dedup_zipfian

done · peak unseen_mismatch 91.7

Configuration

n_layer: 2
n_head: 4
n_embd: 256
base_seed: 42

Figures

Results

seen_* generated at each grid point's own training α · unseen_* = held-out pairings, uniform (α-independent) · 1000 pairs per condition

α	seen_match	seen_mismatch	unseen_match	unseen_mismatch
0.0	100.0 ± 0.0	22.8 ± 21.2	100.0 ± 0.0	22.1 ± 19.6
0.1	100.0 ± 0.0	21.0 ± 24.5	100.0 ± 0.0	20.9 ± 24.7
0.2	100.0 ± 0.0	4.6 ± 12.8	100.0 ± 0.0	4.6 ± 12.5
0.3	100.0 ± 0.0	10.8 ± 18.6	100.0 ± 0.0	10.6 ± 18.0
0.4	100.0 ± 0.0	9.0 ± 14.8	100.0 ± 0.0	9.3 ± 15.0
0.5	100.0 ± 0.0	43.2 ± 37.1	100.0 ± 0.0	43.8 ± 37.4
0.6	100.0 ± 0.0	13.7 ± 19.7	100.0 ± 0.0	13.9 ± 19.9
0.7	100.0 ± 0.0	9.6 ± 16.0	100.0 ± 0.0	9.4 ± 16.0
0.8	100.0 ± 0.0	28.1 ± 26.8	100.0 ± 0.0	27.4 ± 26.8
0.9	100.0 ± 0.0	33.7 ± 30.5	100.0 ± 0.0	33.4 ± 30.6
1.0	100.0 ± 0.0	44.7 ± 37.6	100.0 ± 0.0	44.0 ± 37.1
1.1	100.0 ± 0.0	59.2 ± 31.5	100.0 ± 0.0	58.8 ± 31.8
1.2	100.0 ± 0.0	65.8 ± 28.9	100.0 ± 0.0	64.0 ± 29.4
1.3	99.9 ± 0.1	91.4 ± 8.6	99.8 ± 0.4	87.6 ± 10.2
1.4	99.8 ± 0.4	91.4 ± 12.3	99.1 ± 1.4	86.1 ± 17.3
1.5	99.8 ± 0.2	96.8 ± 2.5	96.6 ± 3.2	90.1 ± 6.0
1.6	99.7 ± 0.2	98.4 ± 1.1	97.7 ± 2.1	91.7 ± 5.2
1.7	99.4 ± 0.4	98.1 ± 1.3	95.8 ± 2.1	89.2 ± 5.0
1.8	99.6 ± 0.5	98.9 ± 0.9	95.0 ± 2.2	88.2 ± 5.2
1.9	99.6 ± 0.3	97.5 ± 3.2	91.1 ± 2.6	80.8 ± 7.3
2.0	99.6 ± 0.3	98.1 ± 1.4	91.7 ± 3.0	80.9 ± 6.0
2.1	99.6 ± 0.1	99.0 ± 0.5	89.2 ± 2.1	82.4 ± 5.9
2.2	99.5 ± 0.4	99.2 ± 0.6	89.4 ± 2.6	81.7 ± 9.5
2.3	99.6 ± 0.3	99.0 ± 0.4	86.8 ± 3.4	77.7 ± 7.8
2.4	99.7 ± 0.2	99.1 ± 0.5	86.6 ± 2.6	76.2 ± 6.4
2.5	99.8 ± 0.1	99.5 ± 0.2	85.8 ± 3.6	78.1 ± 6.2
2.6	99.6 ± 0.2	99.4 ± 0.2	84.6 ± 3.3	74.8 ± 5.2
2.7	99.8 ± 0.1	99.5 ± 0.3	81.4 ± 3.6	73.0 ± 5.6
2.8	99.8 ± 0.1	99.6 ± 0.2	83.2 ± 3.4	75.8 ± 5.0
2.9	99.8 ± 0.2	99.7 ± 0.3	81.1 ± 3.6	72.1 ± 5.9
3.0	99.8 ± 0.2	99.8 ± 0.1	81.6 ± 2.7	71.9 ± 6.2

dedup_zipfian

Configuration

Figures

Results

Downloads