kev-gemma4-e2b / train.log
JohnP1's picture
Kev on Gemma 4 E2B, 1-epoch prototype (g4-e2b-e1-fp32/00-trial-0): dev 0.794, transfer 0.569, T=1.48
2836114 verified
Raw History Blame Contribute Delete
14.5 kB
config.json: 0%| | 0.00/4.91k [00:00<?, ?B/s] config.json: 100%|██████████| 4.91k/4.91k [00:00<00:00, 29.3MB/s]
tokenizer_config.json: 0%| | 0.00/906 [00:00<?, ?B/s] tokenizer_config.json: 100%|██████████| 906/906 [00:00<00:00, 4.75MB/s]
tokenizer.json: reconstructing file: 0%| | 0.00B / 32.2MB  tokenizer.json: downloading bytes: | 0.00B tokenizer.json: downloading bytes: ██████████| 8.75MB, 871kB/s tokenizer.json: downloading bytes: ██████████| 8.75MB, 871kB/s
tokenizer.json: reconstructing file: 100%|██████████| 32.2MB / 32.2MB, 3.20MB/s
model.safetensors: reconstructing file: 0%| | 0.00B / 10.2GB  model.safetensors: downloading bytes: | 0.00B
model.safetensors: reconstructing file: 2%|▏ | 202MB / 10.2GB, 16.4MB/s  model.safetensors: downloading bytes: | 119MB, 9.10MB/s
model.safetensors: reconstructing file: 5%|▌ | 554MB / 10.2GB, 46.4MB/s  model.safetensors: downloading bytes: ▎ | 284MB, 23.7MB/s model.safetensors: downloading bytes: ▉ | 944MB, 75.7MB/s
model.safetensors: reconstructing file: 11%|█ | 1.11GB / 10.2GB, 58.0MB/s  model.safetensors: downloading bytes: █▊ | 1.88GB, 150MB/s
model.safetensors: reconstructing file: 23%|██▎ | 2.39GB / 10.2GB, 182MB/s  model.safetensors: downloading bytes: ██▊ | 2.94GB, 228MB/s
model.safetensors: reconstructing file: 35%|███▌ | 3.59GB / 10.2GB, 272MB/s  model.safetensors: downloading bytes: ████ | 4.13GB, 308MB/s
model.safetensors: reconstructing file: 46%|████▌ | 4.67GB / 10.2GB, 345MB/s  model.safetensors: downloading bytes: █████ | 5.22GB, 374MB/s model.safetensors: downloading bytes: ██████ | 6.21GB, 424MB/s
model.safetensors: reconstructing file: 57%|█████▋ | 5.81GB / 10.2GB, 350MB/s  model.safetensors: downloading bytes: ██████▉ | 7.16GB, 364MB/s
model.safetensors: reconstructing file: 64%|██████▍ | 6.55GB / 10.2GB, 304MB/s  model.safetensors: downloading bytes: ███████▋ | 7.92GB, 310MB/s
model.safetensors: reconstructing file: 71%|███████ | 7.27GB / 10.2GB, 283MB/s  model.safetensors: downloading bytes: ████████▎ | 8.46GB, 293MB/s
model.safetensors: reconstructing file: 77%|███████▋ | 7.90GB / 10.2GB, 270MB/s 
model.safetensors: reconstructing file: 81%|████████ | 8.29GB / 10.2GB, 261MB/s 
model.safetensors: reconstructing file: 85%|████████▌ | 8.76GB / 10.2GB, 255MB/s 
model.safetensors: reconstructing file: 89%|████████▉ | 9.12GB / 10.2GB, 251MB/s 
model.safetensors: reconstructing file: 93%|█████████▎| 9.48GB / 10.2GB, 246MB/s 
model.safetensors: reconstructing file: 97%|█████████▋| 9.92GB / 10.2GB, 243MB/s 
model.safetensors: reconstructing file: 99%|█████████▉| 10.2GB / 10.2GB, 238MB/s  model.safetensors: downloading bytes: ██████████| 8.66GB, 272MB/s model.safetensors: downloading bytes: ██████████| 8.66GB, 272MB/s
model.safetensors: reconstructing file: 100%|██████████| 10.2GB / 10.2GB, 238MB/s
Loading weights: 0%| | 0/1951 [00:00<?, ?it/s] Loading weights: 100%|██████████| 1951/1951 [00:00<00:00, 6084.11it/s]
generation_config.json: 0%| | 0.00/181 [00:00<?, ?B/s] generation_config.json: 100%|██████████| 181/181 [00:00<00:00, 904kB/s]
device=cuda trainable params=24.9M
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, 789kB/s v7/decision-v7/train.jsonl: downloading bytes: ██████████| 8.02MB, 789kB/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@a88f56db53
dropped 70 of 12576 records that exceed the training context (384 state / 960 branch / 2048 packed tokens)
12506 training requests (holdout=[]), questions by type {'score': 3448, 'noul': 5224, 'choice': 6834}
ep0 step 10/1564 loss 2.494 kl 0.000 anchor 0.000 0.444s/rec
ep0 step 20/1564 loss 2.089 kl 0.000 anchor 0.000 0.383s/rec
ep0 step 30/1564 loss 1.886 kl 0.000 anchor 0.000 0.358s/rec
ep0 step 40/1564 loss 2.217 kl 0.000 anchor 0.000 0.383s/rec
ep0 step 50/1564 loss 1.778 kl 0.000 anchor 0.000 0.382s/rec
ep0 step 60/1564 loss 1.879 kl 0.000 anchor 0.000 0.377s/rec
ep0 step 70/1564 loss 1.540 kl 0.000 anchor 0.000 0.356s/rec
ep0 step 80/1564 loss 1.700 kl 0.000 anchor 0.000 0.363s/rec
ep0 step 90/1564 loss 1.608 kl 0.000 anchor 0.000 0.357s/rec
ep0 step 100/1564 loss 1.627 kl 0.000 anchor 0.000 0.355s/rec
ep0 step 110/1564 loss 1.473 kl 0.000 anchor 0.000 0.348s/rec
ep0 step 120/1564 loss 1.457 kl 0.000 anchor 0.000 0.347s/rec
ep0 step 130/1564 loss 1.059 kl 0.000 anchor 0.000 0.350s/rec
ep0 step 140/1564 loss 1.110 kl 0.000 anchor 0.000 0.352s/rec
ep0 step 150/1564 loss 1.593 kl 0.000 anchor 0.000 0.350s/rec
ep0 step 160/1564 loss 0.870 kl 0.000 anchor 0.000 0.346s/rec
ep0 step 170/1564 loss 1.391 kl 0.000 anchor 0.000 0.346s/rec
ep0 step 180/1564 loss 1.205 kl 0.000 anchor 0.000 0.349s/rec
ep0 step 190/1564 loss 0.786 kl 0.000 anchor 0.000 0.345s/rec
ep0 step 200/1564 loss 0.964 kl 0.000 anchor 0.000 0.338s/rec
ep0 step 210/1564 loss 1.072 kl 0.000 anchor 0.000 0.334s/rec
ep0 step 220/1564 loss 0.788 kl 0.000 anchor 0.000 0.338s/rec
ep0 step 230/1564 loss 1.440 kl 0.000 anchor 0.000 0.344s/rec
ep0 step 240/1564 loss 1.064 kl 0.000 anchor 0.000 0.342s/rec
ep0 step 250/1564 loss 0.909 kl 0.000 anchor 0.000 0.341s/rec
ep0 step 260/1564 loss 0.880 kl 0.000 anchor 0.000 0.341s/rec
ep0 step 270/1564 loss 0.853 kl 0.000 anchor 0.000 0.342s/rec
ep0 step 280/1564 loss 0.939 kl 0.000 anchor 0.000 0.344s/rec
ep0 step 290/1564 loss 0.841 kl 0.000 anchor 0.000 0.343s/rec
ep0 step 300/1564 loss 0.791 kl 0.000 anchor 0.000 0.345s/rec
ep0 step 310/1564 loss 1.095 kl 0.000 anchor 0.000 0.346s/rec
ep0 step 320/1564 loss 1.164 kl 0.000 anchor 0.000 0.348s/rec
ep0 step 330/1564 loss 0.823 kl 0.000 anchor 0.000 0.344s/rec
ep0 step 340/1564 loss 0.710 kl 0.000 anchor 0.000 0.343s/rec
ep0 step 350/1564 loss 0.876 kl 0.000 anchor 0.000 0.343s/rec
ep0 step 360/1564 loss 0.983 kl 0.000 anchor 0.000 0.341s/rec
ep0 step 370/1564 loss 0.673 kl 0.000 anchor 0.000 0.337s/rec
ep0 step 380/1564 loss 1.054 kl 0.000 anchor 0.000 0.335s/rec
ep0 step 390/1564 loss 0.603 kl 0.000 anchor 0.000 0.335s/rec
ep0 step 400/1564 loss 0.934 kl 0.000 anchor 0.000 0.334s/rec
ep0 step 410/1564 loss 0.793 kl 0.000 anchor 0.000 0.332s/rec
ep0 step 420/1564 loss 0.735 kl 0.000 anchor 0.000 0.331s/rec
ep0 step 430/1564 loss 0.823 kl 0.000 anchor 0.000 0.334s/rec
ep0 step 440/1564 loss 0.741 kl 0.000 anchor 0.000 0.334s/rec
ep0 step 450/1564 loss 0.651 kl 0.000 anchor 0.000 0.334s/rec
ep0 step 460/1564 loss 0.821 kl 0.000 anchor 0.000 0.333s/rec
ep0 step 470/1564 loss 0.634 kl 0.000 anchor 0.000 0.331s/rec
ep0 step 480/1564 loss 0.620 kl 0.000 anchor 0.000 0.330s/rec
ep0 step 490/1564 loss 0.596 kl 0.000 anchor 0.000 0.331s/rec
ep0 step 500/1564 loss 1.017 kl 0.000 anchor 0.000 0.328s/rec
ep0 step 510/1564 loss 0.903 kl 0.000 anchor 0.000 0.328s/rec
ep0 step 520/1564 loss 0.672 kl 0.000 anchor 0.000 0.327s/rec
ep0 step 530/1564 loss 0.966 kl 0.000 anchor 0.000 0.326s/rec
ep0 step 540/1564 loss 0.609 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 550/1564 loss 0.519 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 560/1564 loss 0.658 kl 0.000 anchor 0.000 0.322s/rec
ep0 step 570/1564 loss 0.776 kl 0.000 anchor 0.000 0.322s/rec
ep0 step 580/1564 loss 0.752 kl 0.000 anchor 0.000 0.321s/rec
ep0 step 590/1564 loss 0.576 kl 0.000 anchor 0.000 0.321s/rec
ep0 step 600/1564 loss 0.713 kl 0.000 anchor 0.000 0.321s/rec
ep0 step 610/1564 loss 0.660 kl 0.000 anchor 0.000 0.320s/rec
ep0 step 620/1564 loss 0.724 kl 0.000 anchor 0.000 0.321s/rec
ep0 step 630/1564 loss 0.746 kl 0.000 anchor 0.000 0.323s/rec
ep0 step 640/1564 loss 0.483 kl 0.000 anchor 0.000 0.323s/rec
ep0 step 650/1564 loss 0.813 kl 0.000 anchor 0.000 0.323s/rec
ep0 step 660/1564 loss 0.664 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 670/1564 loss 0.578 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 680/1564 loss 0.602 kl 0.000 anchor 0.000 0.325s/rec
ep0 step 690/1564 loss 0.622 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 700/1564 loss 0.738 kl 0.000 anchor 0.000 0.325s/rec
ep0 step 710/1564 loss 0.598 kl 0.000 anchor 0.000 0.326s/rec
ep0 step 720/1564 loss 0.842 kl 0.000 anchor 0.000 0.326s/rec
ep0 step 730/1564 loss 0.680 kl 0.000 anchor 0.000 0.329s/rec
ep0 step 740/1564 loss 0.602 kl 0.000 anchor 0.000 0.330s/rec
ep0 step 750/1564 loss 0.381 kl 0.000 anchor 0.000 0.329s/rec
ep0 step 760/1564 loss 0.818 kl 0.000 anchor 0.000 0.329s/rec
ep0 step 770/1564 loss 0.586 kl 0.000 anchor 0.000 0.328s/rec
ep0 step 780/1564 loss 0.696 kl 0.000 anchor 0.000 0.327s/rec
ep0 step 790/1564 loss 0.460 kl 0.000 anchor 0.000 0.327s/rec
ep0 step 800/1564 loss 0.778 kl 0.000 anchor 0.000 0.329s/rec
ep0 step 810/1564 loss 0.440 kl 0.000 anchor 0.000 0.328s/rec
ep0 step 820/1564 loss 0.575 kl 0.000 anchor 0.000 0.329s/rec
ep0 step 830/1564 loss 0.609 kl 0.000 anchor 0.000 0.328s/rec
ep0 step 840/1564 loss 0.666 kl 0.000 anchor 0.000 0.329s/rec
ep0 step 850/1564 loss 0.605 kl 0.000 anchor 0.000 0.328s/rec
ep0 step 860/1564 loss 0.432 kl 0.000 anchor 0.000 0.327s/rec
ep0 step 870/1564 loss 0.615 kl 0.000 anchor 0.000 0.327s/rec
ep0 step 880/1564 loss 0.750 kl 0.000 anchor 0.000 0.328s/rec
ep0 step 890/1564 loss 0.348 kl 0.000 anchor 0.000 0.328s/rec
ep0 step 900/1564 loss 0.652 kl 0.000 anchor 0.000 0.328s/rec
ep0 step 910/1564 loss 0.504 kl 0.000 anchor 0.000 0.328s/rec
ep0 step 920/1564 loss 0.657 kl 0.000 anchor 0.000 0.328s/rec
ep0 step 930/1564 loss 0.748 kl 0.000 anchor 0.000 0.329s/rec
ep0 step 940/1564 loss 0.553 kl 0.000 anchor 0.000 0.330s/rec
ep0 step 950/1564 loss 0.634 kl 0.000 anchor 0.000 0.328s/rec
ep0 step 960/1564 loss 0.499 kl 0.000 anchor 0.000 0.328s/rec
ep0 step 970/1564 loss 0.509 kl 0.000 anchor 0.000 0.327s/rec
ep0 step 980/1564 loss 0.697 kl 0.000 anchor 0.000 0.326s/rec
ep0 step 990/1564 loss 0.688 kl 0.000 anchor 0.000 0.325s/rec
ep0 step 1000/1564 loss 0.660 kl 0.000 anchor 0.000 0.325s/rec
ep0 step 1010/1564 loss 0.695 kl 0.000 anchor 0.000 0.326s/rec
ep0 step 1020/1564 loss 0.691 kl 0.000 anchor 0.000 0.327s/rec
ep0 step 1030/1564 loss 0.569 kl 0.000 anchor 0.000 0.326s/rec
ep0 step 1040/1564 loss 0.532 kl 0.000 anchor 0.000 0.325s/rec
ep0 step 1050/1564 loss 0.553 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1060/1564 loss 0.766 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1070/1564 loss 0.463 kl 0.000 anchor 0.000 0.323s/rec
ep0 step 1080/1564 loss 0.436 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1090/1564 loss 0.715 kl 0.000 anchor 0.000 0.323s/rec
ep0 step 1100/1564 loss 0.501 kl 0.000 anchor 0.000 0.322s/rec
ep0 step 1110/1564 loss 0.495 kl 0.000 anchor 0.000 0.322s/rec
ep0 step 1120/1564 loss 0.607 kl 0.000 anchor 0.000 0.322s/rec
ep0 step 1130/1564 loss 0.642 kl 0.000 anchor 0.000 0.323s/rec
ep0 step 1140/1564 loss 0.716 kl 0.000 anchor 0.000 0.323s/rec
ep0 step 1150/1564 loss 0.327 kl 0.000 anchor 0.000 0.323s/rec
ep0 step 1160/1564 loss 0.847 kl 0.000 anchor 0.000 0.323s/rec
ep0 step 1170/1564 loss 0.885 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1180/1564 loss 0.411 kl 0.000 anchor 0.000 0.323s/rec
ep0 step 1190/1564 loss 0.498 kl 0.000 anchor 0.000 0.323s/rec
ep0 step 1200/1564 loss 0.659 kl 0.000 anchor 0.000 0.323s/rec
ep0 step 1210/1564 loss 0.316 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1220/1564 loss 0.395 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1230/1564 loss 0.505 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1240/1564 loss 0.628 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1250/1564 loss 0.645 kl 0.000 anchor 0.000 0.325s/rec
ep0 step 1260/1564 loss 0.628 kl 0.000 anchor 0.000 0.325s/rec
ep0 step 1270/1564 loss 0.576 kl 0.000 anchor 0.000 0.325s/rec
ep0 step 1280/1564 loss 0.712 kl 0.000 anchor 0.000 0.326s/rec
ep0 step 1290/1564 loss 0.604 kl 0.000 anchor 0.000 0.325s/rec
ep0 step 1300/1564 loss 0.398 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1310/1564 loss 0.543 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1320/1564 loss 0.427 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1330/1564 loss 0.734 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1340/1564 loss 0.468 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1350/1564 loss 0.377 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1360/1564 loss 0.454 kl 0.000 anchor 0.000 0.325s/rec
ep0 step 1370/1564 loss 0.639 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1380/1564 loss 0.494 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1390/1564 loss 0.553 kl 0.000 anchor 0.000 0.324s/rec
ep0 step 1400/1564 loss 0.371 kl 0.000 anchor 0.000 0.323s/rec
ep0 step 1410/1564 loss 0.453 kl 0.000 anchor 0.000 0.323s/rec
ep0 step 1420/1564 loss 0.393 kl 0.000 anchor 0.000 0.322s/rec
ep0 step 1430/1564 loss 0.590 kl 0.000 anchor 0.000 0.322s/rec
ep0 step 1440/1564 loss 0.457 kl 0.000 anchor 0.000 0.322s/rec
ep0 step 1450/1564 loss 0.762 kl 0.000 anchor 0.000 0.322s/rec
ep0 step 1460/1564 loss 0.683 kl 0.000 anchor 0.000 0.321s/rec
ep0 step 1470/1564 loss 0.459 kl 0.000 anchor 0.000 0.321s/rec
ep0 step 1480/1564 loss 0.696 kl 0.000 anchor 0.000 0.321s/rec
ep0 step 1490/1564 loss 0.246 kl 0.000 anchor 0.000 0.321s/rec
ep0 step 1500/1564 loss 0.592 kl 0.000 anchor 0.000 0.320s/rec
ep0 step 1510/1564 loss 0.554 kl 0.000 anchor 0.000 0.320s/rec
ep0 step 1520/1564 loss 0.434 kl 0.000 anchor 0.000 0.320s/rec
ep0 step 1530/1564 loss 0.455 kl 0.000 anchor 0.000 0.320s/rec
ep0 step 1540/1564 loss 0.457 kl 0.000 anchor 0.000 0.320s/rec
ep0 step 1550/1564 loss 0.505 kl 0.000 anchor 0.000 0.321s/rec
ep0 step 1560/1564 loss 0.613 kl 0.000 anchor 0.000 0.322s/rec
saved runs/g4-e2b-e1