d1a-e4b / train.log
JohnP1's picture
D1A-E4B v0.5: Kev's skills back on top of v0.4 (hard 55->71, devtools 64->70); PR blast/severity trade-off in the card
e7db32f verified
Raw History Blame Contribute Delete
8.76 kB
Loading weights: 0%| | 0/2076 [00:00<?, ?it/s] Loading weights: 100%|██████████| 2076/2076 [00:00<00:00, 7255.76it/s]
delta: warm start from /init/start: 516 lora tensors and the pointer head loaded
device=cuda trainable params=36.2M
v7/decision-v7/train.jsonl: reconstructing file: 0%| | 0.00B / 18.7MB  v7/decision-v7/train.jsonl: downloading bytes: | 0.00B v7/decision-v7/train.jsonl: downloading bytes: ██████████| 8.02MB, 791kB/s v7/decision-v7/train.jsonl: downloading bytes: ██████████| 8.02MB, 791kB/s
v7/decision-v7/train.jsonl: reconstructing file: 100%|██████████| 18.7MB / 18.7MB, 1.84MB/s
fetched v7/decision-v7/train.jsonl from jaredpalmer/kev-suites@cc4bac803e
replay: 1000 of 12576 suite training records mixed with 14320 from /data/train.jsonl
15320 training requests (holdout=[]), questions by type {'choice': 14042, 'noul': 7662, 'score': 778}
ep0 step 10/1915 loss 0.467 0.508s/rec
ep0 step 20/1915 loss 0.639 0.425s/rec
ep0 step 30/1915 loss 0.458 0.421s/rec
ep0 step 40/1915 loss 0.545 0.419s/rec
ep0 step 50/1915 loss 0.428 0.423s/rec
ep0 step 60/1915 loss 0.529 0.412s/rec
ep0 step 70/1915 loss 0.510 0.422s/rec
ep0 step 80/1915 loss 0.589 0.422s/rec
ep0 step 90/1915 loss 0.643 0.422s/rec
ep0 step 100/1915 loss 0.637 0.428s/rec
ep0 step 110/1915 loss 0.523 0.426s/rec
ep0 step 120/1915 loss 0.696 0.426s/rec
ep0 step 130/1915 loss 0.661 0.425s/rec
ep0 step 140/1915 loss 0.593 0.426s/rec
ep0 step 150/1915 loss 0.575 0.430s/rec
ep0 step 160/1915 loss 0.600 0.432s/rec
ep0 step 170/1915 loss 0.736 0.430s/rec
ep0 step 180/1915 loss 0.590 0.431s/rec
ep0 step 190/1915 loss 0.635 0.434s/rec
ep0 step 200/1915 loss 0.606 0.435s/rec
ep0 step 210/1915 loss 0.603 0.433s/rec
ep0 step 220/1915 loss 0.477 0.437s/rec
ep0 step 230/1915 loss 0.642 0.438s/rec
ep0 step 240/1915 loss 0.544 0.439s/rec
ep0 step 250/1915 loss 0.555 0.440s/rec
ep0 step 260/1915 loss 0.603 0.438s/rec
ep0 step 270/1915 loss 0.510 0.439s/rec
ep0 step 280/1915 loss 0.654 0.437s/rec
ep0 step 290/1915 loss 0.769 0.436s/rec
ep0 step 300/1915 loss 0.460 0.436s/rec
ep0 step 310/1915 loss 0.669 0.436s/rec
ep0 step 320/1915 loss 0.374 0.436s/rec
ep0 step 330/1915 loss 0.701 0.435s/rec
ep0 step 340/1915 loss 0.498 0.434s/rec
ep0 step 350/1915 loss 0.593 0.434s/rec
ep0 step 360/1915 loss 0.623 0.436s/rec
ep0 step 370/1915 loss 0.578 0.436s/rec
ep0 step 380/1915 loss 0.677 0.436s/rec
ep0 step 390/1915 loss 0.453 0.437s/rec
ep0 step 400/1915 loss 0.591 0.437s/rec
ep0 step 410/1915 loss 0.520 0.435s/rec
ep0 step 420/1915 loss 0.790 0.433s/rec
ep0 step 430/1915 loss 0.499 0.433s/rec
ep0 step 440/1915 loss 0.692 0.433s/rec
ep0 step 450/1915 loss 0.368 0.434s/rec
ep0 step 460/1915 loss 0.634 0.433s/rec
ep0 step 470/1915 loss 0.530 0.434s/rec
ep0 step 480/1915 loss 0.645 0.432s/rec
ep0 step 490/1915 loss 0.732 0.434s/rec
ep0 step 500/1915 loss 0.587 0.433s/rec
ep0 step 510/1915 loss 0.709 0.433s/rec
ep0 step 520/1915 loss 0.538 0.433s/rec
ep0 step 530/1915 loss 0.515 0.433s/rec
ep0 step 540/1915 loss 0.506 0.434s/rec
ep0 step 550/1915 loss 0.647 0.434s/rec
ep0 step 560/1915 loss 0.432 0.435s/rec
ep0 step 570/1915 loss 0.632 0.435s/rec
ep0 step 580/1915 loss 0.580 0.435s/rec
ep0 step 590/1915 loss 0.573 0.436s/rec
ep0 step 600/1915 loss 0.594 0.436s/rec
ep0 step 610/1915 loss 0.821 0.436s/rec
ep0 step 620/1915 loss 0.518 0.438s/rec
ep0 step 630/1915 loss 0.462 0.438s/rec
ep0 step 640/1915 loss 0.497 0.438s/rec
ep0 step 650/1915 loss 0.557 0.436s/rec
ep0 step 660/1915 loss 0.583 0.437s/rec
ep0 step 670/1915 loss 0.518 0.436s/rec
ep0 step 680/1915 loss 0.430 0.436s/rec
ep0 step 690/1915 loss 0.665 0.436s/rec
ep0 step 700/1915 loss 0.532 0.437s/rec
ep0 step 710/1915 loss 0.535 0.438s/rec
ep0 step 720/1915 loss 0.457 0.440s/rec
ep0 step 730/1915 loss 0.541 0.439s/rec
ep0 step 740/1915 loss 0.510 0.440s/rec
ep0 step 750/1915 loss 0.509 0.440s/rec
ep0 step 760/1915 loss 0.568 0.439s/rec
ep0 step 770/1915 loss 0.599 0.440s/rec
ep0 step 780/1915 loss 0.599 0.441s/rec
ep0 step 790/1915 loss 0.505 0.440s/rec
ep0 step 800/1915 loss 0.499 0.441s/rec
ep0 step 810/1915 loss 0.498 0.440s/rec
ep0 step 820/1915 loss 0.629 0.440s/rec
ep0 step 830/1915 loss 0.600 0.440s/rec
ep0 step 840/1915 loss 0.426 0.440s/rec
ep0 step 850/1915 loss 0.592 0.441s/rec
ep0 step 860/1915 loss 0.593 0.440s/rec
ep0 step 870/1915 loss 0.516 0.441s/rec
ep0 step 880/1915 loss 0.521 0.441s/rec
ep0 step 890/1915 loss 0.693 0.441s/rec
ep0 step 900/1915 loss 0.477 0.441s/rec
ep0 step 910/1915 loss 0.465 0.440s/rec
ep0 step 920/1915 loss 0.530 0.441s/rec
ep0 step 930/1915 loss 0.481 0.441s/rec
ep0 step 940/1915 loss 0.570 0.441s/rec
ep0 step 950/1915 loss 0.491 0.442s/rec
ep0 step 960/1915 loss 0.502 0.442s/rec
ep0 step 970/1915 loss 0.557 0.442s/rec
ep0 step 980/1915 loss 0.436 0.442s/rec
ep0 step 990/1915 loss 0.554 0.442s/rec
ep0 step 1000/1915 loss 0.542 0.443s/rec
ep0 step 1010/1915 loss 0.392 0.443s/rec
ep0 step 1020/1915 loss 0.423 0.443s/rec
ep0 step 1030/1915 loss 0.652 0.443s/rec
ep0 step 1040/1915 loss 0.431 0.443s/rec
ep0 step 1050/1915 loss 0.660 0.442s/rec
ep0 step 1060/1915 loss 0.419 0.442s/rec
ep0 step 1070/1915 loss 0.520 0.442s/rec
ep0 step 1080/1915 loss 0.500 0.442s/rec
ep0 step 1090/1915 loss 0.707 0.441s/rec
ep0 step 1100/1915 loss 0.441 0.442s/rec
ep0 step 1110/1915 loss 0.614 0.442s/rec
ep0 step 1120/1915 loss 0.399 0.442s/rec
ep0 step 1130/1915 loss 0.457 0.442s/rec
ep0 step 1140/1915 loss 0.656 0.441s/rec
ep0 step 1150/1915 loss 0.497 0.441s/rec
ep0 step 1160/1915 loss 0.756 0.440s/rec
ep0 step 1170/1915 loss 0.612 0.441s/rec
ep0 step 1180/1915 loss 0.519 0.441s/rec
ep0 step 1190/1915 loss 0.592 0.441s/rec
ep0 step 1200/1915 loss 0.535 0.441s/rec
ep0 step 1210/1915 loss 0.590 0.442s/rec
ep0 step 1220/1915 loss 0.436 0.442s/rec
ep0 step 1230/1915 loss 0.427 0.442s/rec
ep0 step 1240/1915 loss 0.356 0.442s/rec
ep0 step 1250/1915 loss 0.522 0.443s/rec
ep0 step 1260/1915 loss 0.664 0.443s/rec
ep0 step 1270/1915 loss 0.490 0.443s/rec
ep0 step 1280/1915 loss 0.508 0.443s/rec
ep0 step 1290/1915 loss 0.388 0.443s/rec
ep0 step 1300/1915 loss 0.438 0.444s/rec
ep0 step 1310/1915 loss 0.580 0.444s/rec
ep0 step 1320/1915 loss 0.459 0.444s/rec
ep0 step 1330/1915 loss 0.619 0.444s/rec
ep0 step 1340/1915 loss 0.451 0.444s/rec
ep0 step 1350/1915 loss 0.633 0.444s/rec
ep0 step 1360/1915 loss 0.440 0.444s/rec
ep0 step 1370/1915 loss 0.524 0.444s/rec
ep0 step 1380/1915 loss 0.584 0.444s/rec
ep0 step 1390/1915 loss 0.543 0.444s/rec
ep0 step 1400/1915 loss 0.501 0.444s/rec
ep0 step 1410/1915 loss 0.370 0.443s/rec
ep0 step 1420/1915 loss 0.389 0.443s/rec
ep0 step 1430/1915 loss 0.466 0.442s/rec
ep0 step 1440/1915 loss 0.449 0.442s/rec
ep0 step 1450/1915 loss 0.385 0.442s/rec
ep0 step 1460/1915 loss 0.440 0.442s/rec
ep0 step 1470/1915 loss 0.394 0.442s/rec
ep0 step 1480/1915 loss 0.556 0.442s/rec
ep0 step 1490/1915 loss 0.600 0.442s/rec
ep0 step 1500/1915 loss 0.381 0.442s/rec
ep0 step 1510/1915 loss 0.602 0.442s/rec
ep0 step 1520/1915 loss 0.468 0.442s/rec
ep0 step 1530/1915 loss 0.510 0.443s/rec
ep0 step 1540/1915 loss 0.608 0.442s/rec
ep0 step 1550/1915 loss 0.496 0.442s/rec
ep0 step 1560/1915 loss 0.605 0.442s/rec
ep0 step 1570/1915 loss 0.384 0.442s/rec
ep0 step 1580/1915 loss 0.544 0.442s/rec
ep0 step 1590/1915 loss 0.307 0.442s/rec
ep0 step 1600/1915 loss 0.492 0.442s/rec
ep0 step 1610/1915 loss 0.592 0.442s/rec
ep0 step 1620/1915 loss 0.550 0.442s/rec
ep0 step 1630/1915 loss 0.466 0.441s/rec
ep0 step 1640/1915 loss 0.559 0.441s/rec
ep0 step 1650/1915 loss 0.413 0.441s/rec
ep0 step 1660/1915 loss 0.556 0.441s/rec
ep0 step 1670/1915 loss 0.609 0.441s/rec
ep0 step 1680/1915 loss 0.507 0.441s/rec
ep0 step 1690/1915 loss 0.457 0.442s/rec
ep0 step 1700/1915 loss 0.360 0.441s/rec
ep0 step 1710/1915 loss 0.457 0.442s/rec
ep0 step 1720/1915 loss 0.406 0.442s/rec
ep0 step 1730/1915 loss 0.492 0.442s/rec
ep0 step 1740/1915 loss 0.504 0.442s/rec
ep0 step 1750/1915 loss 0.425 0.442s/rec
ep0 step 1760/1915 loss 0.517 0.443s/rec
ep0 step 1770/1915 loss 0.533 0.442s/rec
ep0 step 1780/1915 loss 0.487 0.442s/rec
ep0 step 1790/1915 loss 0.400 0.441s/rec
ep0 step 1800/1915 loss 0.550 0.442s/rec
ep0 step 1810/1915 loss 0.378 0.442s/rec
ep0 step 1820/1915 loss 0.438 0.442s/rec
ep0 step 1830/1915 loss 0.434 0.442s/rec
ep0 step 1840/1915 loss 0.387 0.442s/rec
ep0 step 1850/1915 loss 0.491 0.442s/rec
ep0 step 1860/1915 loss 0.474 0.442s/rec
ep0 step 1870/1915 loss 0.439 0.442s/rec
ep0 step 1880/1915 loss 0.340 0.442s/rec
ep0 step 1890/1915 loss 0.418 0.442s/rec
ep0 step 1900/1915 loss 0.468 0.442s/rec
ep0 step 1910/1915 loss 0.389 0.442s/rec
saved /ckpt/skills-v1