tinystories-diff-ar / log_refill.txt
Asilarkness's picture
Super-squash branch 'main' using huggingface_hub
a6c51a2
Raw History Blame Contribute Delete
4.85 kB
launch refill
01:28:56 loaded d1 step=72000 seen=1179.6M
01:28:59 loaded ar step=65000 seen=1065.0M
01:29:08 === baselines ===
01:29:10 AR (ceiling) calls=128 0.3469s 23617 tok/s nll=2.8144 dup=0.0177 d2=0.5155
01:29:10 DIFF K=2 calls=2 0.0159s 515364 tok/s nll=8.4220 dup=0.0495 d2=0.5907
01:29:10 DIFF K=4 calls=4 0.0319s 256512 tok/s nll=5.8584 dup=0.0586 d2=0.4783
01:29:11 DIFF K=6 calls=6 0.0472s 173597 tok/s nll=4.8534 dup=0.0607 d2=0.4547
01:29:11 DIFF K=8 calls=8 0.0625s 131150 tok/s nll=4.3593 dup=0.0535 d2=0.4401
01:29:12 DIFF K=10 calls=10 0.0783s 104655 tok/s nll=4.1867 dup=0.0544 d2=0.4279
01:29:12 DIFF K=12 calls=12 0.0940s 87121 tok/s nll=3.8919 dup=0.0528 d2=0.4172
01:29:13 DIFF K=16 calls=16 0.1246s 65766 tok/s nll=3.5531 dup=0.0460 d2=0.4199
01:29:13 DIFF K=20 calls=20 0.1564s 52366 tok/s nll=3.2530 dup=0.0433 d2=0.4119
01:29:14 DIFF K=24 calls=24 0.1866s 43903 tok/s nll=2.9899 dup=0.0392 d2=0.3936
01:29:14 === diffusion + refill, matched call budgets ===
01:29:14 K=4 p=0.10 R=1 calls=5 0.0370s 221113 tok/s nll=4.9766 dup=0.0582 d2=0.4325
01:29:15 K=4 p=0.10 R=2 calls=6 0.0428s 191548 tok/s nll=4.7899 dup=0.0594 d2=0.4222
01:29:15 K=4 p=0.10 R=4 calls=8 0.0543s 150894 tok/s nll=4.3275 dup=0.0479 d2=0.4278
01:29:15 K=4 p=0.20 R=1 calls=5 0.0373s 219508 tok/s nll=4.9760 dup=0.0571 d2=0.4402
01:29:16 K=4 p=0.20 R=2 calls=6 0.0433s 189256 tok/s nll=4.3944 dup=0.0486 d2=0.4386
01:29:16 K=4 p=0.20 R=4 calls=8 0.0549s 149345 tok/s nll=3.6246 dup=0.0422 d2=0.4466
01:29:16 K=4 p=0.30 R=1 calls=5 0.0374s 219003 tok/s nll=4.7735 dup=0.0570 d2=0.4427
01:29:17 K=4 p=0.30 R=2 calls=6 0.0435s 188193 tok/s nll=4.1703 dup=0.0420 d2=0.4502
01:29:17 K=4 p=0.30 R=4 calls=8 0.0559s 146419 tok/s nll=3.5979 dup=0.0402 d2=0.4345
01:29:17 K=4 p=0.50 R=1 calls=5 0.0379s 216238 tok/s nll=5.1200 dup=0.0460 d2=0.4922
01:29:18 K=4 p=0.50 R=2 calls=6 0.0446s 183568 tok/s nll=4.7294 dup=0.0399 d2=0.5059
01:29:18 K=4 p=0.50 R=4 calls=8 0.0577s 142003 tok/s nll=4.9303 dup=0.0440 d2=0.5007
01:29:19 K=8 p=0.10 R=1 calls=9 0.0682s 120178 tok/s nll=3.9082 dup=0.0519 d2=0.4197
01:29:19 K=8 p=0.10 R=2 calls=10 0.0740s 110647 tok/s nll=3.7144 dup=0.0493 d2=0.4158
01:29:19 K=8 p=0.10 R=4 calls=12 0.0853s 96021 tok/s nll=3.3582 dup=0.0415 d2=0.4229
01:29:20 K=8 p=0.20 R=1 calls=9 0.0686s 119364 tok/s nll=3.8277 dup=0.0490 d2=0.4280
01:29:20 K=8 p=0.20 R=2 calls=10 0.0743s 110238 tok/s nll=3.4856 dup=0.0395 d2=0.4192
01:29:21 K=8 p=0.20 R=4 calls=12 0.0857s 95562 tok/s nll=3.0254 dup=0.0352 d2=0.4211
01:29:21 K=8 p=0.30 R=1 calls=9 0.0686s 119427 tok/s nll=4.0013 dup=0.0464 d2=0.4438
01:29:22 K=8 p=0.30 R=2 calls=10 0.0744s 110039 tok/s nll=3.4876 dup=0.0378 d2=0.4374
01:29:22 K=8 p=0.30 R=4 calls=12 0.0866s 94619 tok/s nll=3.3252 dup=0.0338 d2=0.4433
01:29:22 K=8 p=0.50 R=1 calls=9 0.0693s 118288 tok/s nll=4.0792 dup=0.0436 d2=0.4488
01:29:23 K=8 p=0.50 R=2 calls=10 0.0760s 107800 tok/s nll=4.3721 dup=0.0381 d2=0.4774
01:29:23 K=8 p=0.50 R=4 calls=12 0.0899s 91114 tok/s nll=4.6417 dup=0.0426 d2=0.4953
01:29:24 K=12 p=0.10 R=1 calls=13 0.0993s 82532 tok/s nll=3.6624 dup=0.0503 d2=0.4170
01:29:24 K=12 p=0.10 R=2 calls=14 0.1058s 77422 tok/s nll=3.3433 dup=0.0391 d2=0.4181
01:29:25 K=12 p=0.10 R=4 calls=16 0.1154s 70984 tok/s nll=3.0165 dup=0.0351 d2=0.4262
01:29:25 K=12 p=0.20 R=1 calls=13 0.0993s 82501 tok/s nll=3.3954 dup=0.0423 d2=0.4222
01:29:26 K=12 p=0.20 R=2 calls=14 0.1051s 77928 tok/s nll=3.2046 dup=0.0383 d2=0.4235
01:29:27 K=12 p=0.20 R=4 calls=16 0.1173s 69812 tok/s nll=2.7801 dup=0.0315 d2=0.4131
01:29:27 K=12 p=0.30 R=1 calls=13 0.1001s 81803 tok/s nll=3.4976 dup=0.0386 d2=0.4461
01:29:28 K=12 p=0.30 R=2 calls=14 0.1054s 77752 tok/s nll=3.1743 dup=0.0372 d2=0.4304
01:29:28 K=12 p=0.30 R=4 calls=16 0.1174s 69781 tok/s nll=3.1914 dup=0.0283 d2=0.4381
01:29:29 K=12 p=0.50 R=1 calls=13 0.0999s 82015 tok/s nll=3.8986 dup=0.0328 d2=0.4703
01:29:29 K=12 p=0.50 R=2 calls=14 0.1069s 76610 tok/s nll=4.1343 dup=0.0346 d2=0.4808
01:29:30 K=12 p=0.50 R=4 calls=16 0.1196s 68478 tok/s nll=4.5074 dup=0.0343 d2=0.5069
01:29:30 REAL TEXT nll=0.9623 dup=0.0156 d2=0.5253