romanevrwld/Random-Router-Baseline-5000-Multimodality
02
1{2 "best_global_step": null,3 "best_metric": null,4 "best_model_checkpoint": null,5 "epoch": 25.64,6 "eval_steps": 500,7 "global_step": 5000,8 "is_hyper_param_search": false,9 "is_local_process_zero": true,10 "is_world_process_zero": true,11 "log_history": [12 {13 "balancing_entropy": 0.0,14 "balancing_loss": 0.0,15 "epoch": 0.0512,16 "grad_norm": 0.86328125,17 "learning_rate": 3.076923076923077e-05,18 "loss": 12.4822,19 "router_z_loss": 0.0,20 "step": 1021 },22 {23 "balancing_entropy": 0.0,24 "balancing_loss": 0.0,25 "epoch": 0.1024,26 "grad_norm": 0.90625,27 "learning_rate": 6.153846153846154e-05,28 "loss": 12.2583,29 "router_z_loss": 0.0,30 "step": 2031 },32 {33 "balancing_entropy": 0.0,34 "balancing_loss": 0.0,35 "epoch": 0.1536,36 "grad_norm": 0.84765625,37 "learning_rate": 9.230769230769232e-05,38 "loss": 11.7958,39 "router_z_loss": 0.0,40 "step": 3041 },42 {43 "balancing_entropy": 0.0,44 "balancing_loss": 0.0,45 "epoch": 0.2048,46 "grad_norm": 0.8203125,47 "learning_rate": 0.00012307692307692307,48 "loss": 11.1369,49 "router_z_loss": 0.0,50 "step": 4051 },52 {53 "balancing_entropy": 0.0,54 "balancing_loss": 0.0,55 "epoch": 0.256,56 "grad_norm": 1.015625,57 "learning_rate": 0.00015384615384615385,58 "loss": 10.2261,59 "router_z_loss": 0.0,60 "step": 5061 },62 {63 "balancing_entropy": 0.0,64 "balancing_loss": 0.0,65 "epoch": 0.3072,66 "grad_norm": 0.9140625,67 "learning_rate": 0.00018461538461538463,68 "loss": 9.3587,69 "router_z_loss": 0.0,70 "step": 6071 },72 {73 "balancing_entropy": 0.0,74 "balancing_loss": 0.0,75 "epoch": 0.3584,76 "grad_norm": 0.5546875,77 "learning_rate": 0.0002153846153846154,78 "loss": 8.7066,79 "router_z_loss": 0.0,80 "step": 7081 },82 {83 "balancing_entropy": 0.0,84 "balancing_loss": 0.0,85 "epoch": 0.4096,86 "grad_norm": 4.03125,87 "learning_rate": 0.00024615384615384614,88 "loss": 8.2878,89 "router_z_loss": 0.0,90 "step": 8091 },92 {93 "balancing_entropy": 0.0,94 "balancing_loss": 0.0,95 "epoch": 0.4608,96 "grad_norm": 2.375,97 "learning_rate": 0.00027692307692307695,98 "loss": 7.9893,99 "router_z_loss": 0.0,100 "step": 90101 },102 {103 "balancing_entropy": 0.0,104 "balancing_loss": 0.0,105 "epoch": 0.512,106 "grad_norm": 2.109375,107 "learning_rate": 0.0003076923076923077,108 "loss": 7.6368,109 "router_z_loss": 0.0,110 "step": 100111 },112 {113 "balancing_entropy": 0.0,114 "balancing_loss": 0.0,115 "epoch": 0.5632,116 "grad_norm": 0.5078125,117 "learning_rate": 0.00033846153846153846,118 "loss": 7.1774,119 "router_z_loss": 0.0,120 "step": 110121 },122 {123 "balancing_entropy": 0.0,124 "balancing_loss": 0.0,125 "epoch": 0.6144,126 "grad_norm": 0.5703125,127 "learning_rate": 0.00036923076923076927,128 "loss": 6.6833,129 "router_z_loss": 0.0,130 "step": 120131 },132 {133 "balancing_entropy": 0.0,134 "balancing_loss": 0.0,135 "epoch": 0.6656,136 "grad_norm": 1.609375,137 "learning_rate": 0.0004,138 "loss": 6.2922,139 "router_z_loss": 0.0,140 "step": 130141 },142 {143 "balancing_entropy": 0.0,144 "balancing_loss": 0.0,145 "epoch": 0.7168,146 "grad_norm": 0.75390625,147 "learning_rate": 0.0004307692307692308,148 "loss": 6.0386,149 "router_z_loss": 0.0,150 "step": 140151 },152 {153 "balancing_entropy": 0.0,154 "balancing_loss": 0.0,155 "epoch": 0.768,156 "grad_norm": 0.71484375,157 "learning_rate": 0.0004615384615384616,158 "loss": 5.8256,159 "router_z_loss": 0.0,160 "step": 150161 },162 {163 "balancing_entropy": 0.0,164 "balancing_loss": 0.0,165 "epoch": 0.8192,166 "grad_norm": 1.1875,167 "learning_rate": 0.0004923076923076923,168 "loss": 5.6412,169 "router_z_loss": 0.0,170 "step": 160171 },172 {173 "balancing_entropy": 0.0,174 "balancing_loss": 0.0,175 "epoch": 0.8704,176 "grad_norm": 0.91015625,177 "learning_rate": 0.0005230769230769231,178 "loss": 5.4734,179 "router_z_loss": 0.0,180 "step": 170181 },182 {183 "balancing_entropy": 0.0,184 "balancing_loss": 0.0,185 "epoch": 0.9216,186 "grad_norm": 1.0625,187 "learning_rate": 0.0005538461538461539,188 "loss": 5.3141,189 "router_z_loss": 0.0,190 "step": 180191 },192 {193 "balancing_entropy": 0.0,194 "balancing_loss": 0.0,195 "epoch": 0.9728,196 "grad_norm": 0.9375,197 "learning_rate": 0.0005846153846153846,198 "loss": 5.1415,199 "router_z_loss": 0.0,200 "step": 190201 },202 {203 "balancing_entropy": 0.0,204 "balancing_loss": 0.0,205 "epoch": 1.0256,206 "grad_norm": 0.81640625,207 "learning_rate": 0.0006153846153846154,208 "loss": 5.5058,209 "router_z_loss": 0.0,210 "step": 200211 },212 {213 "balancing_entropy": 0.0,214 "balancing_loss": 0.0,215 "epoch": 1.0768,216 "grad_norm": 0.5546875,217 "learning_rate": 0.0006461538461538462,218 "loss": 4.8577,219 "router_z_loss": 0.0,220 "step": 210221 },222 {223 "balancing_entropy": 0.0,224 "balancing_loss": 0.0,225 "epoch": 1.1280000000000001,226 "grad_norm": 0.859375,227 "learning_rate": 0.0006769230769230769,228 "loss": 4.7405,229 "router_z_loss": 0.0,230 "step": 220231 },232 {233 "balancing_entropy": 0.0,234 "balancing_loss": 0.0,235 "epoch": 1.1792,236 "grad_norm": 0.75,237 "learning_rate": 0.0007076923076923077,238 "loss": 4.6311,239 "router_z_loss": 0.0,240 "step": 230241 },242 {243 "balancing_entropy": 0.0,244 "balancing_loss": 0.0,245 "epoch": 1.2304,246 "grad_norm": 1.5625,247 "learning_rate": 0.0007384615384615385,248 "loss": 4.4939,249 "router_z_loss": 0.0,250 "step": 240251 },252 {253 "balancing_entropy": 0.0,254 "balancing_loss": 0.0,255 "epoch": 1.2816,256 "grad_norm": 1.8046875,257 "learning_rate": 0.0007692307692307692,258 "loss": 4.4278,259 "router_z_loss": 0.0,260 "step": 250261 },262 {263 "balancing_entropy": 0.0,264 "balancing_loss": 0.0,265 "epoch": 1.3328,266 "grad_norm": 1.4453125,267 "learning_rate": 0.0008,268 "loss": 4.3309,269 "router_z_loss": 0.0,270 "step": 260271 },272 {273 "balancing_entropy": 0.0,274 "balancing_loss": 0.0,275 "epoch": 1.384,276 "grad_norm": 0.52734375,277 "learning_rate": 0.0008307692307692308,278 "loss": 4.2308,279 "router_z_loss": 0.0,280 "step": 270281 },282 {283 "balancing_entropy": 0.0,284 "balancing_loss": 0.0,285 "epoch": 1.4352,286 "grad_norm": 1.0546875,287 "learning_rate": 0.0008615384615384615,288 "loss": 4.1648,289 "router_z_loss": 0.0,290 "step": 280291 },292 {293 "balancing_entropy": 0.0,294 "balancing_loss": 0.0,295 "epoch": 1.4864,296 "grad_norm": 1.0859375,297 "learning_rate": 0.0008923076923076924,298 "loss": 4.0882,299 "router_z_loss": 0.0,300 "step": 290301 },302 {303 "balancing_entropy": 0.0,304 "balancing_loss": 0.0,305 "epoch": 1.5375999999999999,306 "grad_norm": 0.58203125,307 "learning_rate": 0.0009230769230769232,308 "loss": 4.0039,309 "router_z_loss": 0.0,310 "step": 300311 },312 {313 "balancing_entropy": 0.0,314 "balancing_loss": 0.0,315 "epoch": 1.5888,316 "grad_norm": 0.68359375,317 "learning_rate": 0.0009538461538461538,318 "loss": 3.9642,319 "router_z_loss": 0.0,320 "step": 310321 },322 {323 "balancing_entropy": 0.0,324 "balancing_loss": 0.0,325 "epoch": 1.6400000000000001,326 "grad_norm": 0.77734375,327 "learning_rate": 0.0009846153846153846,328 "loss": 3.8901,329 "router_z_loss": 0.0,330 "step": 320331 },332 {333 "balancing_entropy": 0.0,334 "balancing_loss": 0.0,335 "epoch": 1.6912,336 "grad_norm": 0.9453125,337 "learning_rate": 0.0010153846153846155,338 "loss": 3.8254,339 "router_z_loss": 0.0,340 "step": 330341 },342 {343 "balancing_entropy": 0.0,344 "balancing_loss": 0.0,345 "epoch": 1.7424,346 "grad_norm": 0.71484375,347 "learning_rate": 0.0010461538461538462,348 "loss": 3.7841,349 "router_z_loss": 0.0,350 "step": 340351 },352 {353 "balancing_entropy": 0.0,354 "balancing_loss": 0.0,355 "epoch": 1.7936,356 "grad_norm": 0.79296875,357 "learning_rate": 0.0010769230769230769,358 "loss": 3.751,359 "router_z_loss": 0.0,360 "step": 350361 },362 {363 "balancing_entropy": 0.0,364 "balancing_loss": 0.0,365 "epoch": 1.8448,366 "grad_norm": 0.890625,367 "learning_rate": 0.0011076923076923078,368 "loss": 3.6586,369 "router_z_loss": 0.0,370 "step": 360371 },372 {373 "balancing_entropy": 0.0,374 "balancing_loss": 0.0,375 "epoch": 1.896,376 "grad_norm": 1.4140625,377 "learning_rate": 0.0011384615384615385,378 "loss": 3.6627,379 "router_z_loss": 0.0,380 "step": 370381 },382 {383 "balancing_entropy": 0.0,384 "balancing_loss": 0.0,385 "epoch": 1.9472,386 "grad_norm": 0.5859375,387 "learning_rate": 0.0011692307692307692,388 "loss": 3.5893,389 "router_z_loss": 0.0,390 "step": 380391 },392 {393 "balancing_entropy": 0.0,394 "balancing_loss": 0.0,395 "epoch": 1.9984,396 "grad_norm": 0.5703125,397 "learning_rate": 0.0012000000000000001,398 "loss": 3.5727,399 "router_z_loss": 0.0,400 "step": 390401 },402 {403 "balancing_entropy": 0.0,404 "balancing_loss": 0.0,405 "epoch": 2.0512,406 "grad_norm": 0.400390625,407 "learning_rate": 0.0012307692307692308,408 "loss": 3.8897,409 "router_z_loss": 0.0,410 "step": 400411 },412 {413 "balancing_entropy": 0.0,414 "balancing_loss": 0.0,415 "epoch": 2.1024,416 "grad_norm": 0.486328125,417 "learning_rate": 0.0012615384615384615,418 "loss": 3.501,419 "router_z_loss": 0.0,420 "step": 410421 },422 {423 "balancing_entropy": 0.0,424 "balancing_loss": 0.0,425 "epoch": 2.1536,426 "grad_norm": 0.87109375,427 "learning_rate": 0.0012923076923076924,428 "loss": 3.4462,429 "router_z_loss": 0.0,430 "step": 420431 },432 {433 "balancing_entropy": 0.0,434 "balancing_loss": 0.0,435 "epoch": 2.2048,436 "grad_norm": 0.5078125,437 "learning_rate": 0.0013230769230769231,438 "loss": 3.4335,439 "router_z_loss": 0.0,440 "step": 430441 },442 {443 "balancing_entropy": 0.0,444 "balancing_loss": 0.0,445 "epoch": 2.2560000000000002,446 "grad_norm": 0.53515625,447 "learning_rate": 0.0013538461538461538,448 "loss": 3.3947,449 "router_z_loss": 0.0,450 "step": 440451 },452 {453 "balancing_entropy": 0.0,454 "balancing_loss": 0.0,455 "epoch": 2.3072,456 "grad_norm": 0.66796875,457 "learning_rate": 0.0013846153846153847,458 "loss": 3.37,459 "router_z_loss": 0.0,460 "step": 450461 },462 {463 "balancing_entropy": 0.0,464 "balancing_loss": 0.0,465 "epoch": 2.3584,466 "grad_norm": 0.5703125,467 "learning_rate": 0.0014153846153846154,468 "loss": 3.345,469 "router_z_loss": 0.0,470 "step": 460471 },472 {473 "balancing_entropy": 0.0,474 "balancing_loss": 0.0,475 "epoch": 2.4096,476 "grad_norm": 0.447265625,477 "learning_rate": 0.0014461538461538461,478 "loss": 3.355,479 "router_z_loss": 0.0,480 "step": 470481 },482 {483 "balancing_entropy": 0.0,484 "balancing_loss": 0.0,485 "epoch": 2.4608,486 "grad_norm": 0.8671875,487 "learning_rate": 0.001476923076923077,488 "loss": 3.3138,489 "router_z_loss": 0.0,490 "step": 480491 },492 {493 "balancing_entropy": 0.0,494 "balancing_loss": 0.0,495 "epoch": 2.512,496 "grad_norm": 0.6171875,497 "learning_rate": 0.0015076923076923078,498 "loss": 3.2843,499 "router_z_loss": 0.0,500 "step": 490501 },502 {503 "balancing_entropy": 0.0,504 "balancing_loss": 0.0,505 "epoch": 2.5632,506 "grad_norm": 0.60546875,507 "learning_rate": 0.0015384615384615385,508 "loss": 3.2832,509 "router_z_loss": 0.0,510 "step": 500511 },512 {513 "balancing_entropy": 0.0,514 "balancing_loss": 0.0,515 "epoch": 2.6144,516 "grad_norm": 0.53125,517 "learning_rate": 0.0015692307692307694,518 "loss": 3.2416,519 "router_z_loss": 0.0,520 "step": 510521 },522 {523 "balancing_entropy": 0.0,524 "balancing_loss": 0.0,525 "epoch": 2.6656,526 "grad_norm": 0.421875,527 "learning_rate": 0.0016,528 "loss": 3.2343,529 "router_z_loss": 0.0,530 "step": 520531 },532 {533 "balancing_entropy": 0.0,534 "balancing_loss": 0.0,535 "epoch": 2.7168,536 "grad_norm": 0.404296875,537 "learning_rate": 0.0016307692307692308,538 "loss": 3.2167,539 "router_z_loss": 0.0,540 "step": 530541 },542 {543 "balancing_entropy": 0.0,544 "balancing_loss": 0.0,545 "epoch": 2.768,546 "grad_norm": 0.4765625,547 "learning_rate": 0.0016615384615384617,548 "loss": 3.1902,549 "router_z_loss": 0.0,550 "step": 540551 },552 {553 "balancing_entropy": 0.0,554 "balancing_loss": 0.0,555 "epoch": 2.8192,556 "grad_norm": 0.4921875,557 "learning_rate": 0.0016923076923076924,558 "loss": 3.2147,559 "router_z_loss": 0.0,560 "step": 550561 },562 {563 "balancing_entropy": 0.0,564 "balancing_loss": 0.0,565 "epoch": 2.8704,566 "grad_norm": 0.640625,567 "learning_rate": 0.001723076923076923,568 "loss": 3.1914,569 "router_z_loss": 0.0,570 "step": 560571 },572 {573 "balancing_entropy": 0.0,574 "balancing_loss": 0.0,575 "epoch": 2.9215999999999998,576 "grad_norm": 0.396484375,577 "learning_rate": 0.001753846153846154,578 "loss": 3.1648,579 "router_z_loss": 0.0,580 "step": 570581 },582 {583 "balancing_entropy": 0.0,584 "balancing_loss": 0.0,585 "epoch": 2.9728,586 "grad_norm": 0.56640625,587 "learning_rate": 0.0017846153846153847,588 "loss": 3.1473,589 "router_z_loss": 0.0,590 "step": 580591 },592 {593 "balancing_entropy": 0.0,594 "balancing_loss": 0.0,595 "epoch": 3.0256,596 "grad_norm": 1.984375,597 "learning_rate": 0.0018153846153846154,598 "loss": 3.555,599 "router_z_loss": 0.0,600 "step": 590601 },602 {603 "balancing_entropy": 0.0,604 "balancing_loss": 0.0,605 "epoch": 3.0768,606 "grad_norm": 0.88671875,607 "learning_rate": 0.0018461538461538463,608 "loss": 3.2716,609 "router_z_loss": 0.0,610 "step": 600611 },612 {613 "balancing_entropy": 0.0,614 "balancing_loss": 0.0,615 "epoch": 3.128,616 "grad_norm": 0.3515625,617 "learning_rate": 0.001876923076923077,618 "loss": 3.1852,619 "router_z_loss": 0.0,620 "step": 610621 },622 {623 "balancing_entropy": 0.0,624 "balancing_loss": 0.0,625 "epoch": 3.1792,626 "grad_norm": 0.345703125,627 "learning_rate": 0.0019076923076923075,628 "loss": 3.1403,629 "router_z_loss": 0.0,630 "step": 620631 },632 {633 "balancing_entropy": 0.0,634 "balancing_loss": 0.0,635 "epoch": 3.2304,636 "grad_norm": 0.4296875,637 "learning_rate": 0.0019384615384615386,638 "loss": 3.1375,639 "router_z_loss": 0.0,640 "step": 630641 },642 {643 "balancing_entropy": 0.0,644 "balancing_loss": 0.0,645 "epoch": 3.2816,646 "grad_norm": 0.359375,647 "learning_rate": 0.001969230769230769,648 "loss": 3.1062,649 "router_z_loss": 0.0,650 "step": 640651 },652 {653 "balancing_entropy": 0.0,654 "balancing_loss": 0.0,655 "epoch": 3.3327999999999998,656 "grad_norm": 0.486328125,657 "learning_rate": 0.002,658 "loss": 3.0988,659 "router_z_loss": 0.0,660 "step": 650661 },662 {663 "balancing_entropy": 0.0,664 "balancing_loss": 0.0,665 "epoch": 3.384,666 "grad_norm": 0.458984375,667 "learning_rate": 0.002030769230769231,668 "loss": 3.0966,669 "router_z_loss": 0.0,670 "step": 660671 },672 {673 "balancing_entropy": 0.0,674 "balancing_loss": 0.0,675 "epoch": 3.4352,676 "grad_norm": 0.33203125,677 "learning_rate": 0.0020615384615384614,678 "loss": 3.0842,679 "router_z_loss": 0.0,680 "step": 670681 },682 {683 "balancing_entropy": 0.0,684 "balancing_loss": 0.0,685 "epoch": 3.4864,686 "grad_norm": 0.306640625,687 "learning_rate": 0.0020923076923076924,688 "loss": 3.0998,689 "router_z_loss": 0.0,690 "step": 680691 },692 {693 "balancing_entropy": 0.0,694 "balancing_loss": 0.0,695 "epoch": 3.5376,696 "grad_norm": 0.4140625,697 "learning_rate": 0.0021230769230769233,698 "loss": 3.074,699 "router_z_loss": 0.0,700 "step": 690701 },702 {703 "balancing_entropy": 0.0,704 "balancing_loss": 0.0,705 "epoch": 3.5888,706 "grad_norm": 0.408203125,707 "learning_rate": 0.0021538461538461538,708 "loss": 3.0512,709 "router_z_loss": 0.0,710 "step": 700711 },712 {713 "balancing_entropy": 0.0,714 "balancing_loss": 0.0,715 "epoch": 3.64,716 "grad_norm": 0.271484375,717 "learning_rate": 0.0021846153846153847,718 "loss": 3.0719,719 "router_z_loss": 0.0,720 "step": 710721 },722 {723 "balancing_entropy": 0.0,724 "balancing_loss": 0.0,725 "epoch": 3.6912000000000003,726 "grad_norm": 0.3828125,727 "learning_rate": 0.0022153846153846156,728 "loss": 3.0421,729 "router_z_loss": 0.0,730 "step": 720731 },732 {733 "balancing_entropy": 0.0,734 "balancing_loss": 0.0,735 "epoch": 3.7424,736 "grad_norm": 0.337890625,737 "learning_rate": 0.002246153846153846,738 "loss": 3.061,739 "router_z_loss": 0.0,740 "step": 730741 },742 {743 "balancing_entropy": 0.0,744 "balancing_loss": 0.0,745 "epoch": 3.7936,746 "grad_norm": 0.26171875,747 "learning_rate": 0.002276923076923077,748 "loss": 3.0319,749 "router_z_loss": 0.0,750 "step": 740751 },752 {753 "balancing_entropy": 0.0,754 "balancing_loss": 0.0,755 "epoch": 3.8448,756 "grad_norm": 1.453125,757 "learning_rate": 0.002307692307692308,758 "loss": 3.0467,759 "router_z_loss": 0.0,760 "step": 750761 },762 {763 "balancing_entropy": 0.0,764 "balancing_loss": 0.0,765 "epoch": 3.896,766 "grad_norm": 0.296875,767 "learning_rate": 0.0023384615384615384,768 "loss": 3.0424,769 "router_z_loss": 0.0,770 "step": 760771 },772 {773 "balancing_entropy": 0.0,774 "balancing_loss": 0.0,775 "epoch": 3.9472,776 "grad_norm": 0.333984375,777 "learning_rate": 0.0023692307692307693,778 "loss": 3.0409,779 "router_z_loss": 0.0,780 "step": 770781 },782 {783 "balancing_entropy": 0.0,784 "balancing_loss": 0.0,785 "epoch": 3.9984,786 "grad_norm": 0.41015625,787 "learning_rate": 0.0024000000000000002,788 "loss": 3.0104,789 "router_z_loss": 0.0,790 "step": 780791 },792 {793 "balancing_entropy": 0.0,794 "balancing_loss": 0.0,795 "epoch": 4.0512,796 "grad_norm": 0.2578125,797 "learning_rate": 0.0024307692307692307,798 "loss": 3.3914,799 "router_z_loss": 0.0,800 "step": 790801 },802 {803 "balancing_entropy": 0.0,804 "balancing_loss": 0.0,805 "epoch": 4.1024,806 "grad_norm": 0.26171875,807 "learning_rate": 0.0024615384615384616,808 "loss": 2.9982,809 "router_z_loss": 0.0,810 "step": 800811 },812 {813 "balancing_entropy": 0.0,814 "balancing_loss": 0.0,815 "epoch": 4.1536,816 "grad_norm": 0.2890625,817 "learning_rate": 0.0024923076923076925,818 "loss": 3.0196,819 "router_z_loss": 0.0,820 "step": 810821 },822 {823 "balancing_entropy": 0.0,824 "balancing_loss": 0.0,825 "epoch": 4.2048,826 "grad_norm": 0.2578125,827 "learning_rate": 0.002523076923076923,828 "loss": 2.9782,829 "router_z_loss": 0.0,830 "step": 820831 },832 {833 "balancing_entropy": 0.0,834 "balancing_loss": 0.0,835 "epoch": 4.256,836 "grad_norm": 0.271484375,837 "learning_rate": 0.002553846153846154,838 "loss": 2.9809,839 "router_z_loss": 0.0,840 "step": 830841 },842 {843 "balancing_entropy": 0.0,844 "balancing_loss": 0.0,845 "epoch": 4.3072,846 "grad_norm": 0.359375,847 "learning_rate": 0.002584615384615385,848 "loss": 2.9719,849 "router_z_loss": 0.0,850 "step": 840851 },852 {853 "balancing_entropy": 0.0,854 "balancing_loss": 0.0,855 "epoch": 4.3584,856 "grad_norm": 0.275390625,857 "learning_rate": 0.0026153846153846153,858 "loss": 2.967,859 "router_z_loss": 0.0,860 "step": 850861 },862 {863 "balancing_entropy": 0.0,864 "balancing_loss": 0.0,865 "epoch": 4.4096,866 "grad_norm": 0.2490234375,867 "learning_rate": 0.0026461538461538463,868 "loss": 2.951,869 "router_z_loss": 0.0,870 "step": 860871 },872 {873 "balancing_entropy": 0.0,874 "balancing_loss": 0.0,875 "epoch": 4.4608,876 "grad_norm": 0.279296875,877 "learning_rate": 0.002676923076923077,878 "loss": 2.9363,879 "router_z_loss": 0.0,880 "step": 870881 },882 {883 "balancing_entropy": 0.0,884 "balancing_loss": 0.0,885 "epoch": 4.5120000000000005,886 "grad_norm": 0.26171875,887 "learning_rate": 0.0027076923076923077,888 "loss": 2.9139,889 "router_z_loss": 0.0,890 "step": 880891 },892 {893 "balancing_entropy": 0.0,894 "balancing_loss": 0.0,895 "epoch": 4.5632,896 "grad_norm": 0.255859375,897 "learning_rate": 0.0027384615384615386,898 "loss": 2.9451,899 "router_z_loss": 0.0,900 "step": 890901 },902 {903 "balancing_entropy": 0.0,904 "balancing_loss": 0.0,905 "epoch": 4.6144,906 "grad_norm": 0.2578125,907 "learning_rate": 0.0027692307692307695,908 "loss": 2.9361,909 "router_z_loss": 0.0,910 "step": 900911 },912 {913 "balancing_entropy": 0.0,914 "balancing_loss": 0.0,915 "epoch": 4.6655999999999995,916 "grad_norm": 0.255859375,917 "learning_rate": 0.0028,918 "loss": 2.9243,919 "router_z_loss": 0.0,920 "step": 910921 },922 {923 "balancing_entropy": 0.0,924 "balancing_loss": 0.0,925 "epoch": 4.7168,926 "grad_norm": 0.337890625,927 "learning_rate": 0.002830769230769231,928 "loss": 2.9238,929 "router_z_loss": 0.0,930 "step": 920931 },932 {933 "balancing_entropy": 0.0,934 "balancing_loss": 0.0,935 "epoch": 4.768,936 "grad_norm": 0.2177734375,937 "learning_rate": 0.002861538461538462,938 "loss": 2.9097,939 "router_z_loss": 0.0,940 "step": 930941 },942 {943 "balancing_entropy": 0.0,944 "balancing_loss": 0.0,945 "epoch": 4.8192,946 "grad_norm": 0.365234375,947 "learning_rate": 0.0028923076923076923,948 "loss": 2.9009,949 "router_z_loss": 0.0,950 "step": 940951 },952 {953 "balancing_entropy": 0.0,954 "balancing_loss": 0.0,955 "epoch": 4.8704,956 "grad_norm": 0.248046875,957 "learning_rate": 0.002923076923076923,958 "loss": 2.9211,959 "router_z_loss": 0.0,960 "step": 950961 },962 {963 "balancing_entropy": 0.0,964 "balancing_loss": 0.0,965 "epoch": 4.9216,966 "grad_norm": 0.39453125,967 "learning_rate": 0.002953846153846154,968 "loss": 2.8885,969 "router_z_loss": 0.0,970 "step": 960971 },972 {973 "balancing_entropy": 0.0,974 "balancing_loss": 0.0,975 "epoch": 4.9728,976 "grad_norm": 0.34765625,977 "learning_rate": 0.0029846153846153846,978 "loss": 2.8904,979 "router_z_loss": 0.0,980 "step": 970981 },982 {983 "balancing_entropy": 0.0,984 "balancing_loss": 0.0,985 "epoch": 5.0256,986 "grad_norm": 0.26953125,987 "learning_rate": 0.003,988 "loss": 3.1845,989 "router_z_loss": 0.0,990 "step": 980991 },992 {993 "balancing_entropy": 0.0,994 "balancing_loss": 0.0,995 "epoch": 5.0768,996 "grad_norm": 0.25390625,997 "learning_rate": 0.003,998 "loss": 2.8775,999 "router_z_loss": 0.0,1000 "step": 9901001 },1002 {1003 "balancing_entropy": 0.0,1004 "balancing_loss": 0.0,1005 "epoch": 5.128,1006 "grad_norm": 0.26953125,1007 "learning_rate": 0.003,1008 "loss": 2.8545,1009 "router_z_loss": 0.0,1010 "step": 10001011 },1012 {1013 "balancing_entropy": 0.0,1014 "balancing_loss": 0.0,1015 "epoch": 5.1792,1016 "grad_norm": 0.265625,1017 "learning_rate": 0.003,1018 "loss": 2.8552,1019 "router_z_loss": 0.0,1020 "step": 10101021 },1022 {1023 "balancing_entropy": 0.0,1024 "balancing_loss": 0.0,1025 "epoch": 5.2304,1026 "grad_norm": 0.279296875,1027 "learning_rate": 0.003,1028 "loss": 2.8244,1029 "router_z_loss": 0.0,1030 "step": 10201031 },1032 {1033 "balancing_entropy": 0.0,1034 "balancing_loss": 0.0,1035 "epoch": 5.2816,1036 "grad_norm": 0.30078125,1037 "learning_rate": 0.003,1038 "loss": 2.836,1039 "router_z_loss": 0.0,1040 "step": 10301041 },1042 {1043 "balancing_entropy": 0.0,1044 "balancing_loss": 0.0,1045 "epoch": 5.3328,1046 "grad_norm": 0.2265625,1047 "learning_rate": 0.003,1048 "loss": 2.8203,1049 "router_z_loss": 0.0,1050 "step": 10401051 },1052 {1053 "balancing_entropy": 0.0,1054 "balancing_loss": 0.0,1055 "epoch": 5.384,1056 "grad_norm": 0.330078125,1057 "learning_rate": 0.003,1058 "loss": 2.8179,1059 "router_z_loss": 0.0,1060 "step": 10501061 },1062 {1063 "balancing_entropy": 0.0,1064 "balancing_loss": 0.0,1065 "epoch": 5.4352,1066 "grad_norm": 0.375,1067 "learning_rate": 0.003,1068 "loss": 2.8163,1069 "router_z_loss": 0.0,1070 "step": 10601071 },1072 {1073 "balancing_entropy": 0.0,1074 "balancing_loss": 0.0,1075 "epoch": 5.4864,1076 "grad_norm": 0.2138671875,1077 "learning_rate": 0.003,1078 "loss": 2.8234,1079 "router_z_loss": 0.0,1080 "step": 10701081 },1082 {1083 "balancing_entropy": 0.0,1084 "balancing_loss": 0.0,1085 "epoch": 5.5376,1086 "grad_norm": 0.234375,1087 "learning_rate": 0.003,1088 "loss": 2.7997,1089 "router_z_loss": 0.0,1090 "step": 10801091 },1092 {1093 "balancing_entropy": 0.0,1094 "balancing_loss": 0.0,1095 "epoch": 5.5888,1096 "grad_norm": 0.244140625,1097 "learning_rate": 0.003,1098 "loss": 2.7923,1099 "router_z_loss": 0.0,1100 "step": 10901101 },1102 {1103 "balancing_entropy": 0.0,1104 "balancing_loss": 0.0,1105 "epoch": 5.64,1106 "grad_norm": 0.2197265625,1107 "learning_rate": 0.003,1108 "loss": 2.7918,1109 "router_z_loss": 0.0,1110 "step": 11001111 },1112 {1113 "balancing_entropy": 0.0,1114 "balancing_loss": 0.0,1115 "epoch": 5.6912,1116 "grad_norm": 0.2353515625,1117 "learning_rate": 0.003,1118 "loss": 2.7794,1119 "router_z_loss": 0.0,1120 "step": 11101121 },1122 {1123 "balancing_entropy": 0.0,1124 "balancing_loss": 0.0,1125 "epoch": 5.7424,1126 "grad_norm": 0.361328125,1127 "learning_rate": 0.003,1128 "loss": 2.7793,1129 "router_z_loss": 0.0,1130 "step": 11201131 },1132 {1133 "balancing_entropy": 0.0,1134 "balancing_loss": 0.0,1135 "epoch": 5.7936,1136 "grad_norm": 0.20703125,1137 "learning_rate": 0.003,1138 "loss": 2.793,1139 "router_z_loss": 0.0,1140 "step": 11301141 },1142 {1143 "balancing_entropy": 0.0,1144 "balancing_loss": 0.0,1145 "epoch": 5.8448,1146 "grad_norm": 0.2060546875,1147 "learning_rate": 0.003,1148 "loss": 2.7649,1149 "router_z_loss": 0.0,1150 "step": 11401151 },1152 {1153 "balancing_entropy": 0.0,1154 "balancing_loss": 0.0,1155 "epoch": 5.896,1156 "grad_norm": 0.26171875,1157 "learning_rate": 0.003,1158 "loss": 2.7741,1159 "router_z_loss": 0.0,1160 "step": 11501161 },1162 {1163 "balancing_entropy": 0.0,1164 "balancing_loss": 0.0,1165 "epoch": 5.9472000000000005,1166 "grad_norm": 0.251953125,1167 "learning_rate": 0.003,1168 "loss": 2.7515,1169 "router_z_loss": 0.0,1170 "step": 11601171 },1172 {1173 "balancing_entropy": 0.0,1174 "balancing_loss": 0.0,1175 "epoch": 5.9984,1176 "grad_norm": 0.19921875,1177 "learning_rate": 0.003,1178 "loss": 2.7619,1179 "router_z_loss": 0.0,1180 "step": 11701181 },1182 {1183 "balancing_entropy": 0.0,1184 "balancing_loss": 0.0,1185 "epoch": 6.0512,1186 "grad_norm": 0.2421875,1187 "learning_rate": 0.003,1188 "loss": 3.0268,1189 "router_z_loss": 0.0,1190 "step": 11801191 },1192 {1193 "balancing_entropy": 0.0,1194 "balancing_loss": 0.0,1195 "epoch": 6.1024,1196 "grad_norm": 0.341796875,1197 "learning_rate": 0.003,1198 "loss": 2.7404,1199 "router_z_loss": 0.0,1200 "step": 1190