DiffuRefill-1B / pilots /objective.log
Asilarkness's picture
Super-squash branch 'main' using huggingface_hub
a02b918
Raw History Blame Contribute Delete
1.34 kB
03:15:29 === objectives: base / span / selfrepair / both ===
03:19:16 base 3000 loss=4.3888 (219s)
03:23:08 base 6000 loss=3.4443 (451s)
03:23:14 base t=0.5 K=4 tok=0.5338 pair=0.3032 K=8 tok=0.5358 pair=0.3135 K=16 tok=0.5335 pair=0.3147
03:23:14 base t=0.9 K=4 tok=0.1697 pair=0.0663 K=8 tok=0.1623 pair=0.0693 K=16 tok=0.1577 pair=0.0707
03:27:00 span 3000 loss=4.4491 (226s)
03:30:34 span 6000 loss=3.4527 (439s)
03:30:39 span t=0.5 K=4 tok=0.5325 pair=0.3069 K=8 tok=0.5362 pair=0.3198 K=16 tok=0.5341 pair=0.3214
03:30:40 span t=0.9 K=4 tok=0.1712 pair=0.0683 K=8 tok=0.1650 pair=0.0716 K=16 tok=0.1620 pair=0.0731
03:34:22 selfrep 3000 loss=4.2847 (222s)
03:37:58 selfrep 6000 loss=3.2314 (438s)
03:38:03 selfrep t=0.5 K=4 tok=0.5302 pair=0.3011 K=8 tok=0.5319 pair=0.3084 K=16 tok=0.5331 pair=0.3183
03:38:04 selfrep t=0.9 K=4 tok=0.1640 pair=0.0625 K=8 tok=0.1595 pair=0.0668 K=16 tok=0.1543 pair=0.0672
03:41:50 both 3000 loss=4.3134 (226s)
03:45:33 both 6000 loss=3.2216 (448s)
03:45:38 both t=0.5 K=4 tok=0.5345 pair=0.3094 K=8 tok=0.5367 pair=0.3201 K=16 tok=0.5342 pair=0.3242
03:45:39 both t=0.9 K=4 tok=0.1688 pair=0.0652 K=8 tok=0.1592 pair=0.0670 K=16 tok=0.1572 pair=0.0705
03:45:39 published pilots/objective.json