
+step:274 train loss:5.699788
+step:275 train loss:5.808730
+step:276 train loss:5.759241
+step:277 train loss:5.721187
+step:278 train loss:5.750453
+step:279 train loss:5.729042
+step:280 train loss:5.756226
+step:281 train loss:5.825380
+step:282 train loss:5.731263
+step:283 train loss:5.722671
+step:284 train loss:5.741635
+step:285 train loss:5.757961
+step:286 train loss:5.728635
+step:287 train loss:5.741354
+step:288 train loss:5.703504
+step:289 train loss:5.752779
+step:290 train loss:5.761066
+step:291 train loss:5.719710
+step:292 train loss:5.757103
+step:293 train loss:5.743849
+step:294 train loss:5.707259
+step:295 train loss:5.792729
+step:296 train loss:5.743566
+step:297 train loss:5.782072
+step:298 train loss:5.710087
+step:299 train loss:5.742099
+step:300 train loss:5.678102
+step:301 train loss:5.720137
+step:302 train loss:5.717441
+step:303 train loss:5.693343
+step:304 train loss:5.749202
+step:305 train loss:5.675310
+step:306 train loss:5.693254
+step:307 train loss:5.717690
+step:308 train loss:5.653879
+step:309 train loss:5.751453
+step:310 train loss:5.731911
+step:311 train loss:5.682853
+step:312 train loss:5.714482
+step:313 train loss:5.705677
+step:314 train loss:5.719196
+step:315 train loss:5.675369
+step:316 train loss:5.646206
+step:317 train loss:5.636640
+step:318 train loss:5.653316
+step:319 train loss:5.697946
+step:320 train loss:5.660789
+step:321 train loss:5.691439
+step:322 train loss:5.686815
+step:323 train loss:5.737013
+step:324 train loss:5.705341
+step:325 train loss:5.698605
+step:326 train loss:5.708385
+step:327 train loss:5.711277
+step:328 train loss:5.697366
+step:329 train loss:5.665713
+step:330 train loss:5.641728
+step:331 train loss:5.659482
+step:332 train loss:5.616224
+step:333 train loss:5.608384
+step:334 train loss:5.681748
+step:335 train loss:5.715389
+step:336 train loss:5.886877
+step:337 train loss:5.711463
+step:338 train loss:5.629460
+step:339 train loss:5.635762
+step:340 train loss:5.628103
+step:341 train loss:5.603494
+step:342 train loss:5.683806
+step:343 train loss:5.627341
+step:344 train loss:5.670202
+step:345 train loss:5.585654
+step:346 train loss:5.626231
+step:347 train loss:5.590869
+step:348 train loss:5.580972
+step:349 train loss:5.514933
+step:350 train loss:5.586253
+step:351 train loss:5.645633
+step:352 train loss:5.556901
+step:353 train loss:5.636076
+step:354 train loss:5.603020
+step:355 train loss:5.613499
+step:356 train loss:5.600191
+step:357 train loss:5.665856
+step:358 train loss:5.681579
+step:359 train loss:5.516611
+step:360 train loss:5.693751
+step:361 train loss:5.635183
+step:362 train loss:5.577714
+step:363 train loss:5.634393
+step:364 train loss:5.628038
+step:365 train loss:5.658887
+step:366 train loss:5.602283
+step:367 train loss:5.626603
+step:368 train loss:5.613094
+step:369 train loss:5.553933
+step:370 train loss:5.643659
+step:371 train loss:5.569902
+step:372 train loss:5.635789
+step:373 train loss:5.576738
+step:374 train loss:5.570609
+step:375 train loss:5.630656
+step:376 train loss:5.578053
+step:377 train loss:5.498881
+step:378 train loss:5.599305
+step:379 train loss:5.615202
+step:380 train loss:5.559390
+step:381 train loss:5.627434
+step:382 train loss:5.599933
+step:383 train loss:5.574039
+step:384 train loss:5.547537
+step:385 train loss:5.573546
+step:386 train loss:5.589078
+step:387 train loss:5.617164
+step:388 train loss:5.590526
+step:389 train loss:5.520203
+step:390 train loss:5.590436
+step:391 train loss:5.523112
+step:392 train loss:5.593059
+step:393 train loss:5.609905
+step:394 train loss:5.566988
+step:395 train loss:5.553980
+step:396 train loss:5.457115
+step:397 train loss:5.629135
+step:398 train loss:5.544865
+step:399 train loss:5.528760
+step:400 train loss:5.588879
+step:401 train loss:5.521435
+step:402 train loss:5.565764
+step:403 train loss:5.565872
+step:404 train loss:5.526936
+step:405 train loss:5.555054
+step:406 train loss:5.563742
+step:407 train loss:5.623383
+step:408 train loss:5.579584
+step:409 train loss:5.521736
+step:410 train loss:5.529564
+step:411 train loss:5.540267
+step:412 train loss:5.595044
+step:413 train loss:5.504812
+step:414 train loss:5.569297
+step:415 train loss:5.593074
+step:416 train loss:5.525351
+step:417 train loss:5.522644
+step:418 train loss:5.518416
+step:419 train loss:5.532266
+step:420 train loss:5.519451
+step:421 train loss:5.511771
+step:422 train loss:5.497843
+step:423 train loss:5.512751
+step:424 train loss:5.516061
+step:425 train loss:5.553890
+step:426 train loss:5.492079
+step:427 train loss:5.500880
+step:428 train loss:5.491987
+step:429 train loss:5.490248
+step:430 train loss:5.480282
+step:431 train loss:5.574747
+step:432 train loss:5.501834
+step:433 train loss:5.563447
+step:434 train loss:5.524998
+step:435 train loss:5.542096
+step:436 train loss:5.559086
+step:437 train loss:5.535322
+step:438 train loss:5.519480
+step:439 train loss:5.499341
+step:440 train loss:5.490997
+step:441 train loss:5.491372
+step:442 train loss:5.494779
+step:443 train loss:5.497370
+step:444 train loss:5.549359
+step:445 train loss:5.501597
+step:446 train loss:5.514030
+step:447 train loss:5.513056
+step:448 train loss:5.528311
+step:449 train loss:5.500819
+step:450 train loss:5.486154
+step:451 train loss:5.572198
+step:452 train loss:5.482662
+step:453 train loss:5.525918
+step:454 train loss:5.441131
+step:455 train loss:5.509805
+step:456 train loss:5.498071
+step:457 train loss:5.483315
+step:458 train loss:5.497639
+step:459 train loss:5.479138
+step:460 train loss:5.540781
+step:461 train loss:5.506932
+step:462 train loss:5.406518
+step:463 train loss:5.530899
+step:464 train loss:5.517510
+step:465 train loss:5.480126
+step:466 train loss:5.505144
+step:467 train loss:5.499542
+step:468 train loss:5.508452
+step:469 train loss:5.480200
+step:470 train loss:5.431437
+step:471 train loss:5.580779
+step:472 train loss:5.481690
+step:473 train loss:5.493110
+step:474 train loss:5.503496
+step:475 train loss:5.514795
+step:476 train loss:5.455892
+step:477 train loss:5.467666
+step:478 train loss:5.472945
+step:479 train loss:5.428407
+step:480 train loss:5.519136
+step:481 train loss:5.456609
+step:482 train loss:5.399099
+step:483 train loss:5.494047
+step:484 train loss:5.468803
+step:485 train loss:5.434709
+step:486 train loss:5.497624
+step:487 train loss:5.452084
+step:488 train loss:5.450278
+step:489 train loss:5.472017
+step:490 train loss:5.437662
+step:491 train loss:5.435221
+step:492 train loss:5.510285
+step:493 train loss:5.423973
+step:494 train loss:5.430881
+step:495 train loss:5.490462
+step:496 train loss:5.434524
+step:497 train loss:5.476788
+step:498 train loss:5.470979
+step:499 train loss:5.493303
+step:500 validation loss:5.454007
+step:500 train loss:5.451817
+step:501 train loss:5.508993
+step:502 train loss:5.399317
+step:503 train loss:5.412842
+step:504 train loss:5.471434
+step:505 train loss:5.461817
+step:506 train loss:5.398189
+step:507 train loss:5.513207
+step:508 train loss:5.433660
+step:509 train loss:5.454690
+step:510 train loss:5.432570
+step:511 train loss:5.375609
+step:512 train loss:5.444704
+step:513 train loss:5.452021
+step:514 train loss:5.471115
+step:515 train loss:5.446244
+step:516 train loss:5.471374
+step:517 train loss:5.474245
+step:518 train loss:5.426744
+step:519 train loss:5.496078
+step:520 train loss:5.427716
+step:521 train loss:5.397823
+step:522 train loss:5.475583
+step:523 train loss:5.408001
+step:524 train loss:5.452801
+step:525 train loss:5.367191
+step:526 train loss:5.456772
+step:527 train loss:5.416982
+step:528 train loss:5.443983
+step:529 train loss:5.428410
+step:530 train loss:5.424571
+step:531 train loss:5.399049
+step:532 train loss:5.414495
+step:533 train loss:5.425094
+step:534 train loss:5.445930
+step:535 train loss:5.494602
+step:536 train loss:5.402138
+step:537 train loss:5.384509
+step:538 train loss:5.368041
+step:539 train loss:5.503932
+step:540 train loss:5.357913
+step:541 train loss:5.460105
+step:542 train loss:5.453763
+step:543 train loss:5.418608
+step:544 train loss:5.448548
+step:545 train loss:5.386162
+step:546 train loss:5.369950
+step:547 train loss:5.398388
+step:548 train loss:5.406528
+step:549 train loss:5.379954
+step:550 train loss:5.410077
+step:551 train loss:5.423714
+step:552 train loss:5.506493
+step:553 train loss:5.408998
+step:554 train loss:5.415755
+step:555 train loss:5.487288
+step:556 train loss:5.394042
+step:557 train loss:5.386750
+step:558 train loss:5.401230
+step:559 train loss:5.472310
+step:560 train loss:5.396537
+step:561 train loss:5.383673
+step:562 train loss:5.387293
+step:563 train loss:5.387549
+step:564 train loss:5.437938
+step:565 train loss:5.409759
+step:566 train loss:5.422817
+step:567 train loss:5.399188
+step:568 train loss:5.429495
+step:569 train loss:5.401677
+step:570 train loss:5.384977
+step:571 train loss:5.373534
+step:572 train loss:5.380027
+step:573 train loss:5.375136
+step:574 train loss:5.431943
+step:575 train loss:5.364378
+step:576 train loss:5.378711
+step:577 train loss:5.442268
+step:578 train loss:5.399104
+step:579 train loss:5.390522
+step:580 train loss:5.407828
+step:581 train loss:5.411538
+step:582 train loss:5.410308
+step:583 train loss:5.402235
+step:584 train loss:5.373002
+step:585 train loss:5.390867
+step:586 train loss:5.398320
+step:587 train loss:5.411067
+step:588 train loss:5.373336
+step:589 train loss:5.408735
+step:590 train loss:5.451546
+step:591 train loss:5.338789
+step:592 train loss:5.334523
+step:593 train loss:5.389324
+step:594 train loss:5.337025
+step:595 train loss:5.416079
+step:596 train loss:5.396035
+step:597 train loss:5.350927
+step:598 train loss:5.395294
+step:599 train loss:5.385292
+step:600 train loss:5.368805
+step:601 train loss:5.320899
+step:602 train loss:5.354075
+step:603 train loss:5.408556
+step:604 train loss:5.403461
+step:605 train loss:5.399994
+step:606 train loss:5.304999
+step:607 train loss:5.384509
+step:608 train loss:5.335430
+step:609 train loss:5.346176
+step:610 train loss:5.349381
+step:611 train loss:5.374700
+step:612 train loss:5.327247
+step:613 train loss:5.337881
+step:614 train loss:5.386038
+step:615 train loss:5.371188
+step:616 train loss:5.348369
+step:617 train loss:5.364955
+step:618 train loss:5.332599
+step:619 train loss:5.401806
+step:620 train loss:5.339460
+step:621 train loss:5.395061
+step:622 train loss:5.349128
+step:623 train loss:5.391778
+step:624 train loss:5.365892
+step:625 train loss:5.366135
+step:626 train loss:5.391775
+step:627 train loss:5.324592
+step:628 train loss:5.342957
+step:629 train loss:5.292043
+step:630 train loss:5.341146
+step:631 train loss:5.347159
+step:632 train loss:5.333919
+step:633 train loss:5.353924
+step:634 train loss:5.363871
+step:635 train loss:5.350885
+step:636 train loss:5.310527
+step:637 train loss:5.286098
+step:638 train loss:5.311001
+step:639 train loss:5.338397
+step:640 train loss:5.358508
+step:641 train loss:5.355056
+step:642 train loss:5.352121
+step:643 train loss:5.348762
+step:644 train loss:5.330700
+step:645 train loss:5.354775
+step:646 train loss:5.333034
+step:647 train loss:5.355504
+step:648 train loss:5.422951
+step:649 train loss:5.357683
+step:650 train loss:5.341992
+step:651 train loss:5.279242
+step:652 train loss:5.295226
+step:653 train loss:5.349594
+step:654 train loss:5.264517
+step:655 train loss:5.341141
+step:656 train loss:5.357602
+step:657 train loss:5.281108
+step:658 train loss:5.353639
+step:659 train loss:5.306508
+step:660 train loss:5.371544
+step:661 train loss:5.363912
+step:662 train loss:5.332908
+step:663 train loss:5.328643
+step:664 train loss:5.279181
+step:665 train loss:5.292587
+step:666 train loss:5.296033
+step:667 train loss:5.315818
+step:668 train loss:5.320154
+step:669 train loss:5.319018
+step:670 train loss:5.289472
+step:671 train loss:5.277840
+step:672 train loss:5.365256
+step:673 train loss:5.327517
+step:674 train loss:5.312008
+step:675 train loss:5.289848
+step:676 train loss:5.308525
+step:677 train loss:5.311978
+step:678 train loss:5.257811
+step:679 train loss:5.322995
+step:680 train loss:5.304199
+step:681 train loss:5.282611
+step:682 train loss:5.321229
+step:683 train loss:5.353194
+step:684 train loss:5.264623
+step:685 train loss:5.252784
+step:686 train loss:5.380219
+step:687 train loss:5.297470
+step:688 train loss:5.233657
+step:689 train loss:5.325809
+step:690 train loss:5.268259
+step:691 train loss:5.268190
+step:692 train loss:5.331085
+step:693 train loss:5.258225
+step:694 train loss:5.249166
+step:695 train loss:5.260811
+step:696 train loss:5.283231
+step:697 train loss:5.276121
+step:698 train loss:5.268666
+step:699 train loss:5.309930
+step:700 train loss:5.288491
+step:701 train loss:5.256346
+step:702 train loss:5.307794
+step:703 train loss:5.206234
+step:704 train loss:5.251676

logs_qkvo_pure/muon_lr_search_new/mode_muon_adam_lr_0.002_muon_lr_0.02_seed_42.log

@@ -0,0 +1,2373 @@
+"""
+Reference code for GPT-2 training and inference.
+Will save the model weights into files, to be read from C as initialization.
+
+References:
+1) the official GPT-2 TensorFlow implementation released by OpenAI:
+https://github.com/openai/gpt-2/blob/master/src/model.py
+2) huggingface/transformers PyTorch implementation:
+https://github.com/huggingface/transformers/blob/main/src/transformers/models/gpt2/modeling_gpt2.py
+
+Example launches to only benchmark the speed of bfloat16 compiled GPU training:
+1 GPU:
+python train_gpt2.py --write_tensors=0 --num_iterations=50 --sequence_length=1024 --compile=1 --tensorcores=1 --dtype=bfloat16
+you can also turn on flash-attention by appending --flash=1
+4 GPU:
+torchrun --standalone --nproc_per_node=4 train_gpt2.py --write_tensors=0 --num_iterations=50 --sequence_length=1024 --compile=1 --tensorcores=1 --dtype=bfloat16
+"""
+import sys
+with open(sys.argv[0]) as f:
+    code = f.read() # read the code of this file ASAP, for logging
+
+import os
+import math
+import glob
+import struct
+import inspect
+from contextlib import nullcontext
+from dataclasses import dataclass
+import random
+
+import numpy as np
+import torch
+from torch import Tensor
+import torch.nn as nn
+from torch.nn import functional as F
+import torch._inductor.config as config
+from torch.nn.parallel import DistributedDataParallel as DDP
+from torch.distributed import init_process_group, destroy_process_group
+from torch.distributed.optim import ZeroRedundancyOptimizer
+import torch.distributed as dist
+
+# Import Muon optimizer
+import sys
+sys.path.append("/home/aiops/zhangfz/MUON_theory_copy/MUON_theory/modded-nanogpt/optimizers")
+from MUON_fix import Muon
+
+# Import GPT model
+sys.path.append("/home/aiops/zhangfz/MUON_theory_copy/MUON_theory/modded-nanogpt/models")
+from nano_GPT_qkvo_pure import GPT, GPTConfig
+
+
+# -----------------------------------------------------------------------------
+# Our own simple Distributed Data Loader
+
+def _peek_data_shard(filename):
+    # only reads the header, returns header data
+    with open(filename, "rb") as f:
+        # first read the header, which is 256 int32 integers (4 bytes each)
+        header = np.frombuffer(f.read(256*4), dtype=np.int32)
+    if header[0] != 20240520:
+        print("ERROR: magic number mismatch in the data .bin file!")
+        print("---> HINT: Are you passing in a correct file with --input_bin?")
+        print("---> HINT: Dataset encoding changed recently, re-run data prepro or refer again to README")
+        print("---> HINT: For example re-run: `python dev/data/tinyshakespeare.py`, then re-try")
+        exit(1)
+    assert header[1] == 1, "unsupported version"
+    ntok = header[2] # number of tokens (claimed)
+    return ntok # for now just return the number of tokens
+
+def _load_data_shard(filename):
+    with open(filename, "rb") as f:
+        # first read the header, which is 256 int32 integers (4 bytes each)
+        header = np.frombuffer(f.read(256*4), dtype=np.int32)
+        assert header[0] == 20240520, "magic number mismatch in the data .bin file"
+        assert header[1] == 1, "unsupported version"
+        ntok = header[2] # number of tokens (claimed)
+        # the rest of it are tokens, stored as uint16
+        tokens = np.frombuffer(f.read(), dtype=np.uint16)
+    assert len(tokens) == ntok, "number of tokens read does not match header?"
+    return tokens
+
+class DistributedDataLoader:
+    def __init__(self, filename_pattern, B, T, process_rank, num_processes):
+        self.process_rank = process_rank
+        self.num_processes = num_processes
+        self.B = B
+        self.T = T
+
+        # glob files that match the pattern
+        self.files = sorted(glob.glob(filename_pattern))
+        assert len(self.files) > 0, f"did not find any files that match the pattern {filename_pattern}"
+
+        # load and validate all data shards, count number of tokens in total
+        ntok_total = 0
+        for fname in self.files:
+            shard_ntok = _peek_data_shard(fname)
+            assert shard_ntok >= num_processes * B * T + 1
+            ntok_total += shard_ntok
+        self.ntok_total = ntok_total
+        print0(f"DataLoader: total number of tokens: {ntok_total:,} across {len(self.files)} files")
+
+        # kick things off
+        self.current_shard = None
+        self.reset()
+
+    def reset(self):
+        # we're being a bit clever here: if we already had shard 0 loaded,
+        # then don't do the work to reload it, just reset the pointer
+        if self.current_shard != 0:
+            self.current_shard = 0
+            self.tokens = _load_data_shard(self.files[self.current_shard])
+        self.current_position = self.process_rank * self.B * self.T
+
+    def advance(self): # advance to next data shard
+        self.current_shard = (self.current_shard + 1) % len(self.files)
+        self.current_position = self.process_rank * self.B * self.T
+        self.tokens = _load_data_shard(self.files[self.current_shard])
+
+    def next_batch(self):
+        B = self.B
+        T = self.T
+        buf = self.tokens[self.current_position : self.current_position+B*T+1]
+        buf = torch.tensor(buf.astype(np.int32), dtype=torch.long)
+        x = (buf[:-1]).view(B, T) # inputs
+        y = (buf[1:]).view(B, T) # targets
+        # advance the start pointer in current shard
+        self.current_position += B * T * self.num_processes
+        # if loading the next batch would be out of bounds advance the shard
+        if self.current_position + (B * T * self.num_processes + 1) > len(self.tokens):
+            self.advance()
+        return x, y
+
+# -----------------------------------------------------------------------------
+# Python -> C bridge utilities for saving params/grads/activations to .bin files
+
+def write_fp32(tensor, file):
+    t = tensor.detach().cpu().to(torch.float32)
+    b = t.numpy().tobytes()
+    file.write(b)
+
+def write_bf16(tensor, file):
+    t = tensor.detach().cpu().to(torch.bfloat16)
+    # numpy doesn't have bf16 datatype so we have to trick it
+    t = t.view(torch.int16) # trick: reinterpret as int16
+    b = t.numpy().tobytes()
+    file.write(b)
+
+def write_tensors(model_tensors, L, file, dtype):
+    # writes the GPT-2 model's weights to a binary file
+    assert dtype in {"float32", "bfloat16"}
+    write_fun = write_fp32 if dtype == "float32" else write_bf16
+    write_fun(model_tensors["transformer.wte.weight"], file) # (V, C)
+    write_fun(model_tensors["transformer.wpe.weight"], file) # (T, C)
+    for i in range(L): # (L, C)
+        write_fun(model_tensors[f"transformer.h.{i}.ln_1.weight"], file)
+    for i in range(L): # (L, C)
+        write_fun(model_tensors[f"transformer.h.{i}.ln_1.bias"], file)
+    for i in range(L): # (L, 3C, C)
+        write_fun(model_tensors[f"transformer.h.{i}.attn.c_attn.weight"], file)
+    for i in range(L): # (L, 3C)
+        write_fun(model_tensors[f"transformer.h.{i}.attn.c_attn.bias"], file)
+    for i in range(L): # (L, C, C)
+        write_fun(model_tensors[f"transformer.h.{i}.attn.c_proj.weight"], file)
+    for i in range(L): # (L, C)
+        write_fun(model_tensors[f"transformer.h.{i}.attn.c_proj.bias"], file)
+    for i in range(L): # (L, C)
+        write_fun(model_tensors[f"transformer.h.{i}.ln_2.weight"], file)
+    for i in range(L): # (L, C)
+        write_fun(model_tensors[f"transformer.h.{i}.ln_2.bias"], file)
+    for i in range(L): # (L, 4C, C)
+        write_fun(model_tensors[f"transformer.h.{i}.mlp.c_fc.weight"], file)
+    for i in range(L): # (L, 4C)
+        write_fun(model_tensors[f"transformer.h.{i}.mlp.c_fc.bias"], file)
+    for i in range(L): # (L, C, 4C)
+        write_fun(model_tensors[f"transformer.h.{i}.mlp.c_proj.weight"], file)
+    for i in range(L): # (L, C)
+        write_fun(model_tensors[f"transformer.h.{i}.mlp.c_proj.bias"], file)
+    write_fun(model_tensors["transformer.ln_f.weight"], file) # (C, )
+    write_fun(model_tensors["transformer.ln_f.bias"], file) # (C, )
+
+@torch.no_grad()
+def pad_vocab(tensor, multiple=128, value=0):
+    """
+    The dimension of the vocab size in GPT-2 is 50,257
+    which is unfortunately a very unfriendly number for a lot of
+    matrix operations on the GPU. So we pad it to the nearest
+    friendlier multiple, e.g. 50,304 if multiple=128 when we
+    export the weights into C land. This is a NOOP algorithmically
+    and is only done to make the tensor operations more efficient.
+    """
+    assert tensor.ndim == 2
+    V, C = tensor.shape
+    assert V == 50257, "just being defensive here"
+    # calculate padded vocab size by rounding up to nearest multiple
+    Vp = ((V + multiple - 1) // multiple) * multiple
+    # pad the tensor
+    pad_rows = Vp - V
+    padded = tensor if pad_rows == 0 else F.pad(tensor, (0, 0, 0, pad_rows), value=value)
+    assert padded.shape == (Vp, C)
+    return padded
+
+def write_model(model, filename, dtype):
+    # everything we need to instantiate the model
+    # 1) header is: version int, GPTConfig ints, padding to 1024 bytes
+    assert dtype in {"float32", "bfloat16"} # float16 todo maybe later
+    version = {
+        "float32": 3, # 3: all tensors are fp32, padded vocab
+        "bfloat16": 5, # 5: all tensors are bf16, padded vocab
+    }[dtype]
+    header = torch.zeros(256, dtype=torch.int32)
+    header[0] = 20240326 # magic
+    header[1] = version # checkpoint version
+    header[2] = model.config.block_size
+    header[3] = model.config.vocab_size
+    header[4] = model.config.n_layer
+    header[5] = model.config.n_head
+    header[6] = model.config.n_embd
+    # 2) the parameters follow the header
+    params = {name: param.cpu() for name, param in model.named_parameters()}
+    # pad the vocab to a multiple of 128 here at export, for efficiency in C
+    wte = params["transformer.wte.weight"] # (V, C)
+    wte_padded = pad_vocab(wte) # (Vp, C)
+    params["transformer.wte.weight"] = wte_padded # (Vp, C)
+    print(f"padded vocab size from {wte.size(0)} to {wte_padded.size(0)}")
+    header[7] = wte_padded.size(0) # padded vocab size store in header
+    # now write to file
+    with open(filename, "wb") as file:
+        file.write(header.numpy().tobytes()) # header
+        write_tensors(params, model.config.n_layer, file, dtype) # params
+    print(f"wrote {filename}")
+
+def write_state(model, x, y, logits, loss, filename):
+    # the state is used for debugging.
+    # it contains information about the input, logits, loss, and the parameter gradients
+    # this can be used for checking the computation correctness in C
+    header = torch.zeros(256, dtype=torch.int32)
+    header[0] = 20240327 # magic
+    header[1] = 2 # run state version = 2 (1 -> 2 for padded vocab changes)
+    header[2] = x.size(0) # batch size of the batch, B
+    header[3] = x.size(1) # temporal extent of the batch, T
+    grads = {name: param.grad.cpu() for name, param in model.named_parameters()}
+    # pad the vocab grads here as well, to mirror write_model
+    wte_grad = grads["transformer.wte.weight"] # (V, C)
+    wte_grad_padded = pad_vocab(wte_grad, value=0) # (Vp, C) # TODO later maybe pad with nan?
+    grads["transformer.wte.weight"] = wte_grad_padded # (Vp, C)
+    print(f"padded vocab size in reference grads from {wte_grad.size(0)} to {wte_grad_padded.size(0)}")
+    with open(filename, "wb") as file:
+        # header
+        file.write(header.numpy().tobytes())
+        # input x
+        file.write(x.cpu().numpy().astype("int32").tobytes()) # (B, T)
+        # targets y
+        file.write(y.cpu().numpy().astype("int32").tobytes()) # (B, T)
+        # logits (result of the model forward pass)
+        write_fp32(logits.cpu(), file)
+        # loss (single float, result of the cross entropy loss)
+        write_fp32(loss.cpu(), file)
+        # gradients
+        write_tensors(grads, model.config.n_layer, file, "float32")
+    print(f"wrote {filename}")
+
+def write_tokenizer(enc, filename):
+    n = enc.max_token_value + 1
+    header = torch.zeros(256, dtype=torch.int32)
+    header[0] = 20240328 # magic
+    header[1] = 2 # tokenizer version = 2 (1 -> 2: includes EOT token)
+    header[2] = n # number of tokens
+    header[3] = enc.eot_token # EOT token
+    with open(filename, "wb") as file:
+        file.write(header.numpy().tobytes())
+        for i in range(n):
+            b = enc.decode_bytes([i])
+            length = len(b)
+            assert length < 256, f"Token length exceeds 255: {length}"
+            file.write(struct.pack("<B", length))  # Write the length as a 1-byte unsigned integer
+            file.write(b)  # Write the actual bytes
+    print(f"wrote {filename}")
+
+def set_seed(seed):
+    random.seed(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    if torch.cuda.is_available():
+        torch.cuda.manual_seed_all(seed)
+    print(f"PRINT: Set seed to {seed}", flush=True) # Print immediately for all ranks
+
+# -----------------------------------------------------------------------------
+# int main
+
+def print0(*args, **kwargs):
+    # modified print that only prints from the master process
+    # if this is not a distributed run, it's just a print
+    if int(os.environ.get("RANK", 0)) == 0:
+        print(*args, **kwargs)
+
+if __name__ == "__main__":
+    import time
+    import argparse
+    import tiktoken
+    print0(f"Running pytorch {torch.version.__version__}")
+
+    # default settings will overfit a tiny batch of data
+    # and save model weights and debug state to disk on the first iteration
+    parser = argparse.ArgumentParser()
+    # file system input / output
+    parser.add_argument("--input_bin", type=str, default="dev/data/tinyshakespeare/tiny_shakespeare_val.bin", help="input .bin to train on")
+    parser.add_argument("--input_val_bin", type=str, default="", help="input .bin to eval validation loss on")
+    parser.add_argument("--output_dir", type=str, default="", help="output directory to which to write logs and checkpoints")
+    parser.add_argument("--model", type=str, default="gpt2", help="gpt2|gpt2-medium|gpt2-large|gpt2-xl|d12|d24|d36|d48")
+    # token layout for each step of the optimization
+    parser.add_argument("--batch_size", type=int, default=4, help="batch size, in units of #batch dimensions")
+    parser.add_argument("--sequence_length", type=int, default=64, help="sequence length")
+    parser.add_argument("--total_batch_size", type=int, default=256, help="total desired batch size, in units of #tokens")
+    # workload (number of steps)
+    parser.add_argument("--num_iterations", type=int, default=10, help="number of iterations to run")
+    parser.add_argument("--inference_only", type=int, default=0, help="only run inference")
+    # optimization
+    parser.add_argument("--adam_lr", type=float, default=1e-4, help="learning rate warmup iterations")
+    parser.add_argument("--warmup_iters", type=int, default=0, help="learning rate warmup iterations")
+    parser.add_argument("--lr_decay_frac", type=float, default=1.0, help="learning rate warmup iterations")
+    parser.add_argument("--weight_decay", type=float, default=0.0, help="weight decay")
+    parser.add_argument("--grad_clip", type=float, default=1.0, help="maximum gradient magnitude")
+    # evaluation
+    parser.add_argument("--val_loss_every", type=int, default=0, help="every how mant steps to evaluate val loss?")
+    parser.add_argument("--val_max_steps", type=int, default=20, help="how many batches of val to average?")
+    parser.add_argument("--sample_every", type=int, default=0, help="how often to sample from the model?")
+    # debugging
+    parser.add_argument("--overfit_single_batch", type=int, default=1, help="overfit just one batch of data")
+    # numerics
+    parser.add_argument("--tensorcores", type=int, default=0, help="use tensorcores")
+    # memory management
+    parser.add_argument("--device", type=str, default="", help="by default we autodetect, or set it here")
+    parser.add_argument("--compile", type=int, default=0, help="torch.compile the model")
+    parser.add_argument("--flash", type=int, default=0, help="use flash attention")
+    parser.add_argument("--dtype", type=str, default="float32", help="float32|float16|bfloat16")
+    parser.add_argument("--zero_stage", type=int, default=0, help="zero redundancy optimizer stage (0/1/2/3)")
+    # Muon optimizer specific arguments
+    parser.add_argument("--optimizer", type=str, default="adam", help="optimizer to use: adam|muon")
+    parser.add_argument("--muon_lr", type=float, default=0.02, help="learning rate for Muon optimizer")
+    parser.add_argument("--muon_momentum", type=float, default=0.95, help="momentum for Muon optimizer")
+    parser.add_argument("--muon_weight_decay", type=float, default=0.00, help="weight decay for Muon optimizer")
+    parser.add_argument("--muon_ns_steps", type=int, default=5, help="number of Newton-Schulz steps for Muon")
+    parser.add_argument("--muon_nesterov", type=bool, default=False, help="use Nesterov momentum for Muon (0/1)")
+    # python -> C bridge
+    parser.add_argument("--write_tensors", type=int, default=1, help="write tensors to disk")
+    parser.add_argument("--seed", type=int, default=42, help="random seed")
+    args = parser.parse_args()
+
+    # args error checking and convenience variables
+    B, T = args.batch_size, args.sequence_length
+    assert 1 <= T <= 1024
+    assert args.dtype in {"float32", "float16", "bfloat16"}
+    assert args.model in {"gpt2", "gpt2-medium", "gpt2-large", "gpt2-xl", "d12", "d24", "d36", "d48"}
+    assert args.optimizer in {"adam", "muon"}
+
+    set_seed(args.seed)
+
+    # set up DDP (distributed data parallel). torchrun sets this env variable
+    ddp = int(os.environ.get('RANK', -1)) != -1 # is this a ddp run?
+    if ddp:
+        # use of DDP atm demands CUDA, we set the device appropriately according to rank
+        assert torch.cuda.is_available(), "for now i think we need CUDA for DDP"
+        init_process_group(backend='nccl')
+        ddp_rank = int(os.environ['RANK'])
+        ddp_local_rank = int(os.environ['LOCAL_RANK'])
+        ddp_world_size = int(os.environ['WORLD_SIZE'])
+        device = f'cuda:{ddp_local_rank}'
+        torch.cuda.set_device(device)
+        master_process = ddp_rank == 0 # this process will do logging, checkpointing etc.
+        seed_offset = 0 # each process gets the exact same seed
+        zero_stage = args.zero_stage
+    else:
+        ddp_rank = 0
+        ddp_local_rank = 0
+        zero_stage = 0
+        ddp_world_size = 1
+        master_process = True
+        seed_offset = 0
+        # select the device
+        if args.device:
+            # provided explicitly by the user
+            device = args.device
+        else:
+            # attempt to autodetect the device
+            device = "cpu"
+            if torch.cuda.is_available():
+                device = "cuda"
+            elif hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
+                device = "mps"
+    print(f"using device: {device}")
+    device_type = 'cuda' if 'cuda' in device else 'cpu'
+
+    # calculate gradient accumulation from the desired total batch size and the current run configuration
+    tokens_per_fwdbwd = B * T * ddp_world_size
+    assert args.total_batch_size % tokens_per_fwdbwd == 0
+    grad_accum_steps = args.total_batch_size // tokens_per_fwdbwd
+    print0(f"total desired batch size: {args.total_batch_size}")
+    print0(f"=> calculated gradient accumulation steps: {grad_accum_steps}")
+
+    # set up a context manager following the desired dtype and device
+    ptdtype = {'float32': torch.float32, 'bfloat16': torch.bfloat16, 'float16': torch.float16}[args.dtype]
+    ctx = torch.amp.autocast(device_type=device_type, dtype=ptdtype) if device_type == "cuda" else nullcontext()
+
+    # rng / reproducibility
+    torch.manual_seed(42)
+    if torch.cuda.is_available():
+        torch.cuda.manual_seed(42)
+
+    # set the torch precision mode to use TensorFloat32 (TF32) for matmuls
+    # docs https://pytorch.org/docs/stable/generated/torch.set_float32_matmul_precision.html
+    if args.tensorcores:
+        torch.set_float32_matmul_precision('high')
+
+    # turn on/off flash attention
+    assert args.flash in {0, 1}
+    FLASH = args.flash
+
+    # init (and write) the tokenizer
+    enc = tiktoken.get_encoding("gpt2")
+    if master_process and args.write_tensors: # tokenizer is technically not tensors but ok
+        write_tokenizer(enc, "gpt2_tokenizer.bin")
+
+    # init the model, either from scratch or from OpenAI pretrained checkpoint
+    if args.model[0] == "d":
+        # from scratch (random weights)
+        model_config = {
+            "d12": GPTConfig(block_size=1024, vocab_size=50257, n_layer=12, n_head=12, n_embd=768),
+            "d24": GPTConfig(block_size=1024, vocab_size=50257, n_layer=24, n_head=16, n_embd=1024),
+            "d36": GPTConfig(block_size=1024, vocab_size=50257, n_layer=36, n_head=20, n_embd=1280),
+            "d48": GPTConfig(block_size=1024, vocab_size=50257, n_layer=48, n_head=25, n_embd=1600),
+        }[args.model]
+        model = GPT(model_config)
+    else:
+        # load the GPT-2 model weights
+        model = GPT.from_pretrained(args.model)
+    model.train()
+    model.to(device)
+    if args.compile:
+        if hasattr(config, "coordinate_descent_tuning"):
+            config.coordinate_descent_tuning = True # suggested by @Chillee
+        print0("compiling the model...")
+        model = torch.compile(model)
+
+    # -------------------------------------------------------------------------
+    # Our own version of a simple DistributedDataLoader
+
+    # load tokens
+    train_loader = DistributedDataLoader(args.input_bin, B, T, ddp_rank, ddp_world_size)
+    val_loader = None
+    if args.input_val_bin:
+        val_loader = DistributedDataLoader(args.input_val_bin, B, T, ddp_rank, ddp_world_size)
+
+    # -------------------------------------------------------------------------
+    # PyTorch -> C bridge: save some weights and state for C to load later as reference
+
+    # do one forward pass to generate ground truth for our C tests
+    if master_process and args.write_tensors and (not args.inference_only):
+        x, y = train_loader.next_batch()
+        x, y = x.to(device), y.to(device)
+        logits, loss = model(x, y)
+        loss.backward()
+        # save model params, in both float32 and bfloat16
+        model_to_size = {"gpt2": "124M", "gpt2-medium": "355M", "gpt2-large": "774M", "gpt2-xl": "1558M"}
+        model_to_size.update({f"d{d}": f"d{d}" for d in [12, 24, 36, 48]})
+        model_size_str = model_to_size[args.model] # e.g. "124M", or "d12"
+        write_model(model, f"gpt2_{model_size_str}.bin", dtype="float32")
+        write_model(model, f"gpt2_{model_size_str}_bf16.bin", dtype="bfloat16")
+        # save x, y, logits, loss, and parameter gradients, for debugging C
+        # always store these in fp32 to have an accurate reference (?)
+        write_state(model, x, y, logits, loss, f"gpt2_{model_size_str}_debug_state.bin")
+        # reset the train_loader for the optimization below
+        train_loader.reset()
+
+    # -------------------------------------------------------------------------
+    # main training loop
+
+    # here we wrap model into DDP container
+    if ddp:
+        model = DDP(model, device_ids=[ddp_local_rank])
+    raw_model = model.module if ddp else model # always contains the "raw" unwrapped model
+
+
+    def configure_adam(model, weight_decay, learning_rate, betas, device_type, zero_stage):
+        # start with all of the candidate parameters
+        param_dict = {pn: p for pn, p in model.named_parameters()}
+        # filter out those that do not require grad
+        param_dict = {pn: p for pn, p in param_dict.items() if p.requires_grad}
+        # create optim groups. Any parameters that is 2D will be weight decayed, otherwise no.
+        # i.e. all weight tensors in matmuls + embeddings decay, all biases and layernorms don't.
+        decay_params = [p for n, p in param_dict.items() if p.dim() >= 2]
+        nodecay_params = [p for n, p in param_dict.items() if p.dim() < 2]
+        optim_groups = [
+            {'params': decay_params, 'weight_decay': weight_decay},
+            {'params': nodecay_params, 'weight_decay': 0.0}
+        ]
+        num_decay_params = sum(p.numel() for p in decay_params)
+        num_nodecay_params = sum(p.numel() for p in nodecay_params)
+        print0(f"num decayed parameter tensors: {len(decay_params)}, with {num_decay_params:,} parameters")
+        print0(f"num non-decayed parameter tensors: {len(nodecay_params)}, with {num_nodecay_params:,} parameters")
+        # Create AdamW optimizer and use the fused version if it is available
+        fused_available = 'fused' in inspect.signature(torch.optim.AdamW).parameters
+        use_fused = fused_available and device_type == 'cuda'
+        print0(f"using fused AdamW: {use_fused}")
+        if zero_stage == 1:
+            print0("using ZeroRedundancyOptimizer")
+            optimizer = ZeroRedundancyOptimizer(**optim_groups[0], optimizer_class=torch.optim.AdamW,
+                                                lr=learning_rate, betas=betas, fused=use_fused)
+            optimizer.add_param_group(optim_groups[1])
+        else:
+            print0("using regular AdamW")
+            optimizer = torch.optim.AdamW(optim_groups, lr=learning_rate, betas=betas, fused=use_fused)
+        return [optimizer]
+
+    def configure_muon(model, weight_decay, adam_lr, muon_lr, momentum, nesterov, ns_steps, device_type, zero_stage, ddp_rank, ddp_world_size):
+        # start with all of the candidate parameters
+        param_dict = {pn: p for pn, p in model.named_parameters()}
+        # filter out those that do not require grad
+        param_dict = {pn: p for pn, p in param_dict.items() if p.requires_grad}
+        
+        # For Muon, we need to separate 2D parameters (which can be orthogonalized) 
+        # from other parameters (which should use standard optimization)
+        muon_params = []  # 2D parameters for Muon
+        other_params = []  # other parameters for AdamW
+
+        muon_name = []
+        other_name = []
+        for n, p in param_dict.items():
+            if "wte.weight" in n :
+                other_params.append(p)
+                other_name.append(n)
+                continue
+
+            if p.dim() >= 2:  # 2D parameters (weight matrices)
+                muon_params.append(p)
+                muon_name.append(n)
+            else:  # 1D parameters (biases, embeddings, etc.)
+                other_params.append(p)
+                other_name.append(n)
+
+        # print("================================================\n")
+        # print(f"Muon parameters: {muon_name}\n")
+        # print(f"Other parameters: {other_name}\n")
+        # print("================================================\n")
+        
+        print0(f"Muon parameters (2D): {len(muon_params)} tensors")
+        print0(f"Other parameters (non-2D): {len(other_params)} tensors")
+        
+        # Create Muon optimizer for 2D parameters
+        muon_optimizer = None
+        if muon_params:
+            muon_optimizer = Muon(
+                params=muon_params,
+                lr=muon_lr,
+                weight_decay=weight_decay,
+                momentum=momentum,
+                nesterov=nesterov,
+                ns_steps=ns_steps,
+                rank=ddp_rank,
+                world_size=ddp_world_size
+            )
+        
+        # Create AdamW optimizer for non-2D parameters
+        adam_optimizer = None
+        if other_params:
+            # create optim groups for AdamW
+            # decay_params = [p for p in other_params if p.dim() >= 2]
+            # nodecay_params = [p for p in other_params if p.dim() < 2]
+            optim_groups = [
+                {'params': other_params, 'weight_decay': weight_decay},
+                # {'params': nodecay_params, 'weight_decay': 0.0}
+            ]
+            
+            # Create AdamW optimizer
+            fused_available = 'fused' in inspect.signature(torch.optim.AdamW).parameters
+            use_fused = fused_available and device_type == 'cuda'
+            print0(f"using fused AdamW for non-Muon params: {use_fused}")
+            
+            if zero_stage == 1:
+                print0("using ZeroRedundancyOptimizer for non-Muon params")
+                adam_optimizer = ZeroRedundancyOptimizer(**optim_groups[0], optimizer_class=torch.optim.AdamW,
+                                                        lr=adam_lr, betas=(0.9, 0.95), fused=use_fused)
+                # adam_optimizer.add_param_group(optim_groups[1])
+            else:
+                print0("using regular AdamW for non-Muon params")
+                adam_optimizer = torch.optim.AdamW(optim_groups, lr=adam_lr, betas=(0.9, 0.95), fused=use_fused)
+        
+        return [muon_optimizer, adam_optimizer]
+
+    # init the optimizer
+    if args.optimizer == "adam":
+        optimizers = configure_adam(model=raw_model, weight_decay=args.weight_decay,
+                                                   learning_rate=args.adam_lr, betas=(0.9, 0.95),
+                                                   device_type=device, zero_stage=zero_stage)
+    elif args.optimizer == "muon":
+        optimizers = configure_muon(
+            model=raw_model, 
+            weight_decay=args.muon_weight_decay,
+            muon_lr=args.muon_lr,
+            adam_lr=args.adam_lr,
+            momentum=args.muon_momentum,
+            nesterov=bool(args.muon_nesterov),
+            ns_steps=args.muon_ns_steps,
+            device_type=device,
+            zero_stage=zero_stage,
+            ddp_rank=ddp_rank,
+            ddp_world_size=ddp_world_size
+        )
+        # We'll use muon_optimizer and adam_optimizer separately
+
+    # learning rate decay scheduler (cosine with warmup)
+    def get_lr(it,base_lr):
+        # if args.optimizer == "adam":
+        #     base_lr = args.adam_lr
+        # else:  # muon
+        #     base_lr = args.muon_lr
+        min_lr = base_lr * args.lr_decay_frac
+        # 1) linear warmup for warmup_iters steps
+        if it < args.warmup_iters:
+            return base_lr * (it+1) / args.warmup_iters
+        # 2) if it > lr_decay_iters, return min learning rate
+        if it > args.num_iterations:
+            return min_lr
+        # 3) in between, use cosine decay down to min learning rate
+        decay_ratio = (it - args.warmup_iters) / (args.num_iterations - args.warmup_iters)
+        assert 0 <= decay_ratio <= 1
+        coeff = 0.5 * (1.0 + math.cos(math.pi * decay_ratio)) # coeff starts at 1 and goes to 0
+        return min_lr + coeff * (base_lr - min_lr)
+
+    def get_wsd_lr(it,base_lr):
+        min_lr = base_lr * args.lr_decay_frac
+        # 1) linear warmup for warmup_iters steps
+        if it < args.warmup_iters:
+            return base_lr * (it+1) / args.warmup_iters
+        else:
+            return base_lr
+
+    # create the logging directory if it does not exist
+    logfile = None
+    file_name  = f"mode_{args.optimizer}_adam_lr_{args.adam_lr}_muon_lr_{args.muon_lr}_seed_{args.seed}.log"
+    if args.output_dir:
+        os.makedirs(args.output_dir, exist_ok=True)
+        logfile = os.path.join(args.output_dir, file_name)
+        # create the log file "main.log" inside it, and wipe it clean
+        with open(logfile, "w") as f:
+            pass
+    if master_process:
+        with open(logfile, "a") as f:
+            f.write(code)
+
+    if device == "cuda":
+        torch.cuda.reset_peak_memory_stats()
+    timings = []
+    norm = -1.0   # dummy value to print in inference-only mode
+    for step in range(args.num_iterations + 1):
+        t0 = time.time()
+        last_step = (step == args.num_iterations)
+
+        # once in a while evaluate the validation dataset
+        if (args.val_loss_every > 0 \
+            and (step % args.val_loss_every == 0 or last_step)) \
+            and (val_loader is not None):
+            model.eval()
+            val_loader.reset()
+            with torch.no_grad():
+                val_loss = 0.0
+                for _ in range(args.val_max_steps):
+                    x, y = val_loader.next_batch()
+                    x, y = x.to(device), y.to(device)
+                    _, loss = model(x, y, return_logits=False)
+                    val_loss += loss.item()
+                val_loss /= args.val_max_steps
+            # log to console and to file
+            print0(f"val loss {val_loss}")
+            if master_process and logfile is not None:
+                with open(logfile, "a") as f:
+                    f.write("step:%d validation loss:%f\n" % (step, val_loss))
+
+        # once in a while perform model inference on the master process
+        if (args.sample_every > 0 \
+            and (step % args.sample_every == 0 or last_step)) \
+            and master_process:
+            model.eval()
+            # before we end, let's also do one round of inference
+            # we'll kick off the generation with "<|endoftext|>", which designates the start of a new sequence
+            start_ids = [enc.eot_token]
+            xg = (torch.tensor(start_ids, dtype=torch.long, device=device)[None, ...])
+            max_new_tokens = 32
+            temperature = 1.0
+            top_k = 40
+            yg = raw_model.generate(xg, max_new_tokens, temperature=temperature, top_k=top_k)
+            print0('---------------')
+            print0(enc.decode(yg[0].tolist()))
+            print0('---------------')
+
+        # bit confusing: we want to make sure to eval and sample on 0th iteration
+        # but also after the very last iteration. so we loop for step <= num_iterations
+        # instead of just < num_iterations (one extra due to <=), only to do
+        # the validation/sampling one last time, and then we break right here as we're done.
+        if last_step:
+            break
+
+        # --------------- TRAINING SECTION BEGIN -----------------
+        model.train()
+        # Zero gradients for the appropriate optimizer(s)
+
+        for optimizer in optimizers:
+            if isinstance(optimizer, ZeroRedundancyOptimizer) or isinstance(optimizer, torch.optim.AdamW):
+                optimizer.zero_grad(set_to_none=True)
+            elif isinstance(optimizer, Muon):
+                optimizer.zero_grad()
+        # if args.optimizer == "adam":
+        #     optimizer.zero_grad(set_to_none=True)
+        # else:  # muon
+        #     if muon_optimizer is not None:
+        #         muon_optimizer.zero_grad()
+        #     if adam_optimizer is not None:
+        #         adam_optimizer.zero_grad(set_to_none=True)
+        # if we are trying to overfit a single batch, we reset the loader here
+        if args.overfit_single_batch:
+            train_loader.reset()
+        # micro-batch loop where we do gradient accumulation to reach desired total batch size
+        lossf = 0.0 # for getting the mean loss (as simple float) over the accumulation steps
+        for micro_step in range(grad_accum_steps):
+            # fetch a batch
+            x, y = train_loader.next_batch()
+            x, y = x.to(device), y.to(device)
+            if ddp:
+                # we want only the last micro-step to sync grads in a DDP model
+                # the official way to do this is with model.no_sync(), but that is a
+                # context manager that bloats the code, so we just toggle this variable
+                model.require_backward_grad_sync = (micro_step == grad_accum_steps - 1)
+            # forward pass
+            with ctx:
+                _, loss = model(x, y, return_logits=False)
+                # we have to scale the loss to account for gradient accumulation,
+                # because the gradients just add on each successive backward().
+                # addition of gradients corresponds to a SUM in the objective, but
+                # instead of a SUM we want MEAN, so we scale the loss here
+                loss = loss / grad_accum_steps
+                lossf += loss.detach() # keep track of the mean loss
+            # backward pass
+            if not args.inference_only:
+                loss.backward()
+        if ddp:
+            dist.all_reduce(lossf, op=dist.ReduceOp.AVG)
+        lossf = lossf.item()
+        norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)
+        # determine and set the learning rate for this iteration
+
+        
+        # Update learning rate and step the appropriate optimizer(s)
+        # if args.optimizer == "adam":
+        #     adam_lr = get_wsd_lr(step,args.adam_lr)
+        #     for param_group in optimizer.param_groups:
+        #         param_group['lr'] = adam_lr
+        #     optimizer.step()
+        # else:  # muon
+        #     if muon_optimizer is not None:
+        #         muon_lr = get_wsd_lr(step,args.muon_lr)
+        #         for param_group in muon_optimizer.param_groups:
+        #             param_group['lr'] = muon_lr
+        #         muon_optimizer.step()
+        #     if adam_optimizer is not None:
+        #         adam_lr = get_wsd_lr(step,args.adam_lr)
+        #         for param_group in adam_optimizer.param_groups:
+        #             param_group['lr'] = adam_lr
+        #         adam_optimizer.step()
+        for optimizer in optimizers:
+            if isinstance(optimizer, ZeroRedundancyOptimizer) or isinstance(optimizer, torch.optim.AdamW):
+                adam_lr = get_wsd_lr(step,args.adam_lr)
+                for param_group in optimizer.param_groups:
+                    param_group['lr'] = adam_lr
+                optimizer.step()
+            elif isinstance(optimizer, Muon):
+                muon_lr = get_wsd_lr(step,args.muon_lr)
+                for param_group in optimizer.param_groups:
+                    param_group['lr'] = muon_lr
+                optimizer.step()
+            else:
+                raise ValueError(f"Unsupported optimizer: {type(optimizer)}")
+        # --------------- TRAINING SECTION END -------------------
+        # everything that follows now is just diagnostics, prints, logging, etc.
+
+        # wait on the CPU for all device work to end so we get accurate per-iteration timings below
+        if device == "mps":
+            torch.mps.synchronize()
+        elif device == "cuda":
+            torch.cuda.synchronize()
+        # time and print
+        t1 = time.time()
+        # the 0th iteration is often an outlier (much slower) => skip logging it
+        tokens_per_second = grad_accum_steps * ddp_world_size * B * T / (t1-t0)
+        print0(f"step {step+1:4d}/{args.num_iterations} | train loss {lossf:.6f} | norm {norm:.4f} | ({(t1-t0)*1000:.2f} ms | {tokens_per_second:.0f} tok/s)")
+        # log to logile
+        if master_process and logfile is not None:
+            with open(logfile, "a") as f:
+                f.write("step:%d train loss:%f\n" % (step, lossf))
+
+        # keep track of smooth timings, last 20 iterations
+        if step > 0 and step > args.num_iterations - 20:
+            timings.append(t1-t0)
+
+    # print the average of the last 20 timings, to get something smooth-ish
+    timings = timings[-20:]
+    print0(f"final {len(timings)} iters avg: {np.mean(timings)*1000:.3f}ms")
+    print0(f"peak memory consumption: {torch.cuda.max_memory_allocated() // 1024 // 1024} MiB")
+
+    # -------------------------------------------------------------------------
+    # clean up nice
+    if ddp:
+        destroy_process_group()
+step:0 validation loss:11.021495
+step:0 train loss:11.015424
+step:1 train loss:11.017307
+step:2 train loss:11.002539
+step:3 train loss:10.987658
+step:4 train loss:10.966583
+step:5 train loss:10.940186
+step:6 train loss:10.906839
+step:7 train loss:10.868830
+step:8 train loss:10.830691
+step:9 train loss:10.780419
+step:10 train loss:10.728622
+step:11 train loss:10.679697
+step:12 train loss:10.610229
+step:13 train loss:10.545015
+step:14 train loss:10.478123
+step:15 train loss:10.409031
+step:16 train loss:10.336840
+step:17 train loss:10.266420
+step:18 train loss:10.193359
+step:19 train loss:10.110159
+step:20 train loss:10.029213
+step:21 train loss:9.952513
+step:22 train loss:9.840858
+step:23 train loss:9.776922
+step:24 train loss:9.665634
+step:25 train loss:9.595795
+step:26 train loss:9.498317
+step:27 train loss:9.395107
+step:28 train loss:9.328226
+step:29 train loss:9.237043
+step:30 train loss:9.148466
+step:31 train loss:9.030872
+step:32 train loss:8.935430
+step:33 train loss:8.850748
+step:34 train loss:8.789619
+step:35 train loss:8.669458
+step:36 train loss:8.587969
+step:37 train loss:8.478680
+step:38 train loss:8.428891
+step:39 train loss:8.318810
+step:40 train loss:8.245317
+step:41 train loss:8.136267
+step:42 train loss:8.097561
+step:43 train loss:7.968538
+step:44 train loss:7.898306
+step:45 train loss:7.837488
+step:46 train loss:7.776301
+step:47 train loss:7.719494
+step:48 train loss:7.618882
+step:49 train loss:7.559425
+step:50 train loss:7.459339
+step:51 train loss:7.429264
+step:52 train loss:7.396437
+step:53 train loss:7.347573
+step:54 train loss:7.306512
+step:55 train loss:7.238846
+step:56 train loss:7.180341
+step:57 train loss:7.193172
+step:58 train loss:7.104615
+step:59 train loss:7.112510
+step:60 train loss:7.082651
+step:61 train loss:7.039591
+step:62 train loss:7.012208
+step:63 train loss:7.050370
+step:64 train loss:6.934842
+step:65 train loss:6.955455
+step:66 train loss:6.945438
+step:67 train loss:6.957443
+step:68 train loss:6.899389
+step:69 train loss:6.870329
+step:70 train loss:6.830938
+step:71 train loss:6.801010
+step:72 train loss:6.820164
+step:73 train loss:6.762726
+step:74 train loss:6.779589
+step:75 train loss:6.713727
+step:76 train loss:6.798114
+step:77 train loss:6.727730
+step:78 train loss:6.478737
+step:79 train loss:6.641900
+step:80 train loss:6.612185
+step:81 train loss:6.701096
+step:82 train loss:6.647059
+step:83 train loss:6.602648
+step:84 train loss:6.559029
+step:85 train loss:6.535501
+step:86 train loss:6.526028
+step:87 train loss:6.499008
+step:88 train loss:6.496197
+step:89 train loss:6.448274
+step:90 train loss:6.492400
+step:91 train loss:6.491443
+step:92 train loss:6.499742
+step:93 train loss:6.450020
+step:94 train loss:6.406896
+step:95 train loss:6.348999
+step:96 train loss:6.452277
+step:97 train loss:6.394535
+step:98 train loss:6.378076
+step:99 train loss:6.346870
+step:100 train loss:6.357969
+step:101 train loss:6.291591
+step:102 train loss:6.310251
+step:103 train loss:6.300294
+step:104 train loss:6.319246
+step:105 train loss:6.374966
+step:106 train loss:6.326587
+step:107 train loss:6.271721
+step:108 train loss:6.299007
+step:109 train loss:6.331908
+step:110 train loss:6.257711
+step:111 train loss:6.278391
+step:112 train loss:6.272741
+step:113 train loss:6.226389
+step:114 train loss:6.275346
+step:115 train loss:6.246730
+step:116 train loss:6.220730
+step:117 train loss:6.168935
+step:118 train loss:6.216222
+step:119 train loss:6.171053
+step:120 train loss:6.192867
+step:121 train loss:6.109637
+step:122 train loss:6.204206
+step:123 train loss:6.134442
+step:124 train loss:6.120174
+step:125 train loss:6.095740
+step:126 train loss:6.198706
+step:127 train loss:6.110419
+step:128 train loss:6.164063
+step:129 train loss:6.131337
+step:130 train loss:6.151025
+step:131 train loss:6.109653
+step:132 train loss:6.055182
+step:133 train loss:6.096249
+step:134 train loss:6.085812
+step:135 train loss:5.997619
+step:136 train loss:6.041600
+step:137 train loss:6.047670
+step:138 train loss:5.993762
+step:139 train loss:6.067573
+step:140 train loss:5.986075
+step:141 train loss:6.069380
+step:142 train loss:6.028786
+step:143 train loss:6.039732
+step:144 train loss:6.015032
+step:145 train loss:5.949231
+step:146 train loss:5.965765
+step:147 train loss:6.013845
+step:148 train loss:6.027832
+step:149 train loss:5.982743
+step:150 train loss:5.980625
+step:151 train loss:5.898334
+step:152 train loss:5.937162
+step:153 train loss:5.922023
+step:154 train loss:5.985418
+step:155 train loss:5.976247
+step:156 train loss:5.991820
+step:157 train loss:5.915868
+step:158 train loss:5.897432
+step:159 train loss:5.923434
+step:160 train loss:5.915922
+step:161 train loss:5.913770
+step:162 train loss:5.874032
+step:163 train loss:5.897184
+step:164 train loss:5.886250
+step:165 train loss:5.913967
+step:166 train loss:5.863026
+step:167 train loss:5.870093
+step:168 train loss:5.851177
+step:169 train loss:5.800628
+step:170 train loss:5.771802
+step:171 train loss:5.883171
+step:172 train loss:5.820835
+step:173 train loss:5.879887
+step:174 train loss:5.872364
+step:175 train loss:5.832941
+step:176 train loss:5.794168
+step:177 train loss:5.826818
+step:178 train loss:5.835801
+step:179 train loss:5.796124
+step:180 train loss:5.763860
+step:181 train loss:5.813160
+step:182 train loss:5.748355
+step:183 train loss:5.827563
+step:184 train loss:5.789713
+step:185 train loss:5.737475
+step:186 train loss:5.855652
+step:187 train loss:5.796614
+step:188 train loss:5.654269
+step:189 train loss:5.776381
+step:190 train loss:5.769153
+step:191 train loss:5.715700
+step:192 train loss:5.652488
+step:193 train loss:5.785297
+step:194 train loss:5.791453
+step:195 train loss:5.777102
+step:196 train loss:5.768545
+step:197 train loss:5.741137
+step:198 train loss:5.708144
+step:199 train loss:5.768876
+step:200 train loss:5.842128
+step:201 train loss:5.739259
+step:202 train loss:5.757513
+step:203 train loss:5.715420
+step:204 train loss:5.728333
+step:205 train loss:5.624386
+step:206 train loss:5.704189
+step:207 train loss:5.722406
+step:208 train loss:5.652603
+step:209 train loss:5.644573
+step:210 train loss:5.654663
+step:211 train loss:5.692886
+step:212 train loss:5.680385
+step:213 train loss:5.665684
+step:214 train loss:5.680418
+step:215 train loss:5.669912
+step:216 train loss:5.632862
+step:217 train loss:5.669870
+step:218 train loss:5.620474
+step:219 train loss:5.596796
+step:220 train loss:5.622748
+step:221 train loss:5.589097
+step:222 train loss:5.628883
+step:223 train loss:5.647436
+step:224 train loss:5.648049
+step:225 train loss:5.576045
+step:226 train loss:5.574128
+step:227 train loss:5.632720
+step:228 train loss:5.597082
+step:229 train loss:5.663924
+step:230 train loss:5.533484
+step:231 train loss:5.617034
+step:232 train loss:5.585238
+step:233 train loss:5.544968
+step:234 train loss:5.581217
+step:235 train loss:5.609170
+step:236 train loss:5.565095
+step:237 train loss:5.648161
+step:238 train loss:5.600225
+step:239 train loss:5.527732
+step:240 train loss:5.600331
+step:241 train loss:5.643022
+step:242 train loss:5.602471
+step:243 train loss:5.529858
+step:244 train loss:5.546822
+step:245 train loss:5.518763
+step:246 train loss:5.504246
+step:247 train loss:5.522960
+step:248 train loss:5.482553
+step:249 train loss:5.532679
+step:250 validation loss:5.511507
+step:250 train loss:5.498803
+step:251 train loss:5.538063
+step:252 train loss:5.491353
+step:253 train loss:5.489568
+step:254 train loss:5.472230
+step:255 train loss:5.495923
+step:256 train loss:5.486245
+step:257 train loss:5.544394
+step:258 train loss:5.458185
+step:259 train loss:5.473286
+step:260 train loss:5.431251
+step:261 train loss:5.440634
+step:262 train loss:5.483522
+step:263 train loss:5.474167
+step:264 train loss:5.425406
+step:265 train loss:5.437705
+step:266 train loss:5.427199
+step:267 train loss:5.449633
+step:268 train loss:5.388786
+step:269 train loss:5.419320
+step:270 train loss:5.446882
+step:271 train loss:5.431798
+step:272 train loss:5.388842
+step:273 train loss:5.460462
+step:274 train loss:5.341482
+step:275 train loss:5.435415
+step:276 train loss:5.389358
+step:277 train loss:5.367981
+step:278 train loss:5.352919
+step:279 train loss:5.349316
+step:280 train loss:5.398752
+step:281 train loss:5.466868
+step:282 train loss:5.327730
+step:283 train loss:5.363546
+step:284 train loss:5.356159
+step:285 train loss:5.396748
+step:286 train loss:5.340771
+step:287 train loss:5.342405
+step:288 train loss:5.293667
+step:289 train loss:5.347423
+step:290 train loss:5.369082
+step:291 train loss:5.306618
+step:292 train loss:5.353625
+step:293 train loss:5.305127
+step:294 train loss:5.354642
+step:295 train loss:5.335859
+step:296 train loss:5.332960
+step:297 train loss:5.347226
+step:298 train loss:5.256810
+step:299 train loss:5.309451
+step:300 train loss:5.251300
+step:301 train loss:5.284420
+step:302 train loss:5.241238
+step:303 train loss:5.261605
+step:304 train loss:5.287779
+step:305 train loss:5.238199
+step:306 train loss:5.228479
+step:307 train loss:5.241763
+step:308 train loss:5.176595
+step:309 train loss:5.302693
+step:310 train loss:5.266294
+step:311 train loss:5.234930
+step:312 train loss:5.223832
+step:313 train loss:5.230992
+step:314 train loss:5.226810
+step:315 train loss:5.161324
+step:316 train loss:5.159932
+step:317 train loss:5.154632
+step:318 train loss:5.159330
+step:319 train loss:5.212348
+step:320 train loss:5.117433
+step:321 train loss:5.190118
+step:322 train loss:5.167397
+step:323 train loss:5.229135
+step:324 train loss:5.183520
+step:325 train loss:5.188560
+step:326 train loss:5.175596
+step:327 train loss:5.191351
+step:328 train loss:5.148170
+step:329 train loss:5.143776
+step:330 train loss:5.089268
+step:331 train loss:5.120739
+step:332 train loss:5.090946
+step:333 train loss:5.030110
+step:334 train loss:5.123396
+step:335 train loss:5.158096
+step:336 train loss:5.362187
+step:337 train loss:5.151428
+step:338 train loss:5.079495
+step:339 train loss:5.043005
+step:340 train loss:5.047062
+step:341 train loss:5.023942
+step:342 train loss:5.099456
+step:343 train loss:5.055502
+step:344 train loss:5.050872
+step:345 train loss:5.001206
+step:346 train loss:5.041559
+step:347 train loss:5.007703
+step:348 train loss:4.979145
+step:349 train loss:4.928859
+step:350 train loss:4.961483
+step:351 train loss:5.025234
+step:352 train loss:4.968834
+step:353 train loss:5.004018
+step:354 train loss:4.957245
+step:355 train loss:4.986541
+step:356 train loss:4.946774
+step:357 train loss:5.019356
+step:358 train loss:5.062447
+step:359 train loss:4.902654
+step:360 train loss:5.010986
+step:361 train loss:4.997455
+step:362 train loss:4.967745
+step:363 train loss:4.960323
+step:364 train loss:4.971955
+step:365 train loss:4.975463
+step:366 train loss:4.911398
+step:367 train loss:4.985394
+step:368 train loss:4.909604
+step:369 train loss:4.918077
+step:370 train loss:4.951027
+step:371 train loss:4.876147
+step:372 train loss:4.967766
+step:373 train loss:4.889700
+step:374 train loss:4.882738
+step:375 train loss:4.939146
+step:376 train loss:4.893357
+step:377 train loss:4.808064
+step:378 train loss:4.873933
+step:379 train loss:4.885363
+step:380 train loss:4.817454
+step:381 train loss:4.924068
+step:382 train loss:4.851533
+step:383 train loss:4.811456
+step:384 train loss:4.841870
+step:385 train loss:4.816648
+step:386 train loss:4.846949
+step:387 train loss:4.890012
+step:388 train loss:4.813462
+step:389 train loss:4.799527
+step:390 train loss:4.801789
+step:391 train loss:4.795315
+step:392 train loss:4.822580
+step:393 train loss:4.801601
+step:394 train loss:4.834167
+step:395 train loss:4.725907
+step:396 train loss:4.721747
+step:397 train loss:4.796023
+step:398 train loss:4.779165
+step:399 train loss:4.740551
+step:400 train loss:4.768107
+step:401 train loss:4.739222
+step:402 train loss:4.754579
+step:403 train loss:4.729427
+step:404 train loss:4.716895
+step:405 train loss:4.711387
+step:406 train loss:4.771140
+step:407 train loss:4.796683
+step:408 train loss:4.749212
+step:409 train loss:4.714596
+step:410 train loss:4.675590
+step:411 train loss:4.663490
+step:412 train loss:4.755675
+step:413 train loss:4.669036
+step:414 train loss:4.723219
+step:415 train loss:4.714604
+step:416 train loss:4.682702
+step:417 train loss:4.707096
+step:418 train loss:4.663315
+step:419 train loss:4.652900
+step:420 train loss:4.594244
+step:421 train loss:4.630994
+step:422 train loss:4.596540
+step:423 train loss:4.643395
+step:424 train loss:4.612567
+step:425 train loss:4.729231
+step:426 train loss:4.604239
+step:427 train loss:4.610860
+step:428 train loss:4.622014
+step:429 train loss:4.561036
+step:430 train loss:4.606919
+step:431 train loss:4.611240
+step:432 train loss:4.639371
+step:433 train loss:4.631077
+step:434 train loss:4.587337
+step:435 train loss:4.631926
+step:436 train loss:4.655809
+step:437 train loss:4.597250
+step:438 train loss:4.615482
+step:439 train loss:4.551104
+step:440 train loss:4.586331
+step:441 train loss:4.508221
+step:442 train loss:4.525724
+step:443 train loss:4.554875
+step:444 train loss:4.617531
+step:445 train loss:4.574759
+step:446 train loss:4.581038
+step:447 train loss:4.515603
+step:448 train loss:4.636861
+step:449 train loss:4.541403
+step:450 train loss:4.566509
+step:451 train loss:4.627400
+step:452 train loss:4.595576
+step:453 train loss:4.602299
+step:454 train loss:4.476817
+step:455 train loss:4.530272
+step:456 train loss:4.534359
+step:457 train loss:4.504605
+step:458 train loss:4.548941
+step:459 train loss:4.527772
+step:460 train loss:4.586617
+step:461 train loss:4.522080
+step:462 train loss:4.464002
+step:463 train loss:4.506139
+step:464 train loss:4.526798
+step:465 train loss:4.504621
+step:466 train loss:4.527721
+step:467 train loss:4.478045
+step:468 train loss:4.544846
+step:469 train loss:4.497848
+step:470 train loss:4.449895
+step:471 train loss:4.544323
+step:472 train loss:4.456493
+step:473 train loss:4.521075
+step:474 train loss:4.481557
+step:475 train loss:4.529571
+step:476 train loss:4.520624
+step:477 train loss:4.432263
+step:478 train loss:4.474338
+step:479 train loss:4.440193
+step:480 train loss:4.515778
+step:481 train loss:4.511002
+step:482 train loss:4.384820
+step:483 train loss:4.499825
+step:484 train loss:4.446632
+step:485 train loss:4.401309
+step:486 train loss:4.444818
+step:487 train loss:4.450336
+step:488 train loss:4.431090
+step:489 train loss:4.454013
+step:490 train loss:4.396312
+step:491 train loss:4.452537
+step:492 train loss:4.440995
+step:493 train loss:4.416666
+step:494 train loss:4.448147
+step:495 train loss:4.412883
+step:496 train loss:4.410392
+step:497 train loss:4.374882
+step:498 train loss:4.450910
+step:499 train loss:4.469098
+step:500 validation loss:4.393310
+step:500 train loss:4.427353
+step:501 train loss:4.419229
+step:502 train loss:4.460567
+step:503 train loss:4.383344
+step:504 train loss:4.449477
+step:505 train loss:4.409599
+step:506 train loss:4.376177
+step:507 train loss:4.421039
+step:508 train loss:4.417836
+step:509 train loss:4.451217
+step:510 train loss:4.332636
+step:511 train loss:4.376929
+step:512 train loss:4.378193
+step:513 train loss:4.394472
+step:514 train loss:4.481476
+step:515 train loss:4.366212
+step:516 train loss:4.480642
+step:517 train loss:4.395843
+step:518 train loss:4.357895
+step:519 train loss:4.470166
+step:520 train loss:4.342049
+step:521 train loss:4.376186
+step:522 train loss:4.410932
+step:523 train loss:4.422513
+step:524 train loss:4.330533
+step:525 train loss:4.334688
+step:526 train loss:4.413133
+step:527 train loss:4.349930
+step:528 train loss:4.371693
+step:529 train loss:4.403624
+step:530 train loss:4.371160
+step:531 train loss:4.368406
+step:532 train loss:4.355752
+step:533 train loss:4.326072
+step:534 train loss:4.375371
+step:535 train loss:4.398804
+step:536 train loss:4.424671
+step:537 train loss:4.314857
+step:538 train loss:4.279737
+step:539 train loss:4.421515
+step:540 train loss:4.444015
+step:541 train loss:4.340447
+step:542 train loss:4.317795
+step:543 train loss:4.343908
+step:544 train loss:4.354474
+step:545 train loss:4.340141
+step:546 train loss:4.325336
+step:547 train loss:4.360598
+step:548 train loss:4.240135
+step:549 train loss:4.352332
+step:550 train loss:4.289812
+step:551 train loss:4.335394
+step:552 train loss:4.442968
+step:553 train loss:4.346203
+step:554 train loss:4.334588
+step:555 train loss:4.380287
+step:556 train loss:4.314001
+step:557 train loss:4.288407
+step:558 train loss:4.281936
+step:559 train loss:4.330935
+step:560 train loss:4.406416
+step:561 train loss:4.281101
+step:562 train loss:4.282530
+step:563 train loss:4.315229
+step:564 train loss:4.325429
+step:565 train loss:4.290172
+step:566 train loss:4.366910
+step:567 train loss:4.302389
+step:568 train loss:4.393555
+step:569 train loss:4.323826
+step:570 train loss:4.268733
+step:571 train loss:4.291211
+step:572 train loss:4.250766
+step:573 train loss:4.324527
+step:574 train loss:4.399329
+step:575 train loss:4.265396
+step:576 train loss:4.325948
+step:577 train loss:4.306362
+step:578 train loss:4.303312
+step:579 train loss:4.357180
+step:580 train loss:4.259871
+step:581 train loss:4.335917
+step:582 train loss:4.352087
+step:583 train loss:4.307730
+step:584 train loss:4.294944
+step:585 train loss:4.297154
+step:586 train loss:4.271719
+step:587 train loss:4.404578
+step:588 train loss:4.254440
+step:589 train loss:4.364561
+step:590 train loss:4.338441
+step:591 train loss:4.238070
+step:592 train loss:4.283438
+step:593 train loss:4.271563
+step:594 train loss:4.231655
+step:595 train loss:4.390388
+step:596 train loss:4.253740
+step:597 train loss:4.278986
+step:598 train loss:4.294640
+step:599 train loss:4.274227
+step:600 train loss:4.262802
+step:601 train loss:4.259683
+step:602 train loss:4.273860
+step:603 train loss:4.292183
+step:604 train loss:4.274671
+step:605 train loss:4.314252
+step:606 train loss:4.258016
+step:607 train loss:4.297611
+step:608 train loss:4.240560
+step:609 train loss:4.239917
+step:610 train loss:4.290593
+step:611 train loss:4.219511
+step:612 train loss:4.252311
+step:613 train loss:4.215436
+step:614 train loss:4.234926
+step:615 train loss:4.276649
+step:616 train loss:4.224585
+step:617 train loss:4.245788
+step:618 train loss:4.254639
+step:619 train loss:4.264497
+step:620 train loss:4.289402
+step:621 train loss:4.264825
+step:622 train loss:4.282917
+step:623 train loss:4.289814
+step:624 train loss:4.254495
+step:625 train loss:4.284570
+step:626 train loss:4.264802
+step:627 train loss:4.245918
+step:628 train loss:4.265225
+step:629 train loss:4.172097
+step:630 train loss:4.243270
+step:631 train loss:4.233099
+step:632 train loss:4.213527
+step:633 train loss:4.221115
+step:634 train loss:4.263871
+step:635 train loss:4.218144
+step:636 train loss:4.241845
+step:637 train loss:4.159383
+step:638 train loss:4.167646
+step:639 train loss:4.269362
+step:640 train loss:4.204619
+step:641 train loss:4.218515
+step:642 train loss:4.309052
+step:643 train loss:4.177016
+step:644 train loss:4.250973
+step:645 train loss:4.225832
+step:646 train loss:4.222608
+step:647 train loss:4.244049
+step:648 train loss:4.328261
+step:649 train loss:4.239064
+step:650 train loss:4.219040
+step:651 train loss:4.215826
+step:652 train loss:4.157939
+step:653 train loss:4.194305
+step:654 train loss:4.164826
+step:655 train loss:4.268385
+step:656 train loss:4.198242
+step:657 train loss:4.243348
+step:658 train loss:4.195463
+step:659 train loss:4.255297
+step:660 train loss:4.254205
+step:661 train loss:4.278774
+step:662 train loss:4.254154
+step:663 train loss:4.259056
+step:664 train loss:4.172908
+step:665 train loss:4.156044
+step:666 train loss:4.225113
+step:667 train loss:4.228320
+step:668 train loss:4.238627
+step:669 train loss:4.209437
+step:670 train loss:4.223588
+step:671 train loss:4.203549
+step:672 train loss:4.191574
+step:673 train loss:4.290041
+step:674 train loss:4.182269
+step:675 train loss:4.184412
+step:676 train loss:4.221931
+step:677 train loss:4.190480
+step:678 train loss:4.166787
+step:679 train loss:4.204130
+step:680 train loss:4.188313
+step:681 train loss:4.213797
+step:682 train loss:4.139986
+step:683 train loss:4.218596
+step:684 train loss:4.253249
+step:685 train loss:4.201429
+step:686 train loss:4.238118
+step:687 train loss:4.256205
+step:688 train loss:4.119943
+step:689 train loss:4.175318
+step:690 train loss:4.191999
+step:691 train loss:4.175777
+step:692 train loss:4.238293
+step:693 train loss:4.171908
+step:694 train loss:4.187183
+step:695 train loss:4.166661
+step:696 train loss:4.154706
+step:697 train loss:4.201894
+step:698 train loss:4.149628
+step:699 train loss:4.198371
+step:700 train loss:4.188437
+step:701 train loss:4.142958
+step:702 train loss:4.162082
+step:703 train loss:4.147178
+step:704 train loss:4.101230
+step:705 train loss:4.171474
+step:706 train loss:4.069380
+step:707 train loss:4.131087
+step:708 train loss:4.188102
+step:709 train loss:4.180864
+step:710 train loss:4.155499
+step:711 train loss:4.163764
+step:712 train loss:4.165518
+step:713 train loss:4.112861
+step:714 train loss:4.185285
+step:715 train loss:4.065869
+step:716 train loss:4.252131
+step:717 train loss:4.157557
+step:718 train loss:4.181412
+step:719 train loss:4.194964
+step:720 train loss:4.189048
+step:721 train loss:4.131832
+step:722 train loss:4.205163
+step:723 train loss:4.182544
+step:724 train loss:4.182381
+step:725 train loss:4.131506
+step:726 train loss:4.115680
+step:727 train loss:4.171064
+step:728 train loss:4.143806
+step:729 train loss:4.136785
+step:730 train loss:4.193211
+step:731 train loss:4.197047
+step:732 train loss:4.182890
+step:733 train loss:4.206203
+step:734 train loss:4.135615
+step:735 train loss:4.237441
+step:736 train loss:4.178821
+step:737 train loss:4.142222
+step:738 train loss:4.187204
+step:739 train loss:4.158437
+step:740 train loss:4.232876
+step:741 train loss:4.187460
+step:742 train loss:4.149143
+step:743 train loss:4.113138
+step:744 train loss:4.180312
+step:745 train loss:4.057991
+step:746 train loss:4.116617
+step:747 train loss:4.158803
+step:748 train loss:4.107720
+step:749 train loss:4.159406
+step:750 validation loss:4.100407
+step:750 train loss:4.090472
+step:751 train loss:4.137776
+step:752 train loss:4.098166
+step:753 train loss:4.144568
+step:754 train loss:4.122446
+step:755 train loss:4.223899
+step:756 train loss:4.137798
+step:757 train loss:4.247702
+step:758 train loss:4.125173
+step:759 train loss:4.132003
+step:760 train loss:4.103979
+step:761 train loss:4.132847
+step:762 train loss:4.098803
+step:763 train loss:4.120800
+step:764 train loss:4.088158
+step:765 train loss:4.106368
+step:766 train loss:4.153855
+step:767 train loss:4.285463
+step:768 train loss:4.153776
+step:769 train loss:4.165555
+step:770 train loss:4.225049
+step:771 train loss:4.142545
+step:772 train loss:4.196192
+step:773 train loss:4.079477
+step:774 train loss:4.110145
+step:775 train loss:4.175649
+step:776 train loss:4.089301
+step:777 train loss:4.093771
+step:778 train loss:4.095689
+step:779 train loss:4.078810
+step:780 train loss:4.130891
+step:781 train loss:4.091107
+step:782 train loss:4.083716
+step:783 train loss:4.102267
+step:784 train loss:4.083166
+step:785 train loss:4.085423
+step:786 train loss:4.107359
+step:787 train loss:4.025120
+step:788 train loss:4.133759
+step:789 train loss:4.093121
+step:790 train loss:4.113019
+step:791 train loss:4.109143
+step:792 train loss:4.219778
+step:793 train loss:4.141148
+step:794 train loss:4.080012
+step:795 train loss:4.086285
+step:796 train loss:4.393638
+step:797 train loss:4.136972
+step:798 train loss:4.045201
+step:799 train loss:4.122879
+step:800 train loss:4.186136
+step:801 train loss:4.081709
+step:802 train loss:4.248178
+step:803 train loss:4.093216
+step:804 train loss:4.080031
+step:805 train loss:4.101005
+step:806 train loss:4.097311
+step:807 train loss:4.046363
+step:808 train loss:4.122826
+step:809 train loss:4.090263
+step:810 train loss:4.093955
+step:811 train loss:4.084640
+step:812 train loss:4.087770
+step:813 train loss:4.136417
+step:814 train loss:4.244318
+step:815 train loss:4.106221
+step:816 train loss:4.065707
+step:817 train loss:4.123710
+step:818 train loss:4.080160
+step:819 train loss:4.049644
+step:820 train loss:4.110887
+step:821 train loss:4.022964
+step:822 train loss:4.042496
+step:823 train loss:4.093456
+step:824 train loss:4.018581
+step:825 train loss:4.014795
+step:826 train loss:4.066572
+step:827 train loss:3.991488
+step:828 train loss:4.074113
+step:829 train loss:4.050081
+step:830 train loss:4.058598
+step:831 train loss:4.122734
+step:832 train loss:4.174010
+step:833 train loss:4.095309
+step:834 train loss:4.073218
+step:835 train loss:4.066260
+step:836 train loss:4.013987
+step:837 train loss:4.084777
+step:838 train loss:4.019686
+step:839 train loss:4.068089
+step:840 train loss:4.087542
+step:841 train loss:4.074968
+step:842 train loss:4.050657
+step:843 train loss:4.036591
+step:844 train loss:4.081662
+step:845 train loss:3.992584
+step:846 train loss:4.088906
+step:847 train loss:4.124359
+step:848 train loss:4.025726
+step:849 train loss:4.074891
+step:850 train loss:4.067529
+step:851 train loss:4.111156
+step:852 train loss:4.090915
+step:853 train loss:4.006748
+step:854 train loss:4.051981
+step:855 train loss:4.070954
+step:856 train loss:4.019452
+step:857 train loss:4.072253
+step:858 train loss:4.058503
+step:859 train loss:3.986559
+step:860 train loss:4.061744
+step:861 train loss:4.078319
+step:862 train loss:3.993880
+step:863 train loss:4.009288
+step:864 train loss:4.062699
+step:865 train loss:4.042477
+step:866 train loss:4.030247
+step:867 train loss:4.196805
+step:868 train loss:4.065006
+step:869 train loss:4.016693
+step:870 train loss:3.980330
+step:871 train loss:4.035506
+step:872 train loss:4.022592
+step:873 train loss:4.056123
+step:874 train loss:4.075579
+step:875 train loss:3.917429
+step:876 train loss:4.059820
+step:877 train loss:3.982265
+step:878 train loss:4.126727
+step:879 train loss:3.978958
+step:880 train loss:4.082157
+step:881 train loss:4.046810
+step:882 train loss:4.003259
+step:883 train loss:4.042023
+step:884 train loss:4.056682
+step:885 train loss:4.036390
+step:886 train loss:3.984952
+step:887 train loss:4.040345
+step:888 train loss:4.147490
+step:889 train loss:4.034448
+step:890 train loss:4.012470
+step:891 train loss:3.950345
+step:892 train loss:3.988950
+step:893 train loss:4.064858
+step:894 train loss:3.980763
+step:895 train loss:3.988190
+step:896 train loss:4.045144
+step:897 train loss:4.016442
+step:898 train loss:4.026119
+step:899 train loss:3.998734
+step:900 train loss:4.088009
+step:901 train loss:3.992893
+step:902 train loss:3.983998
+step:903 train loss:4.193089
+step:904 train loss:4.066815
+step:905 train loss:4.037220
+step:906 train loss:4.030595
+step:907 train loss:4.029167
+step:908 train loss:4.049891
+step:909 train loss:3.996425
+step:910 train loss:4.114166
+step:911 train loss:4.077013
+step:912 train loss:3.972678
+step:913 train loss:4.009423
+step:914 train loss:4.004431
+step:915 train loss:4.020821
+step:916 train loss:4.045393
+step:917 train loss:4.088541
+step:918 train loss:4.060804
+step:919 train loss:4.181206
+step:920 train loss:3.948839
+step:921 train loss:4.056776
+step:922 train loss:3.960792
+step:923 train loss:4.007132
+step:924 train loss:3.979956
+step:925 train loss:3.951149
+step:926 train loss:4.077478
+step:927 train loss:3.944209
+step:928 train loss:4.035848
+step:929 train loss:4.028355
+step:930 train loss:4.045132
+step:931 train loss:4.035497
+step:932 train loss:4.002022
+step:933 train loss:4.088483
+step:934 train loss:3.983762
+step:935 train loss:4.110620
+step:936 train loss:4.021106
+step:937 train loss:4.022020
+step:938 train loss:3.990609
+step:939 train loss:3.925499
+step:940 train loss:3.990597
+step:941 train loss:3.970042
+step:942 train loss:4.033924
+step:943 train loss:3.955588
+step:944 train loss:4.062877
+step:945 train loss:3.994523
+step:946 train loss:4.039502
+step:947 train loss:4.139359
+step:948 train loss:4.028047
+step:949 train loss:3.975475
+step:950 train loss:3.971011
+step:951 train loss:3.969028
+step:952 train loss:4.070521
+step:953 train loss:3.983914
+step:954 train loss:4.000311
+step:955 train loss:3.981298
+step:956 train loss:3.982191
+step:957 train loss:4.010962
+step:958 train loss:4.013579
+step:959 train loss:4.050619
+step:960 train loss:4.032233
+step:961 train loss:4.047357
+step:962 train loss:3.964910
+step:963 train loss:3.960313
+step:964 train loss:4.004961
+step:965 train loss:3.951746
+step:966 train loss:3.951941
+step:967 train loss:3.974328
+step:968 train loss:4.031517
+step:969 train loss:3.966717
+step:970 train loss:4.009393
+step:971 train loss:3.947788
+step:972 train loss:3.961646
+step:973 train loss:3.967481
+step:974 train loss:3.948026
+step:975 train loss:4.070353
+step:976 train loss:3.969415
+step:977 train loss:4.032722
+step:978 train loss:3.977338
+step:979 train loss:3.955246
+step:980 train loss:3.947230
+step:981 train loss:3.984202
+step:982 train loss:3.978875
+step:983 train loss:3.961284
+step:984 train loss:3.997459
+step:985 train loss:3.978372
+step:986 train loss:4.008345
+step:987 train loss:4.056268
+step:988 train loss:3.968272
+step:989 train loss:3.978174
+step:990 train loss:3.935039
+step:991 train loss:3.942272
+step:992 train loss:3.952811
+step:993 train loss:3.970752
+step:994 train loss:3.960445
+step:995 train loss:3.942155
+step:996 train loss:3.949069
+step:997 train loss:3.968068
+step:998 train loss:3.945670
+step:999 train loss:3.963030
+step:1000 validation loss:3.922949
+step:1000 train loss:4.002028
+step:1001 train loss:3.985914
+step:1002 train loss:3.988024
+step:1003 train loss:3.934859
+step:1004 train loss:3.954869
+step:1005 train loss:3.976978
+step:1006 train loss:4.014203
+step:1007 train loss:3.949594
+step:1008 train loss:3.934643
+step:1009 train loss:3.986087
+step:1010 train loss:4.022973
+step:1011 train loss:4.005935
+step:1012 train loss:3.942850
+step:1013 train loss:3.974207
+step:1014 train loss:3.870000
+step:1015 train loss:3.965679
+step:1016 train loss:3.984982
+step:1017 train loss:3.924944
+step:1018 train loss:3.967738
+step:1019 train loss:3.961305
+step:1020 train loss:3.955120
+step:1021 train loss:3.986247
+step:1022 train loss:3.959225
+step:1023 train loss:3.942993
+step:1024 train loss:3.961884
+step:1025 train loss:4.006050
+step:1026 train loss:3.929019
+step:1027 train loss:3.975284
+step:1028 train loss:3.942400
+step:1029 train loss:3.952156
+step:1030 train loss:3.939455
+step:1031 train loss:4.029231
+step:1032 train loss:3.936928
+step:1033 train loss:3.930429
+step:1034 train loss:3.980088
+step:1035 train loss:3.969071
+step:1036 train loss:3.925973
+step:1037 train loss:3.938655
+step:1038 train loss:4.011632
+step:1039 train loss:4.091969
+step:1040 train loss:3.954931
+step:1041 train loss:3.969803
+step:1042 train loss:3.943537
+step:1043 train loss:3.916167
+step:1044 train loss:4.011185
+step:1045 train loss:3.944720
+step:1046 train loss:3.855110
+step:1047 train loss:3.967314
+step:1048 train loss:3.933725
+step:1049 train loss:4.026780
+step:1050 train loss:3.916263
+step:1051 train loss:3.907289
+step:1052 train loss:4.019715
+step:1053 train loss:3.941101
+step:1054 train loss:3.931720
+step:1055 train loss:3.972822
+step:1056 train loss:3.899542
+step:1057 train loss:3.873503
+step:1058 train loss:3.910534
+step:1059 train loss:3.933813
+step:1060 train loss:3.920648
+step:1061 train loss:3.977629
+step:1062 train loss:3.864970
+step:1063 train loss:3.989268
+step:1064 train loss:3.932365
+step:1065 train loss:3.900620
+step:1066 train loss:3.963389
+step:1067 train loss:3.870997
+step:1068 train loss:3.937557
+step:1069 train loss:3.924787
+step:1070 train loss:3.946663
+step:1071 train loss:3.933115
+step:1072 train loss:3.953652
+step:1073 train loss:3.873840
+step:1074 train loss:3.931653
+step:1075 train loss:3.867524
+step:1076 train loss:3.984814
+step:1077 train loss:3.922936
+step:1078 train loss:3.975248
+step:1079 train loss:3.967391
+step:1080 train loss:3.905135
+step:1081 train loss:3.929991
+step:1082 train loss:3.927953
+step:1083 train loss:3.891830
+step:1084 train loss:3.905551
+step:1085 train loss:3.933024
+step:1086 train loss:3.923809
+step:1087 train loss:3.921296
+step:1088 train loss:3.928435
+step:1089 train loss:3.921442
+step:1090 train loss:3.879214
+step:1091 train loss:3.865694
+step:1092 train loss:3.967604
+step:1093 train loss:3.895097
+step:1094 train loss:3.877093
+step:1095 train loss:3.951813
+step:1096 train loss:3.913936
+step:1097 train loss:3.884236
+step:1098 train loss:3.897169
+step:1099 train loss:3.939652
+step:1100 train loss:3.962886
+step:1101 train loss:3.947030
+step:1102 train loss:3.967351
+step:1103 train loss:3.915780
+step:1104 train loss:3.927171
+step:1105 train loss:3.939467
+step:1106 train loss:3.973014
+step:1107 train loss:3.986967
+step:1108 train loss:4.002523
+step:1109 train loss:3.942834
+step:1110 train loss:3.897256
+step:1111 train loss:3.896074
+step:1112 train loss:3.889287
+step:1113 train loss:3.795183
+step:1114 train loss:3.869312
+step:1115 train loss:3.940021
+step:1116 train loss:3.918025
+step:1117 train loss:3.943326
+step:1118 train loss:3.973754
+step:1119 train loss:3.956108
+step:1120 train loss:3.921861
+step:1121 train loss:3.908311
+step:1122 train loss:3.965761
+step:1123 train loss:3.940778
+step:1124 train loss:3.864900
+step:1125 train loss:3.884663
+step:1126 train loss:3.885263
+step:1127 train loss:3.905897
+step:1128 train loss:3.842694
+step:1129 train loss:3.987027
+step:1130 train loss:3.853091
+step:1131 train loss:3.944433
+step:1132 train loss:3.897341
+step:1133 train loss:3.897770
+step:1134 train loss:3.913785
+step:1135 train loss:3.945745
+step:1136 train loss:3.930482
+step:1137 train loss:3.928375
+step:1138 train loss:3.878631
+step:1139 train loss:4.012078
+step:1140 train loss:3.854665
+step:1141 train loss:3.925866
+step:1142 train loss:3.883603
+step:1143 train loss:3.998202
+step:1144 train loss:3.920395
+step:1145 train loss:3.873887
+step:1146 train loss:3.912208
+step:1147 train loss:3.886749
+step:1148 train loss:3.921994
+step:1149 train loss:3.997519
+step:1150 train loss:3.949852
+step:1151 train loss:3.933699
+step:1152 train loss:3.862453
+step:1153 train loss:3.871907
+step:1154 train loss:3.878570
+step:1155 train loss:3.943954
+step:1156 train loss:3.881603
+step:1157 train loss:3.945322
+step:1158 train loss:3.910732
+step:1159 train loss:3.938544
+step:1160 train loss:3.874466
+step:1161 train loss:3.939105
+step:1162 train loss:3.906281
+step:1163 train loss:3.838763
+step:1164 train loss:3.848461
+step:1165 train loss:3.899048
+step:1166 train loss:3.867635
+step:1167 train loss:3.875611
+step:1168 train loss:3.930719
+step:1169 train loss:3.868754
+step:1170 train loss:3.972005
+step:1171 train loss:3.870772
+step:1172 train loss:3.852270
+step:1173 train loss:3.912057
+step:1174 train loss:3.836881
+step:1175 train loss:3.916032
+step:1176 train loss:3.983089
+step:1177 train loss:3.891183
+step:1178 train loss:3.861855
+step:1179 train loss:3.841957
+step:1180 train loss:3.901367
+step:1181 train loss:3.857945
+step:1182 train loss:3.974981
+step:1183 train loss:3.873095
+step:1184 train loss:3.818384
+step:1185 train loss:3.882956
+step:1186 train loss:3.894314
+step:1187 train loss:3.854965
+step:1188 train loss:3.877975
+step:1189 train loss:3.838130
+step:1190 train loss:3.895848
+step:1191 train loss:3.867215
+step:1192 train loss:3.923374
+step:1193 train loss:3.915555
+step:1194 train loss:3.995477
+step:1195 train loss:3.935253
+step:1196 train loss:3.937064
+step:1197 train loss:3.849087
+step:1198 train loss:3.846049
+step:1199 train loss:4.042292
+step:1200 train loss:3.780253
+step:1201 train loss:3.896569
+step:1202 train loss:3.887328
+step:1203 train loss:3.871069
+step:1204 train loss:3.880103
+step:1205 train loss:3.872984
+step:1206 train loss:3.865698
+step:1207 train loss:3.906325
+step:1208 train loss:3.900407
+step:1209 train loss:3.826255
+step:1210 train loss:3.903646
+step:1211 train loss:3.849854
+step:1212 train loss:3.872662
+step:1213 train loss:3.868860
+step:1214 train loss:3.914253
+step:1215 train loss:3.878576
+step:1216 train loss:3.799879
+step:1217 train loss:3.877932
+step:1218 train loss:3.879938
+step:1219 train loss:3.849298
+step:1220 train loss:3.851941
+step:1221 train loss:3.832316
+step:1222 train loss:3.992810
+step:1223 train loss:3.895532
+step:1224 train loss:3.842948
+step:1225 train loss:3.883440
+step:1226 train loss:3.868147
+step:1227 train loss:3.888015
+step:1228 train loss:3.829001
+step:1229 train loss:3.861384
+step:1230 train loss:3.915865
+step:1231 train loss:3.828366
+step:1232 train loss:3.867975
+step:1233 train loss:3.865742
+step:1234 train loss:3.901035
+step:1235 train loss:3.849055
+step:1236 train loss:3.889338
+step:1237 train loss:3.843271
+step:1238 train loss:3.857498
+step:1239 train loss:3.876526
+step:1240 train loss:3.817823
+step:1241 train loss:3.828878
+step:1242 train loss:3.884361
+step:1243 train loss:3.849646
+step:1244 train loss:3.907674
+step:1245 train loss:3.957650
+step:1246 train loss:3.878843
+step:1247 train loss:3.862557
+step:1248 train loss:3.809530
+step:1249 train loss:3.859677
+step:1250 validation loss:3.826611
+step:1250 train loss:3.893163
+step:1251 train loss:3.877445
+step:1252 train loss:3.814513
+step:1253 train loss:3.824313
+step:1254 train loss:3.822617
+step:1255 train loss:3.839972
+step:1256 train loss:3.884478
+step:1257 train loss:3.918450
+step:1258 train loss:3.850078
+step:1259 train loss:3.827213
+step:1260 train loss:3.938088
+step:1261 train loss:3.953469
+step:1262 train loss:3.854068
+step:1263 train loss:3.841062
+step:1264 train loss:3.865801
+step:1265 train loss:3.860689
+step:1266 train loss:3.890187
+step:1267 train loss:3.865319
+step:1268 train loss:3.833306
+step:1269 train loss:3.852707
+step:1270 train loss:3.756754
+step:1271 train loss:3.825117
+step:1272 train loss:3.802334
+step:1273 train loss:3.925034
+step:1274 train loss:3.835282
+step:1275 train loss:3.875565
+step:1276 train loss:3.878094
+step:1277 train loss:3.889197
+step:1278 train loss:3.807224
+step:1279 train loss:3.875180
+step:1280 train loss:3.846453
+step:1281 train loss:3.868839
+step:1282 train loss:3.841353
+step:1283 train loss:3.883650
+step:1284 train loss:3.892779
+step:1285 train loss:3.851052
+step:1286 train loss:3.829006
+step:1287 train loss:3.808278
+step:1288 train loss:3.884778
+step:1289 train loss:3.942343
+step:1290 train loss:3.822271
+step:1291 train loss:3.849371
+step:1292 train loss:3.873369
+step:1293 train loss:3.802647
+step:1294 train loss:3.833345
+step:1295 train loss:3.913428
+step:1296 train loss:3.873119
+step:1297 train loss:3.811568
+step:1298 train loss:3.916762
+step:1299 train loss:3.887358
+step:1300 train loss:3.812883
+step:1301 train loss:3.899492
+step:1302 train loss:3.821421
+step:1303 train loss:3.865287
+step:1304 train loss:3.911485
+step:1305 train loss:3.852350
+step:1306 train loss:3.900435
+step:1307 train loss:3.810319
+step:1308 train loss:3.804298
+step:1309 train loss:3.841265
+step:1310 train loss:3.795567
+step:1311 train loss:3.831146
+step:1312 train loss:3.901070
+step:1313 train loss:3.787488
+step:1314 train loss:3.846757
+step:1315 train loss:3.840177
+step:1316 train loss:3.770811
+step:1317 train loss:3.803823
+step:1318 train loss:3.891773
+step:1319 train loss:3.900659
+step:1320 train loss:3.816842
+step:1321 train loss:3.858860
+step:1322 train loss:3.899477
+step:1323 train loss:3.870969
+step:1324 train loss:3.952270
+step:1325 train loss:3.842304
+step:1326 train loss:3.873678
+step:1327 train loss:3.916664
+step:1328 train loss:3.773400
+step:1329 train loss:3.807928
+step:1330 train loss:3.890032
+step:1331 train loss:3.761925
+step:1332 train loss:3.895957
+step:1333 train loss:3.831857
+step:1334 train loss:3.872089
+step:1335 train loss:3.898932
+step:1336 train loss:3.873413
+step:1337 train loss:3.864900
+step:1338 train loss:3.928332
+step:1339 train loss:3.820869
+step:1340 train loss:3.873405
+step:1341 train loss:3.884965
+step:1342 train loss:3.890243
+step:1343 train loss:3.732008
+step:1344 train loss:3.943747
+step:1345 train loss:3.923591
+step:1346 train loss:3.863621
+step:1347 train loss:3.847536
+step:1348 train loss:3.785698
+step:1349 train loss:3.805537
+step:1350 train loss:3.812785
+step:1351 train loss:3.828299
+step:1352 train loss:3.872229
+step:1353 train loss:3.823015
+step:1354 train loss:3.870942
+step:1355 train loss:3.865641
+step:1356 train loss:3.820480
+step:1357 train loss:3.841460
+step:1358 train loss:3.792540
+step:1359 train loss:3.849933
+step:1360 train loss:4.029719
+step:1361 train loss:3.840365
+step:1362 train loss:3.825315
+step:1363 train loss:3.825120
+step:1364 train loss:3.766476
+step:1365 train loss:3.814825
+step:1366 train loss:3.804184
+step:1367 train loss:3.792404
+step:1368 train loss:3.809503
+step:1369 train loss:3.841174
+step:1370 train loss:3.861629
+step:1371 train loss:3.873272
+step:1372 train loss:3.774899
+step:1373 train loss:3.875476
+step:1374 train loss:3.927669
+step:1375 train loss:3.801474
+step:1376 train loss:3.858068
+step:1377 train loss:3.868899
+step:1378 train loss:3.834599
+step:1379 train loss:3.822088
+step:1380 train loss:3.857166
+step:1381 train loss:3.860100
+step:1382 train loss:3.823347
+step:1383 train loss:3.760774
+step:1384 train loss:3.873748
+step:1385 train loss:3.825006
+step:1386 train loss:3.781129
+step:1387 train loss:3.855453
+step:1388 train loss:3.835993
+step:1389 train loss:3.808662
+step:1390 train loss:3.836672
+step:1391 train loss:3.827110
+step:1392 train loss:3.858364
+step:1393 train loss:3.919786
+step:1394 train loss:3.811592
+step:1395 train loss:3.820995
+step:1396 train loss:3.859890
+step:1397 train loss:3.878932
+step:1398 train loss:3.841174
+step:1399 train loss:3.839590
+step:1400 train loss:3.797957
+step:1401 train loss:3.804448
+step:1402 train loss:3.844164
+step:1403 train loss:3.802745
+step:1404 train loss:3.761545
+step:1405 train loss:3.807818
+step:1406 train loss:3.861656
+step:1407 train loss:3.807901
+step:1408 train loss:3.764139
+step:1409 train loss:3.851090
+step:1410 train loss:3.800162
+step:1411 train loss:3.890156
+step:1412 train loss:3.837520
+step:1413 train loss:3.812350
+step:1414 train loss:3.849592
+step:1415 train loss:3.799215
+step:1416 train loss:3.856426
+step:1417 train loss:3.769259
+step:1418 train loss:3.812027
+step:1419 train loss:3.808388
+step:1420 train loss:3.818705
+step:1421 train loss:3.830803
+step:1422 train loss:3.888608
+step:1423 train loss:3.867841
+step:1424 train loss:3.777815
+step:1425 train loss:3.804133
+step:1426 train loss:3.826435
+step:1427 train loss:3.749098
+step:1428 train loss:3.834265
+step:1429 train loss:3.815309
+step:1430 train loss:3.766468
+step:1431 train loss:3.832919
+step:1432 train loss:3.837049
+step:1433 train loss:3.793988
+step:1434 train loss:3.747606
+step:1435 train loss:3.838619
+step:1436 train loss:3.777488
+step:1437 train loss:3.778184
+step:1438 train loss:3.773561
+step:1439 train loss:3.749923
+step:1440 train loss:3.817825
+step:1441 train loss:3.880980
+step:1442 train loss:3.812835
+step:1443 train loss:3.748357
+step:1444 train loss:3.772249
+step:1445 train loss:3.792529
+step:1446 train loss:3.829174
+step:1447 train loss:3.793631
+step:1448 train loss:3.772240
+step:1449 train loss:3.851706
+step:1450 train loss:3.803333
+step:1451 train loss:3.746500
+step:1452 train loss:3.806686
+step:1453 train loss:3.805840
+step:1454 train loss:3.790751
+step:1455 train loss:3.743222
+step:1456 train loss:3.783026
+step:1457 train loss:3.802218
+step:1458 train loss:3.871597
+step:1459 train loss:3.763002
+step:1460 train loss:3.770894
+step:1461 train loss:3.878401
+step:1462 train loss:3.803609
+step:1463 train loss:3.778724
+step:1464 train loss:3.799502
+step:1465 train loss:3.751837
+step:1466 train loss:3.847315
+step:1467 train loss:3.806876
+step:1468 train loss:3.783533
+step:1469 train loss:3.802670
+step:1470 train loss:3.777166
+step:1471 train loss:3.775966
+step:1472 train loss:3.765676
+step:1473 train loss:3.760405
+step:1474 train loss:3.741285
+step:1475 train loss:3.807868
+step:1476 train loss:3.818596
+step:1477 train loss:3.802255
+step:1478 train loss:3.782046
+step:1479 train loss:3.759466
+step:1480 train loss:3.756655
+step:1481 train loss:3.767833
+step:1482 train loss:3.828053
+step:1483 train loss:3.782181
+step:1484 train loss:3.826070
+step:1485 train loss:3.854391
+step:1486 train loss:3.789398
+step:1487 train loss:3.769118
+step:1488 train loss:3.746739
+step:1489 train loss:3.767208
+step:1490 train loss:3.841179
+step:1491 train loss:3.780541
+step:1492 train loss:3.788824
+step:1493 train loss:3.798223
+step:1494 train loss:3.792627
+step:1495 train loss:3.732736
+step:1496 train loss:3.807983
+step:1497 train loss:3.756237
+step:1498 train loss:3.745954
+step:1499 train loss:3.737407
+step:1500 validation loss:3.757109
+step:1500 train loss:3.815202
+step:1501 train loss:3.740230
+step:1502 train loss:3.718124
+step:1503 train loss:3.816290
+step:1504 train loss:3.659916
+step:1505 train loss:3.776286
+step:1506 train loss:3.738159
+step:1507 train loss:3.734453
+step:1508 train loss:3.719521
+step:1509 train loss:3.807832
+step:1510 train loss:3.723871
+step:1511 train loss:3.758124
+step:1512 train loss:3.765553
+step:1513 train loss:3.755434
+step:1514 train loss:3.785947
+step:1515 train loss:3.808322
+step:1516 train loss:3.735939
+step:1517 train loss:3.829880
+step:1518 train loss:3.800495
+step:1519 train loss:3.817586
+step:1520 train loss:3.778425
+step:1521 train loss:3.778063
+step:1522 train loss:3.797825
+step:1523 train loss:3.763740
+step:1524 train loss:3.750288
+step:1525 train loss:3.763456
+step:1526 train loss:3.753563
+step:1527 train loss:3.798655
+step:1528 train loss:3.803994
+step:1529 train loss:3.816316
+step:1530 train loss:3.793533
+step:1531 train loss:3.740641
+step:1532 train loss:3.832451
+step:1533 train loss:3.772871
+step:1534 train loss:3.763093
+step:1535 train loss:3.778946
+step:1536 train loss:3.821423
+step:1537 train loss:3.795830
+step:1538 train loss:3.764590
+step:1539 train loss:3.780171
+step:1540 train loss:3.743726
+step:1541 train loss:3.831665
+step:1542 train loss:3.776404
+step:1543 train loss:3.879735
+step:1544 train loss:3.733360
+step:1545 train loss:3.736825
+step:1546 train loss:3.750626
+step:1547 train loss:3.782934
+step:1548 train loss:3.743047
+step:1549 train loss:3.805847
+step:1550 train loss:3.774482
+step:1551 train loss:3.786910
+step:1552 train loss:3.798052
+step:1553 train loss:3.800518
+step:1554 train loss:3.789695