ar-scaling-i4-96bfa5e-3fractions-9b9b1699
completed —
Measured behaviour
modelposition-marginal floor
Table view
| optimizer step | model | position-marginal floor |
|---|---|---|
| 300 | 3.6724 | 3.9319 |
| 600 | 3.8067 | 3.9319 |
| 900 | 4.1424 | 3.9319 |
| 1200 | 4.4191 | 3.9319 |
| 1500 | 4.8529 | 3.9319 |
| 1800 | 5.2190 | 3.9319 |
| 2100 | 5.6341 | 3.9319 |
| 2400 | 6.0474 | 3.9319 |
| 2700 | 6.3104 | 3.9319 |
| 300 | 3.6585 | 3.9230 |
| 600 | 3.7210 | 3.9230 |
| 900 | 3.9067 | 3.9230 |
| 1200 | 4.1615 | 3.9230 |
| 1500 | 4.4598 | 3.9230 |
| 1800 | 4.8273 | 3.9230 |
| 2100 | 5.1510 | 3.9230 |
| 2400 | 5.6098 | 3.9230 |
| 2700 | 5.6669 | 3.9230 |
| 300 | 3.6612 | 3.9290 |
| 600 | 3.6480 | 3.9290 |
| 900 | 3.8384 | 3.9290 |
| 1200 | 3.9849 | 3.9290 |
| 1500 | 4.2191 | 3.9290 |
| 1800 | 4.6009 | 3.9290 |
| 2100 | 4.7426 | 3.9290 |
| 2400 | 5.0071 | 3.9290 |
| 2700 | 5.2550 | 3.9290 |
| 3000 | 5.4942 | 3.9290 |
Measured result
Verbatim from the run's summary.json.
arms
epochs_to_best
7.164179104477612
fraction
0.25
held_out_nll
3.6724401549041907
icons
670
marginal_nll
3.9319226814163515
ratio_to_own_floor
0.9340061981028858
selected_step
300
steps_run
2700
trace
fraction
0.25
held_out_nll
3.6724401549041907
marginal_nll
3.9319226814163515
step
300
train_loss
3.285900115966797
fraction
0.25
held_out_nll
3.806695536452591
marginal_nll
3.9319226814163515
step
600
train_loss
2.9582679271698
fraction
0.25
held_out_nll
4.142442508999982
marginal_nll
3.9319226814163515
step
900
train_loss
2.364109516143799
fraction
0.25
held_out_nll
4.419114722451071
marginal_nll
3.9319226814163515
step
1200
train_loss
1.2945629358291626
fraction
0.25
held_out_nll
4.852858317483112
marginal_nll
3.9319226814163515
step
1500
train_loss
2.3625028133392334
fraction
0.25
held_out_nll
5.218995099835457
marginal_nll
3.9319226814163515
step
1800
train_loss
0.8343997001647949
… and 3 more
train_seconds
274.96233393100556
epochs_to_best
3.582089552238806
fraction
0.5
held_out_nll
3.6585002393237525
icons
1340
marginal_nll
3.922965558542167
ratio_to_own_floor
0.9325853578697505
selected_step
300
steps_run
2700
trace
fraction
0.5
held_out_nll
3.6585002393237525
marginal_nll
3.922965558542167
step
300
train_loss
3.2752957344055176
fraction
0.5
held_out_nll
3.7210268176327306
marginal_nll
3.922965558542167
step
600
train_loss
2.934887170791626
fraction
0.5
held_out_nll
3.9066621481955113
marginal_nll
3.922965558542167
step
900
train_loss
2.652989625930786
fraction
0.5
held_out_nll
4.161533230072122
marginal_nll
3.922965558542167
step
1200
train_loss
2.35439133644104
fraction
0.5
held_out_nll
4.459813070374794
marginal_nll
3.922965558542167
step
1500
train_loss
2.799227237701416
fraction
0.5
held_out_nll
4.827342667951118
marginal_nll
3.922965558542167
step
1800
train_loss
1.8788082599639893
… and 3 more
train_seconds
274.7395094890089
epochs_to_best
3.5807534502051475
fraction
1.0
held_out_nll
3.64796488139269
icons
2681
marginal_nll
3.9289882210903344
ratio_to_own_floor
0.9284743746012917
selected_step
600
steps_run
3000
trace
fraction
1.0
held_out_nll
3.6611640545050923
marginal_nll
3.9289882210903344
step
300
train_loss
3.2960057258605957
fraction
1.0
held_out_nll
3.64796488139269
marginal_nll
3.9289882210903344
step
600
train_loss
3.3200597763061523
fraction
1.0
held_out_nll
3.8384189372997466
marginal_nll
3.9289882210903344
step
900
train_loss
2.903935194015503
fraction
1.0
held_out_nll
3.984894900990683
marginal_nll
3.9289882210903344
step
1200
train_loss
1.603338360786438
fraction
1.0
held_out_nll
4.219101442814329
marginal_nll
3.9289882210903344
step
1500
train_loss
2.4615190029144287
fraction
1.0
held_out_nll
4.600927306373807
marginal_nll
3.9289882210903344
step
1800
train_loss
1.678338646888733
… and 4 more
train_seconds
305.2526577640092
checks
monotone_in_data
true
second_doubling_still_buys
true
config_sha256
d913579e4fa98215…criteria
min_second_doubling_share
0.25
deterministic_algorithms
true
device
cuda
gpu
NVIDIA GeForce RTX 4080
improvement_first_doubling
0.013939915580438278
improvement_second_doubling
0.010535357931062528
predeclared_outcome
data_limited
prepare_seconds
102.34559835400432
schema_version
1
second_doubling_share
0.7557691343444485
study_version
ar-scaling-i4
torch_version
2.14.0a0+4fdf77b940.nv26.08
validation_icons
339
State transitions
- planned2026-09-21T12:05:00Z
- completed2026-09-21T12:35:00Zthe predeclared criterion passes and does not support the decision it was written to inform
Run record
This run has no run.yaml. What follows is the identity and configuration carried by its rows in state/runs.jsonl, the append-only registry.
run_id
ar-scaling-i4-96bfa5e-3fractions-9b9b1699
config
configs/learning/ar-scaling-i4.yaml