Across FineWeb-Edu tiers, the paired validation-loss benefit of the sink logit shrinks with scale, while the additional benefit of a gate when combined with the sink logit grows overall; combo2 has the lowest mean validation loss at every tested tier.

Across FineWeb-Edu tiers, the paired validation-loss benefit of the sink logit shrinks with scale, while the additional benefit of a gate when combined with the sink logit grows overall; combo2 has the lowest mean validation loss at every tested tier. · CiteArk

Measurement results

0/69 assessed

0 supported0 disputed or conflicting69 not assessed

The complete result group stays with its claim. Assessed rows come first; their judgments do not replace the whole-claim conclusion. Expand a row for conditions, sources, and evidence history.

Each row is a measurement with its reported value, latest available observation, and cumulative evidence state.
Metric and conditions	Reported	Latest observation	Evidence judgment
arm: baseline · tier: 10M	4.2119 nats	—	Not assessed
arm: baseline · tier: 50M	3.5399 nats	—	Not assessed
arm: baseline · tier: 124M	3.371 nats	—	Not assessed
arm: baseline · tier: 350M	3.0338 nats	—	Not assessed
arm: off-by-one · tier: 10M · comparison: baseline	-0.016 nats	—	Not assessed
arm: off-by-one · tier: 10M · comparison: baseline	0.0027 nats	—	Not assessed
arm: sink token · tier: 10M · comparison: baseline	-0.0025 nats	—	Not assessed
arm: sink token · tier: 10M · comparison: baseline	0.0041 nats	—	Not assessed
arm: sink logit · tier: 10M · comparison: baseline	-0.0185 nats	—	Not assessed
arm: sink logit · tier: 10M · comparison: baseline	0.0033 nats	—	Not assessed
arm: sink logit · tier: 50M · comparison: baseline	-0.0068 nats	—	Not assessed
arm: sink logit · tier: 50M · comparison: baseline	0.0003 nats	—	Not assessed

Across FineWeb-Edu tiers, the paired validation-loss benefit of the sink logit shrinks with scale, while the additional benefit of a gate when combined with the sink logit grows overall; combo2 has the lowest mean validation loss at every tested tier.

Reported and observed measurements

Assessments (0)

Runs (0)

Measurement results