Files
rzen ba73c8a7e8 Phase 4: v1 training run results (val loss 1.52, ppl ~4.6)
- reports/loss-v1.csv: full train/val loss log over 20k steps
- reports/loss-v1.txt: ASCII loss curve (steep descent then flattening tail)
Training ran ~58 min on M1 MPS at 5.7 it/s. No overfitting (train~=val).
2026-07-12 12:35:47 -04:00

1.2 KiB

1itertrain_lossval_losslr
25002.82852.81710.000600
310002.33112.32670.000598
415002.11352.10260.000594
520002.00132.00020.000589
625001.92531.92300.000582
730001.88111.87520.000574
835001.83991.84290.000564
940001.81191.81280.000552
1045001.77751.79050.000540
1150001.75891.76160.000525
1255001.74661.74400.000510
1360001.72481.72610.000494
1465001.70381.70930.000476
1570001.69801.70230.000458
1675001.68291.68900.000438
1780001.67471.67960.000418
1885001.65721.66730.000398
1990001.65061.64520.000377
2095001.64081.63190.000356
21100001.62901.63070.000334
22105001.62021.62150.000313
23110001.61261.61660.000292
24115001.61321.60140.000271
25120001.59621.59030.000250
26125001.58511.57410.000230
27130001.58071.58640.000210
28135001.56661.57540.000191
29140001.56931.56630.000173
30145001.55641.56730.000156
31150001.55711.55290.000141
32155001.54151.54440.000126
33160001.54411.53880.000113
34165001.53961.53560.000101
35170001.53601.52700.000090
36175001.53051.52780.000081
37180001.52811.52630.000073
38185001.51921.52440.000068
39190001.51911.52010.000063
40195001.51221.51980.000061