Dataset Viewer
Auto-converted to Parquet Duplicate
sample_id
int64
0
999
task_type
stringclasses
4 values
sequence_length
int64
256
4.1k
state_dimension
int64
8
32
mean_delta_gate
float64
0.6
1.15
selective_sparsity_ratio
float64
0.75
0.92
mamba_latency_ms
float64
2.56
59.2
transformer_latency_ms
float64
4
631
speedup_factor
float64
0.88
11
vram_reduction_factor
float64
4.7
37.7
task_accuracy
float64
0.97
1
initial_token_entropy
float64
3.21
5.8
state_norm_final
float64
1.2
4.5
token_activation_profile
listlengths
32
32
0
selective_associative_recall
2,048
32
0.7313
0.9035
29.61
161.43
5.45
20.1
0.9784
3.879
2.561
[ 0.2946, 0.7581, 0.1516, 0.0452, 0.0915, 0.2679, 0.0213, 0.313, 0.0894, 0.3117, 0.0638, 0.6058, 0.2338, 0.7534, 0.7561, 0.2045, 0.1089, 0.0089, 0.1515, 0.2582, 0.21, 0.0026, 0.2537, 0.1272, 0.3622, 0.6554, 0.2885, 0.3976, 0.101, 0.0369, 0.0783, 0.3892 ]
1
variable_length_copying
512
32
0.8883
0.8261
7.04
11.5
1.63
6.9
0.988
5.291
1.839
[ 0.5923, 0.2334, 0.813, 0.0443, 0.2985, 0.0967, 0.1906, 0.238, 0.0775, 0.1519, 0.1323, 0.3787, 0.0554, 0.126, 0.3826, 0.4958, 0.1442, 0.0952, 0.0273, 0.069, 0.8629, 0.3262, 0.2162, 0.2476, 0.1832, 0.2655, 1.8957, 0.2389, 0.5419, 0.1713, 0.9713, 0.0068 ]
2
induction_heads_induction
1,024
8
0.845
0.9044
14.01
41.02
2.93
11.3
0.993
3.914
3.499
[ 0.794, 0.1893, 0.9218, 0.0563, 0.3409, 0.0773, 0.7651, 0.1595, 0.0758, 0.5192, 0.4985, 0.191, 0.4428, 0.151, 0.3408, 0.4303, 0.1653, 0.7864, 0.8381, 0.7204, 0.19, 0.5094, 0.8652, 0.3129, 0.0074, 0.5797, 0.91, 0.1416, 0.2607, 0.886, 0.1098, 0.1327 ]
3
noise_filtering_retrieval
4,096
16
0.9517
0.8435
57.82
627.37
10.85
37.7
0.9925
3.805
4.106
[ 0.1002, 0.3793, 0.3575, 0.1473, 0.4318, 0.0627, 0.71, 0.4732, 0.0773, 0.5873, 0.0142, 0.7246, 0.025, 1.2656, 0.1042, 0.0142, 0.631, 0.8829, 0.101, 1.1266, 0.333, 0.1011, 0.3876, 0.3356, 0.3565, 0.3193, 0.629, 0.4228, 0.0024, 0.4182, 0.0365, 0.6461 ]
4
selective_associative_recall
512
16
0.8578
0.8372
6.9
11.81
1.71
6.9
0.9793
4.057
2.607
[ 0.2289, 0.2085, 0.0427, 0.5487, 1.1776, 0.253, 0.1018, 0.4624, 0.2872, 0.0187, 0.0984, 0.371, 0.5042, 0.759, 0.0395, 0.2172, 0.2554, 0.0256, 0.1677, 0.4026, 0.0705, 0.5689, 0.2985, 0.15, 0.0094, 0.4748, 0.3022, 0.0721, 0.159, 0.8102, 0.1589, 0.7645 ]
5
variable_length_copying
256
32
0.7112
0.829
3.04
4
1.32
4.7
0.965
5.01
3.506
[ 0.3763, 0.3559, 0.0179, 0.2483, 0.4659, 0.1998, 0.206, 0.1821, 0.4192, 0.2518, 0.1586, 0.1009, 0.7638, 0.1894, 0.0932, 0.1216, 0.0043, 0.6037, 0.721, 0.5059, 1.5628, 0.1291, 0.1947, 0.1687, 0.1271, 0.0496, 0.6587, 0.0595, 0.0405, 0.3177, 0.0834, 0.1727 ]
6
induction_heads_induction
512
32
0.6765
0.7879
6.63
10.37
1.56
6.9
0.9838
3.227
3.884
[ 0.0882, 1.3184, 0.5192, 0.3302, 0.3493, 0.1824, 0.0205, 0.1778, 0.0614, 0.7206, 0.5571, 1.0414, 0.7678, 0.0055, 0.2967, 0.3771, 0.4628, 0.3815, 0.2352, 0.4973, 1.0716, 0.7008, 0.1079, 0.2141, 0.0372, 0.7204, 0.314, 0.3205, 0.0557, 0.6031, 0.1019, 0.1529 ...
7
noise_filtering_retrieval
1,024
32
1.139
0.8054
14.59
41.76
2.86
11.3
0.993
3.231
3.935
[ 0.0684, 0.5194, 0.0011, 0.3468, 0.588, 0.4251, 0.302, 0.0668, 0.0559, 0.118, 0.3371, 0.5871, 0.0411, 0.0033, 0.3108, 0.0697, 0.7897, 0.4501, 0.2258, 0.0277, 0.1748, 0.076, 0.0636, 0.6098, 0.0722, 0.3074, 0.0615, 0.3478, 1.4146, 0.0138, 1.124, 0.1446 ]
8
selective_associative_recall
2,048
8
0.9869
0.8517
29.15
162.5
5.57
20.1
0.9718
3.883
2.771
[ 0.0707, 0.0147, 0.317, 0.0473, 0.0645, 0.0905, 0.4072, 0.1802, 0.1038, 0.1941, 0.3184, 0.1544, 0.7075, 0.7873, 0.2507, 0.066, 0.1437, 1.0507, 0.4598, 0.5642, 0.1085, 0.0525, 0.2018, 0.4776, 0.3529, 0.1964, 0.0277, 0.5175, 0.9417, 0.0487, 0.1376, 1.4486 ]
9
variable_length_copying
256
32
0.8487
0.796
3.38
4
1.18
4.7
0.9823
3.488
3.039
[ 0.1293, 0.4424, 0.5036, 0.1804, 0.1526, 0.0615, 0.3418, 2.2085, 1.2781, 0.6123, 0.4336, 0.2013, 0.2499, 0.1649, 0.0134, 0.4476, 0.2149, 0.712, 0.1066, 0.5179, 0.6441, 0.1478, 0.1149, 0.1534, 0.0334, 0.0196, 0.2578, 0.0992, 0.1522, 0.4248, 0.3022, 0.1865 ...
10
induction_heads_induction
4,096
16
0.7421
0.7881
57.61
629.12
10.92
37.7
0.9984
3.371
2.205
[ 0.0706, 0.3207, 0.1601, 0.1753, 0.0372, 1.0765, 0.1673, 0.2909, 0.0636, 0.2405, 0.2905, 0.3459, 0.0275, 0.3584, 0.1574, 0.435, 0.1514, 0.0216, 0.8114, 0.3994, 0.0463, 0.3268, 0.3074, 0.0036, 0.006, 0.3369, 0.1407, 0.2824, 0.7545, 0.2968, 0.0692, 0.1099 ]
11
noise_filtering_retrieval
256
32
0.7995
0.8825
4.16
4
0.96
4.7
0.9952
3.62
4.426
[ 0.2078, 0.013, 0.115, 0.0465, 0.5252, 0.3338, 0.5279, 0.5491, 0.0684, 0.3427, 0.3098, 0.4835, 0.174, 0.3739, 0.3505, 0.1359, 0.1358, 0.2418, 0.5551, 1.1901, 0.7722, 0.0042, 0.1654, 0.1549, 0.1903, 0.0825, 0.1247, 0.6147, 0.2695, 0.2913, 0.0034, 0.3529 ]
12
selective_associative_recall
4,096
8
0.9678
0.9122
57.72
628.1
10.88
37.7
0.9914
4.417
1.851
[ 0.2114, 0.0793, 0.1699, 0.0403, 0.0944, 0.0777, 0.0235, 1.7148, 0.6161, 0.3364, 0.3002, 0.0787, 0.7297, 0.0564, 0.5688, 0.5833, 0.335, 0.1825, 0.1968, 1.5436, 0.1084, 0.772, 0.076, 0.4543, 0.181, 0.5141, 0.5919, 0.1829, 0.1145, 0.7995, 0.3174, 0.0699 ]
13
variable_length_copying
4,096
16
0.6973
0.8396
58.05
627.79
10.81
37.7
0.9764
5.253
1.342
[ 0.6332, 0.4036, 0.1876, 0.3328, 0.1998, 0.1452, 0.0919, 0.4535, 0.0819, 0.0703, 0.0675, 0.0054, 0.4773, 0.1585, 0.0157, 0.0868, 0.2222, 0.0561, 0.3497, 0.0903, 0.0272, 0.194, 0.4361, 0.0588, 0.0792, 0.1622, 0.4826, 0.1804, 1.0156, 0.0153, 0.4486, 1.4391 ...
14
induction_heads_induction
1,024
16
0.6814
0.8868
14.69
45.49
3.1
11.3
0.9891
4.975
2.667
[ 0.0195, 0.2462, 0.1018, 1.3514, 0.104, 0.3259, 0.1167, 0.601, 0.2946, 0.3294, 0.5565, 0.9749, 0.4139, 0.0017, 0.0376, 0.2637, 0.4658, 0.0567, 0.236, 0.3329, 0.7708, 0.4638, 0.2614, 0.0432, 0.1572, 0.5909, 0.3341, 0.4001, 0.1452, 0.2727, 0.3133, 0.0278 ]
15
noise_filtering_retrieval
256
16
0.6298
0.7924
3.16
4
1.27
4.7
0.9774
3.599
1.361
[ 0.212, 0.5223, 0.2261, 0.1677, 0.1029, 0.9896, 0.1707, 0.1758, 0.1242, 0.1411, 0.1309, 0.3472, 0.5677, 0.0339, 0.0516, 1.7447, 0.4907, 0.3632, 0.4848, 0.3499, 0.5899, 0.2792, 0.1258, 0.4194, 0.3689, 0.8485, 0.3269, 0.0286, 0.777, 0.0205, 0.3462, 0.1379 ]
16
selective_associative_recall
4,096
32
0.7529
0.9025
58.63
626.74
10.69
37.7
0.9908
3.768
1.889
[ 1.2235, 0.0644, 0.349, 0.2682, 0.3636, 0.0463, 0.1579, 0.037, 0.5433, 0.6051, 0.8995, 0.0308, 0.2045, 0.0286, 1.1609, 0.1946, 0.1265, 1.1446, 0.2867, 0.1997, 0.4349, 0.135, 0.1827, 0.0556, 0.0742, 0.1794, 0.0222, 0.0196, 0.0912, 0.424, 0.0026, 0.1831 ]
17
variable_length_copying
2,048
16
0.7294
0.8147
29.19
163.87
5.61
20.1
0.9841
5.167
3.978
[ 0.2625, 0.4224, 0.7634, 0.1649, 0.6264, 0.2916, 0.6008, 0.1317, 0.3463, 0.0955, 1.0893, 1.007, 0.0696, 0.113, 0.2752, 0.6501, 0.4135, 0.6395, 0.0968, 0.1683, 0.5787, 0.8502, 0.0032, 0.7381, 0.1798, 0.5623, 0.4149, 0.2849, 0.5235, 0.2773, 0.6974, 0.2422 ]
18
induction_heads_induction
2,048
16
0.879
0.8827
29.05
160.31
5.52
20.1
0.9912
5.277
2.138
[ 0.0597, 0.2605, 0.4082, 0.0588, 0.5175, 0.9471, 0.356, 0.1642, 0.2151, 0.827, 0.2624, 0.0526, 0.1161, 0.144, 1.8398, 0.0368, 1.2571, 0.0083, 0.4703, 0.2434, 0.829, 0.3013, 0.0578, 0.1103, 0.0569, 0.2703, 0.1146, 0.0415, 0.0606, 0.511, 0.5916, 0.1082 ]
19
noise_filtering_retrieval
4,096
8
0.8222
0.8276
57.78
626.58
10.84
37.7
0.9715
4.027
4.228
[ 0.2756, 0.2012, 0.5514, 0.2496, 0.7085, 0.0663, 0.4472, 0.8991, 0.2604, 0.2349, 0.0056, 0.1658, 0.0456, 0.183, 0.8391, 0.1229, 0.4939, 0.0986, 0.0806, 2.3107, 0.1948, 0.0527, 0.0685, 0.2857, 0.1664, 0.2623, 0.2131, 0.5041, 0.0119, 0.7552, 0.5304, 0.1943 ...
20
selective_associative_recall
256
8
0.8356
0.9044
3.52
4.23
1.2
4.7
0.9829
4.84
1.21
[ 1.1686, 0.4656, 0.407, 0.3947, 0.0565, 1.4699, 0.243, 0.2117, 0.0121, 0.0705, 0.222, 0.968, 0.678, 0.3327, 0.0649, 0.2078, 0.3331, 1.0087, 0.5759, 0.0799, 0.0744, 0.0267, 0.0017, 0.3467, 0.0343, 0.2535, 0.3747, 0.1738, 0.8407, 0.2298, 0.3484, 0.7956 ]
21
variable_length_copying
2,048
32
0.9501
0.8634
28.74
160.04
5.57
20.1
0.9856
4.868
3.203
[ 0.1983, 0.1278, 0.5935, 0.1117, 0.075, 0.2629, 0.6922, 0.214, 0.2563, 0.3098, 0.0947, 0.1232, 0.3225, 0.1304, 0.3461, 0.955, 0.4175, 0.0791, 0.9282, 0.5251, 0.0266, 0.9571, 0.0714, 0.6009, 0.0685, 1.0909, 0.0727, 0.6616, 0.6748, 0.1677, 0.1814, 0.7774 ]
22
induction_heads_induction
2,048
32
0.9515
0.854
28.46
160.35
5.63
20.1
0.9922
4.814
1.739
[ 0.3624, 0.3895, 0.0773, 0.0209, 0.149, 0.1509, 0.4415, 0.1572, 0.174, 0.4117, 0.1854, 0.6729, 0.4531, 0.8602, 0.112, 0.0603, 0.1134, 0.3313, 0.3876, 0.3391, 0.5544, 0.2849, 0.5216, 0.009, 0.9105, 0.4852, 0.1352, 0.1535, 0.0994, 0.7256, 0.0174, 0.5673 ]
23
noise_filtering_retrieval
2,048
8
0.9589
0.8343
28.38
160.62
5.66
20.1
0.9772
4.908
3.665
[ 0.0631, 0.1962, 0.3752, 0.3581, 0.0046, 0.2453, 0.1314, 0.5566, 0.8224, 0.4076, 0.354, 0.1811, 0.0991, 0.1407, 0.1495, 0.1041, 0.0148, 0.0141, 0.2707, 0.437, 0.019, 0.0623, 0.0479, 0.3835, 0.3986, 0.0835, 0.1869, 0.8332, 0.0804, 0.2168, 0.2416, 0.1246 ]
24
selective_associative_recall
4,096
32
0.628
0.7793
58.33
628.77
10.78
37.7
0.9913
5.292
2.483
[ 0.352, 0.0873, 0.0409, 0.0226, 0.2462, 0.1622, 0.0882, 0.5979, 0.2147, 0.3188, 0.0625, 0.8773, 0.079, 0.0584, 0.4319, 0.2521, 0.1615, 0.2295, 0.0922, 0.9733, 0.0518, 0.2395, 0.2156, 0.1178, 0.6042, 0.1634, 0.4621, 0.162, 0.0973, 0.2483, 0.7694, 0.4658 ]
25
variable_length_copying
2,048
8
0.7701
0.8162
29.44
162.33
5.51
20.1
0.965
4.735
2.996
[ 0.7126, 0.5058, 0.0748, 0.1521, 0.572, 0.0287, 0.0046, 0.3375, 0.828, 0.4059, 0.8488, 0.0245, 0.1992, 0.0941, 0.1239, 0.8462, 0.3235, 0.0775, 0.372, 0.1966, 0.6348, 0.0402, 0.4579, 0.114, 0.0292, 0.0991, 0.3817, 0.2418, 0.1035, 0.0251, 0.7875, 0.4411 ]
26
induction_heads_induction
256
32
0.9586
0.8386
3.71
4
1.08
4.7
0.9727
3.818
3.736
[ 0.0871, 0.2053, 0.1565, 0.0561, 0.9923, 0.0616, 0.2394, 0.0861, 0.0041, 0.1063, 0.3466, 0.2323, 0.1878, 1.8071, 0.0223, 0.0714, 0.3618, 0.0456, 0.2564, 0.1414, 0.0075, 0.2571, 0.3714, 0.1348, 0.7234, 0.332, 0.1667, 0.4022, 0.0986, 0.0007, 0.0153, 0.2788 ...
27
noise_filtering_retrieval
2,048
8
0.961
0.8174
28.74
159.21
5.54
20.1
0.9773
5.611
2.699
[ 0.0325, 0.0053, 0.26, 0.1375, 0.2286, 1.0551, 0.8562, 0.0009, 0.5838, 0.044, 0.0934, 0.0152, 0.0407, 0.2907, 0.5395, 0.4586, 0.1805, 1.0168, 0.0381, 0.3561, 0.0748, 0.3498, 0.1644, 1.1856, 0.0831, 0.1506, 0.4452, 0.2899, 1.3109, 0.1709, 0.2726, 0.6338 ]
28
selective_associative_recall
512
32
1.1334
0.7636
7.1
12.28
1.73
6.9
0.9776
4.978
1.47
[ 0.606, 0.7441, 1.0041, 0.4404, 0.2491, 1.9224, 0.4303, 0.2523, 0.415, 0.04, 0.5942, 0.1868, 0.0875, 0.5528, 0.0561, 0.9077, 0.0981, 0.3019, 0.0106, 0.3622, 0.3503, 0.1905, 0.3529, 0.0172, 0.4448, 0.0774, 0.1667, 0.1576, 0.765, 0.1267, 1.0518, 1.5928 ]
29
variable_length_copying
1,024
32
0.8476
0.9194
14.19
43.48
3.06
11.3
0.9919
5.343
4.209
[ 0.0983, 0.1381, 0.6968, 0.1541, 0.2545, 0.0758, 0.3249, 0.1846, 0.4457, 0.0027, 0.2252, 0.0112, 0.268, 0.0602, 0.0044, 0.0465, 0.287, 0.6516, 0.4006, 0.6487, 0.123, 0.2261, 0.9506, 0.0036, 0.2157, 0.3515, 1.391, 0.0135, 0.1877, 0.2104, 0.4607, 0.1059 ]
30
induction_heads_induction
1,024
32
1.1248
0.8979
14.41
42.59
2.96
11.3
0.9916
4.93
2.385
[ 0.1915, 0.7479, 0.2914, 0.326, 0.585, 0.0389, 0.3284, 1.1164, 0.0203, 1.2384, 0.0633, 0.024, 0.0561, 0.0204, 0.2711, 0.222, 0.0199, 0.3335, 0.3514, 0.8344, 0.1977, 0.283, 0.0751, 0.5474, 0.0237, 0.5915, 0.8654, 0.0189, 0.5706, 0.2838, 0.0253, 1.1048 ]
31
noise_filtering_retrieval
256
8
0.9006
0.9101
3.55
4
1.13
4.7
0.9908
5.185
3.724
[ 0.2114, 0.026, 0.6009, 0.0631, 0.1791, 0.6977, 0.7287, 0.0026, 0.1522, 0.0951, 1.0243, 0.0494, 0.0957, 0.8607, 0.2868, 0.1978, 0.2401, 0.1559, 0.1538, 0.5177, 0.156, 0.609, 0.1408, 0.0055, 0.5638, 0.2501, 0.0233, 0.4488, 0.2033, 0.1476, 1.6667, 0.0686 ]
32
selective_associative_recall
1,024
16
0.8868
0.7668
13.81
40.56
2.94
11.3
0.965
3.802
3.788
[ 0.5733, 0.1132, 0.4632, 1.447, 0.3153, 0.0486, 0.5782, 0.0584, 0.25, 0.2196, 1.082, 0.1869, 0.2212, 0.957, 0.2313, 0.1429, 0.8228, 1.9069, 0.3624, 0.2866, 1.0261, 0.0876, 0.1908, 0.3548, 0.1798, 0.1534, 0.0602, 0.084, 0.0606, 0.0118, 0.062, 0.5895 ]
33
variable_length_copying
256
8
0.8832
0.901
4.01
4
1
4.7
0.9922
5.492
2.036
[ 0.2659, 0.1347, 0.3294, 0.3338, 1.0796, 0.2662, 0.4747, 0.2721, 0.6182, 0.5409, 0.0783, 0.062, 0.632, 0.2162, 0.0138, 0.0504, 0.252, 0.2127, 0.1084, 0.5328, 0.0963, 0.1278, 0.5102, 0.3434, 0.4881, 0.0338, 0.2442, 0.1592, 0.0324, 0.3033, 0.1339, 0.4992 ]
34
induction_heads_induction
1,024
8
0.9422
0.8899
13.71
42.36
3.09
11.3
0.9909
3.839
1.216
[ 0.1344, 0.0975, 0.7145, 0.0231, 0.8335, 1.3898, 0.2622, 0.2539, 0.7608, 0.121, 0.1333, 0.4895, 0.0072, 0.2383, 0.0334, 0.5394, 0.029, 0.15, 0.2187, 0.176, 0.0233, 0.2856, 0.1031, 0.1421, 0.1583, 0.48, 0.1288, 0.7461, 0.2136, 0.0411, 0.134, 0.491 ]
35
noise_filtering_retrieval
1,024
32
1.1467
0.8656
14.45
43.13
2.98
11.3
0.9732
3.893
2.133
[ 1.5836, 1.0293, 0.002, 0.7954, 0.4789, 0.3485, 0.108, 2.1965, 0.2679, 0.4129, 0.4948, 0.0258, 0.4121, 0.1249, 0.1258, 0.0391, 0.0918, 0.0939, 0.0685, 1.0707, 0.3925, 0.2487, 0.0875, 0.4038, 0.2243, 0.2842, 0.147, 0.0005, 1.1203, 0.1033, 0.4138, 0.2386 ]
36
selective_associative_recall
256
16
0.6115
0.8135
3.87
4
1.03
4.7
0.9698
4.384
1.509
[ 0.2886, 0.0199, 0.4014, 0.1193, 0.6017, 0.0341, 0.0226, 0.1957, 0.4504, 0.0317, 0.2497, 0.0919, 0.0773, 0.7332, 0.1859, 0.1344, 0.2299, 0.1776, 0.8678, 0.397, 0.0839, 0.0607, 0.0123, 1.0717, 0.2907, 0.0664, 0.055, 0.1126, 0.2971, 0.1751, 1.0054, 0.5651 ]
37
variable_length_copying
256
8
0.6998
0.8325
3.19
4.05
1.27
4.7
0.965
5.529
1.971
[ 0.339, 0.4434, 0.7867, 0.555, 0.093, 0.2234, 0.0713, 0.1758, 0.0513, 0.136, 0.0593, 0.3667, 0.4265, 0.5803, 0.0971, 0.0977, 0.0745, 0.6844, 0.4876, 0.1789, 0.1858, 0.2613, 0.0545, 0.9673, 0.1665, 0.3478, 1.0499, 0.0214, 0.508, 0.0765, 0.5321, 0.6827 ]
38
induction_heads_induction
4,096
8
0.6459
0.8591
57.78
627.85
10.87
37.7
0.9878
5.633
2.346
[ 0.733, 0.0022, 0.3674, 0.17, 0.4995, 0.0757, 0.0693, 0.2648, 0.3848, 0.4854, 0.1934, 1.4212, 0.5149, 0.2999, 0.2299, 0.5734, 0.0234, 0.1867, 1.426, 0.2165, 0.0253, 0.0572, 0.0074, 0.3535, 0.1417, 0.0175, 0.3027, 0.7109, 0.0127, 0.0116, 0.1533, 0.1168 ]
39
noise_filtering_retrieval
256
32
0.9666
0.7653
3.28
4.64
1.42
4.7
0.9793
3.213
3.488
[ 0.1908, 0.1034, 0.4351, 0.0065, 0.1573, 0.1517, 0.6841, 0.1254, 0.8096, 0.1415, 0.0335, 1.1742, 0.1026, 0.0429, 0.069, 0.4254, 0.2044, 0.2562, 0.0247, 0.0781, 0.2611, 0.0476, 0.0165, 1.5573, 1.1118, 0.9076, 0.6006, 0.1791, 1.1916, 0.2018, 0.1175, 0.7134 ...
40
selective_associative_recall
512
32
0.9705
0.7746
7.88
13.04
1.66
6.9
0.9794
4.306
2.749
[ 0.2204, 0.2494, 0.413, 0.0817, 0.4313, 0.1684, 0.1011, 0.2162, 0.1834, 0.2887, 0.2014, 0.7373, 0.098, 0.2277, 0.8474, 0.1801, 0.079, 0.0169, 0.3508, 0.3431, 0.1243, 0.0456, 0.2667, 0.3025, 0.4676, 1.9765, 0.4267, 0.4492, 0.5738, 0.1095, 0.2405, 1.0401 ]
41
variable_length_copying
2,048
32
0.9733
0.8347
29.81
163.26
5.48
20.1
0.9962
5.093
3.463
[ 0.0781, 0.0794, 0.383, 0.122, 0.1302, 0.6451, 0.6638, 0.9901, 0.0939, 0.4372, 1.0072, 0.2428, 0.104, 0.029, 0.4829, 0.7999, 0.0583, 0.3453, 0.4576, 0.1622, 0.1939, 0.0585, 0.3781, 0.0259, 0.4246, 0.3348, 0.5381, 0.1235, 0.0068, 0.4176, 0.484, 0.1893 ]
42
induction_heads_induction
2,048
8
1.002
0.9056
28.4
164.82
5.8
20.1
0.9915
4.662
1.243
[ 0.2595, 0.8839, 1.1168, 0.8437, 1.0477, 0.14, 0.1575, 0.0965, 0.0248, 0.0172, 0.0203, 0.3719, 0.0294, 0.2001, 0.0786, 0.2046, 0.0682, 0.0518, 0.0155, 0.1228, 0.1895, 0.1236, 0.02, 0.0518, 0.487, 0.207, 0.0046, 0.3322, 0.0329, 0.0298, 1.0228, 0.7966 ]
43
noise_filtering_retrieval
1,024
8
1.0589
0.7596
14.56
45.56
3.13
11.3
0.9872
5.458
3.233
[ 0.0841, 1.1995, 0.0603, 0.2632, 0.0473, 0.4439, 0.11, 0.2124, 0.487, 0.391, 0.3307, 0.168, 0.0556, 0.2117, 0.2113, 1.4262, 0.0219, 0.088, 0.6892, 0.1033, 0.7855, 0.5177, 0.2231, 0.3017, 0.046, 0.4921, 0.851, 0.0464, 0.0799, 0.3851, 0.4145, 0.3104 ]
44
selective_associative_recall
512
8
0.7652
0.7665
7.55
13.1
1.73
6.9
0.9842
4.782
2.298
[ 1.3464, 0.886, 0.3873, 0.0022, 1.4753, 0.0618, 0.4789, 0.1693, 0.5375, 1.7582, 0.8916, 0.4211, 0.3594, 1.3506, 0.312, 0.0023, 0.3845, 0.0199, 0.0149, 1.1417, 0.1029, 0.0728, 0.2583, 0.3581, 0.0773, 0.3518, 0.1451, 0.0256, 0.8115, 0.0911, 0.087, 0.0303 ]
45
variable_length_copying
1,024
8
0.7437
0.8879
14.2
41.39
2.91
11.3
0.9964
4.442
4.148
[ 0.0062, 0.2326, 0.1223, 0.0295, 0.0432, 0.5685, 0.6076, 0.3245, 0.0255, 0.1023, 0.0307, 0.0586, 0.5158, 0.2737, 0.0084, 1.7468, 0.0915, 0.1434, 0.2104, 0.0043, 0.0787, 0.7282, 0.3597, 0.253, 0.1525, 0.0437, 0.2327, 0.3894, 0.1161, 0.0071, 0.3773, 0.525 ]
46
induction_heads_induction
2,048
16
1.0174
0.8548
28.76
161.02
5.6
20.1
0.9833
5.461
3.861
[ 0.134, 0.1323, 0.0039, 0.0497, 0.3858, 0.0895, 0.144, 0.1608, 0.2817, 0.1218, 0.0421, 0.0076, 0.2783, 0.0531, 0.1781, 0.376, 0.0533, 0.0757, 0.2223, 0.3809, 0.0878, 0.0327, 0.101, 0.2769, 1.034, 0.0808, 0.0179, 0.1007, 0.3253, 0.3495, 0.4162, 0.2012 ]
47
noise_filtering_retrieval
1,024
16
1.0959
0.781
14.93
40.6
2.72
11.3
0.9766
4.254
3.778
[ 0.0701, 0.0023, 0.2591, 0.9787, 0.5475, 1.4842, 0.0732, 0.5434, 0.2573, 0.0141, 0.4882, 0.1009, 0.5507, 0.4351, 0.5905, 0.1428, 0.4699, 0.7245, 0.2992, 0.1857, 0.4812, 0.6059, 0.1455, 0.6953, 0.3224, 0.0816, 0.3267, 0.2043, 0.849, 0.4364, 0.7201, 0.1127 ...
48
selective_associative_recall
256
8
1.1282
0.8533
3.69
4
1.08
4.7
0.9839
3.897
3.965
[ 0.0744, 0.4221, 0.2384, 0.0669, 0.0867, 0.1432, 0.0078, 0.236, 0.3243, 0.1997, 1.0957, 0.6961, 0.1, 0.8799, 0.4182, 0.2836, 0.0027, 0.6302, 0.5873, 0.7759, 0.4157, 0.0177, 0.5961, 0.68, 0.0412, 0.292, 0.3419, 0.1808, 0.1452, 0.1231, 0.0925, 0.1021 ]
49
variable_length_copying
256
16
1.0281
0.8521
3.65
4.2
1.15
4.7
0.9922
3.743
3.608
[ 0.1924, 0.2833, 0.1888, 0.1062, 0.4852, 0.1386, 0.1511, 0.8761, 0.2729, 0.0037, 0.031, 0.0549, 0.0496, 0.489, 0.0747, 0.0583, 0.2068, 0.1348, 0.0327, 0.0959, 0.1621, 0.1957, 0.278, 0.1609, 0.3181, 0.4965, 0.3834, 0.3924, 0.3392, 0.1146, 0.1059, 0.0604 ]
50
induction_heads_induction
4,096
16
0.9113
0.8105
58.14
627.31
10.79
37.7
0.9987
5.126
2.846
[ 0.2678, 0.3197, 0.4813, 0.1924, 0.5873, 0.0459, 0.2624, 0.4389, 0.2517, 0.0608, 0.008, 0.0159, 0.1027, 0.2933, 0.1092, 0.259, 0.0785, 0.5035, 0.0129, 0.5353, 0.0215, 0.0038, 0.4012, 0.0711, 0.3114, 0.3833, 0.0541, 0.407, 0.7625, 1.4073, 0.0819, 1.0635 ]
51
noise_filtering_retrieval
4,096
32
0.9011
0.831
58.28
628.97
10.79
37.7
0.9941
4.918
1.704
[ 0.5475, 0.2899, 0.2, 0.1508, 0.779, 0.4563, 0.0172, 0.0247, 0.5523, 0.2706, 0.2827, 0.8851, 1.9639, 0.3625, 0.0708, 0.4012, 0.1606, 0.1523, 0.1072, 1.1833, 0.648, 0.3938, 0.846, 0.0291, 0.1341, 0.2508, 0.1633, 0.1924, 0.1327, 0.1929, 0.7617, 0.224 ]
52
selective_associative_recall
256
8
1.0595
0.7865
3.84
4.4
1.15
4.7
0.9899
4.306
1.805
[ 0.1261, 0.5662, 0.063, 0.577, 0.8506, 0.2534, 0.2021, 0.0978, 0.0501, 0.2897, 0.0196, 0.4023, 0.1393, 0.0922, 0.0841, 0.3272, 0.4642, 0.532, 0.1779, 0.3338, 0.1605, 0.2238, 0.1962, 0.1306, 0.5881, 0.0286, 0.8712, 0.2017, 0.1167, 0.0654, 0.8026, 0.1063 ]
53
variable_length_copying
1,024
16
0.9255
0.7679
14.36
43.17
3.01
11.3
0.9911
3.828
3.527
[ 0.064, 0.4696, 0.2863, 0.8682, 0.7943, 0.0359, 0.3886, 0.2154, 0.2369, 0.2884, 0.0698, 0.2879, 0.4663, 0.0303, 0.0094, 0.2161, 1.292, 0.1012, 0.3876, 0.1733, 0.4099, 1.0818, 0.0453, 0.3404, 1.0055, 0.0943, 0.1597, 0.0467, 0.2735, 0.0687, 0.1065, 0.2761 ]
54
induction_heads_induction
4,096
16
0.9365
0.7985
58.24
626.23
10.75
37.7
0.978
4.883
4.268
[ 0.8019, 0.1449, 0.5383, 0.1208, 0.2717, 0.0055, 0.829, 0.639, 0.7867, 1.3528, 0.1559, 0.665, 0.0972, 0.0381, 0.4351, 0.1374, 0.2501, 0.1387, 0.1992, 0.0074, 0.1421, 0.0663, 0.4657, 0.2065, 0.6964, 0.1084, 0.2294, 0.2753, 0.3173, 0.0936, 0.743, 0.992 ]
55
noise_filtering_retrieval
4,096
16
0.8701
0.7964
57.67
627.08
10.87
37.7
0.9883
3.539
1.977
[ 0.1733, 0.023, 0.1247, 0.4165, 0.846, 0.111, 0.0907, 0.6834, 0.1878, 0.4742, 0.0503, 0.9112, 0.1728, 0.7188, 1.1167, 1.1319, 0.2042, 0.1327, 0.1204, 1.4191, 0.1255, 0.0667, 0.3622, 0.3805, 0.3007, 1.1443, 0.5035, 0.4179, 0.0112, 0.2864, 0.1696, 0.5622 ]
56
selective_associative_recall
2,048
32
0.6247
0.9132
29.12
162.14
5.57
20.1
0.9816
4.329
2.858
[ 0.2945, 0.2837, 0.4369, 0.0905, 0.4431, 0.7254, 0.3778, 0.562, 0.5527, 0.0772, 0.5816, 0.8746, 0.157, 0.1181, 0.3175, 0.662, 0.4131, 0.2433, 0.1839, 0.0138, 0.0674, 0.6071, 0.2947, 0.6867, 0.2742, 0.2755, 0.1801, 0.0667, 0.3461, 0.0125, 0.0727, 0.2868 ]
57
variable_length_copying
256
16
0.7743
0.9192
3.58
4
1.12
4.7
0.9785
4.546
3.267
[ 0.0109, 0.2054, 0.4912, 0.0013, 0.5369, 0.5273, 0.5966, 0.2597, 0.0581, 0.2225, 0.0915, 0.0597, 0.0525, 0.4679, 0.1008, 0.0208, 0.3928, 0.4248, 0.1147, 0.4163, 0.0712, 0.021, 0.0377, 0.1572, 0.16, 0.2246, 0.07, 0.9255, 0.0963, 0.1018, 0.0109, 0.2812 ]
58
induction_heads_induction
1,024
16
0.611
0.7998
14.51
41.53
2.86
11.3
0.9863
4.243
4.38
[ 0.0361, 0.0471, 0.0297, 0.3245, 0.284, 0.8374, 0.25, 0.6257, 0.7901, 1.1984, 0.0805, 0.0817, 0.0103, 0.2298, 0.3686, 0.3454, 0.244, 0.0816, 0.2288, 0.0178, 0.0354, 0.2825, 0.1098, 0.0179, 0.325, 0.0038, 0.478, 0.3679, 0.357, 1.4227, 0.4599, 0.6568 ]
59
noise_filtering_retrieval
256
16
0.7433
0.8944
2.92
4
1.37
4.7
0.9701
5.365
2.791
[ 0.2318, 0.6687, 0.1535, 0.1429, 1.1192, 0.4055, 0.6471, 0.4285, 0.1795, 1.5788, 0.1844, 0.2409, 0.0072, 0.0715, 0.4812, 0.1852, 0.6895, 0.4719, 0.3111, 0.0538, 1.3024, 0.0357, 0.4469, 0.0123, 0.0435, 0.2254, 0.6827, 0.1512, 0.2013, 0.3005, 0.9139, 2.1559...
60
selective_associative_recall
2,048
32
0.7488
0.8534
28.24
162.62
5.76
20.1
0.9759
3.41
1.69
[ 0.3125, 0.6116, 0.227, 0.0664, 0.0094, 0.3349, 0.4804, 0.0604, 0.0627, 0.0719, 0.1341, 1.6147, 0.0853, 0.3132, 0.1576, 0.0161, 0.3473, 0.4436, 0.0828, 0.1402, 0.2567, 0.112, 0.7464, 0.0479, 0.0758, 0.1216, 0.332, 0.433, 0.1122, 0.8585, 0.3321, 1.2529 ]
61
variable_length_copying
1,024
32
0.9438
0.9085
14.58
42.84
2.94
11.3
0.9973
5.399
2.311
[ 0.1538, 0.1913, 0.5304, 0.5282, 0.0746, 0.3553, 0.3117, 0.5854, 0.3092, 0.0354, 0.8222, 0.0697, 0.3766, 0.3746, 0.4165, 0.0326, 0.2319, 0.6613, 0.552, 0.0339, 0.0134, 0.2753, 0.171, 0.4755, 0.105, 0.5382, 0.4607, 0.8287, 0.1117, 0.0624, 0.2159, 0.2962 ]
62
induction_heads_induction
256
32
0.9614
0.8883
3.36
4.02
1.2
4.7
0.9927
3.656
3.927
[ 0.592, 0.028, 0.502, 0.3083, 0.1627, 1.1599, 0.0695, 1.4301, 0.1247, 0.0061, 0.3742, 0.3657, 1.5054, 0.4065, 0.1399, 1.148, 0.1755, 0.4539, 0.4939, 0.2465, 0.0076, 1.3463, 0.1646, 0.1547, 0.1409, 0.6306, 0.1795, 0.3319, 0.1263, 0.6844, 0.0656, 0.196 ]
63
noise_filtering_retrieval
256
32
0.8673
0.8275
3.22
4.99
1.55
4.7
0.9724
5.494
4.108
[ 0.3366, 1.2651, 0.164, 0.5093, 0.1035, 0.4005, 0.0642, 0.8109, 0.6457, 0.9737, 0.6227, 0.3086, 0.8434, 0.394, 0.6669, 0.5207, 0.3514, 0.3128, 0.3141, 0.2817, 0.1211, 0.0989, 0.1003, 0.4618, 1.866, 1.0915, 0.0921, 0.7458, 0.3315, 0.095, 0.1524, 0.2127 ]
64
selective_associative_recall
256
16
0.6886
0.783
3.42
4
1.17
4.7
0.9881
3.917
1.896
[ 0.204, 0.1027, 0.1148, 0.0646, 0.2298, 0.736, 0.4047, 0.1036, 0.0132, 0.5487, 0.0313, 0.0555, 0.4043, 0.045, 0.0864, 0.7281, 0.4772, 0.1116, 0.0729, 0.008, 0.3299, 0.4173, 0.0418, 0.7431, 0.8028, 0.8479, 0.5301, 0.098, 1.3235, 0.4651, 0.1253, 0.4692 ]
65
variable_length_copying
2,048
8
0.882
0.875
28.87
162.9
5.64
20.1
0.9849
5.609
1.303
[ 0.2009, 0.2632, 0.2762, 0.1979, 0.2082, 1.6698, 0.4716, 0.1, 0.0148, 0.1711, 0.0309, 0.5382, 0.0476, 0.4337, 0.214, 0.2061, 0.818, 0.0746, 0.6075, 0.5774, 0.1922, 0.024, 0.005, 0.0471, 1.0534, 0.3408, 0.3878, 0.0428, 0.0132, 0.1187, 0.1243, 0.1345 ]
66
induction_heads_induction
1,024
16
0.7538
0.7744
14.49
43.55
3.01
11.3
0.9805
3.5
1.644
[ 1.0414, 1.735, 0.2208, 0.4466, 0.0247, 0.6394, 0.4481, 0.188, 0.2403, 0.4937, 0.5826, 0.4023, 0.701, 0.6151, 0.1616, 0.773, 0.3558, 0.679, 0.0169, 0.0751, 0.0445, 0.0114, 0.0374, 0.0861, 1.4265, 0.1124, 0.3246, 0.6043, 0.1777, 0.1111, 1.0929, 0.2145 ]
67
noise_filtering_retrieval
512
8
0.8331
0.7706
7.27
11.45
1.58
6.9
0.9918
3.294
4.082
[ 0.0027, 0.9098, 0.4362, 0.2421, 0.4752, 0.3251, 0.2209, 0.2831, 1.0538, 0.3772, 0.2221, 0.5319, 0.9075, 0.1818, 0.2889, 0.0757, 0.7951, 0.1154, 0.145, 0.2517, 0.2841, 0.0975, 0.2753, 0.3054, 0.6683, 0.0088, 0.687, 0.1912, 0.7047, 0.6692, 0.6346, 0.1849 ]
68
selective_associative_recall
1,024
8
0.6151
0.8579
13.61
42.93
3.15
11.3
0.9878
5.235
4.188
[ 0.2461, 0.4011, 0.0146, 0.2464, 0.1103, 0.2973, 0.0677, 0.0067, 0.4667, 0.6199, 0.0223, 0.1009, 0.1059, 0.2717, 0.048, 1.0962, 0.1909, 0.7353, 1.5726, 0.1081, 0.4511, 0.0631, 0.532, 0.2987, 0.2034, 0.0051, 0.1058, 0.2779, 0.0437, 0.0736, 0.0012, 0.514 ]
69
variable_length_copying
1,024
32
0.9261
0.8676
14.81
44.07
2.98
11.3
0.992
5.188
4.421
[ 0.0967, 0.3971, 0.3121, 0.0511, 0.1134, 0.2224, 0.0342, 0.5466, 1.1396, 0.1949, 0.385, 0.031, 0.5425, 0.0669, 0.3042, 0.449, 0.0032, 0.163, 0.0171, 0.3584, 0.3021, 0.0713, 0.0791, 0.8432, 0.0096, 0.0527, 1.5481, 0.2552, 0.2622, 0.3045, 0.3026, 0.6072 ]
70
induction_heads_induction
2,048
16
0.6591
0.8945
29.01
162.46
5.6
20.1
0.965
4.361
2.438
[ 0.227, 0.6078, 1.0168, 0.6767, 0.4357, 0.4862, 0.3153, 0.8394, 1.6872, 0.0586, 0.711, 0.7812, 1.1546, 0.323, 1.2758, 1.026, 0.3781, 0.0501, 0.0183, 0.9781, 0.1874, 0.3323, 0.0173, 1.0451, 1.2242, 0.2821, 0.3518, 0.4128, 0.1282, 0.4781, 0.2309, 0.01 ]
71
noise_filtering_retrieval
512
16
1.0568
0.7596
7.41
10.78
1.45
6.9
0.965
4.679
1.691
[ 0.1703, 0.3331, 0.776, 0.216, 0.4599, 0.0771, 0.792, 0.0649, 0.2331, 0.0036, 0.143, 0.7247, 0.344, 0.2048, 0.2083, 0.4011, 0.6302, 0.1435, 0.4752, 0.4624, 0.0576, 0.416, 0.1807, 0.3454, 0.0693, 0.0233, 0.3242, 0.4038, 0.1847, 0.0336, 0.2827, 0.0371 ]
72
selective_associative_recall
4,096
16
0.7915
0.8775
57.75
629.12
10.89
37.7
0.9681
3.996
4.368
[ 0.2161, 1.3733, 1.4222, 0.1679, 0.0753, 0.0882, 0.2787, 0.1524, 0.0013, 0.0899, 0.9012, 0.3874, 0.1027, 0.5206, 0.3297, 0.5527, 0.0457, 0.6798, 1.6429, 0.223, 0.1846, 0.7593, 0.9615, 0.5294, 0.0039, 0.0714, 0.0902, 0.1616, 0.6853, 0.9521, 0.1323, 0.5697 ...
73
variable_length_copying
1,024
16
0.7682
0.7728
14.97
41.46
2.77
11.3
0.9942
4.779
3.64
[ 0.6317, 0.2482, 0.8325, 0.24, 0.3491, 0.4943, 0.294, 0.3355, 0.4089, 0.1922, 0.4505, 0.0546, 0.453, 0.0071, 0.2941, 0.0626, 0.5292, 0.3868, 0.0319, 1.9025, 0.8161, 0.2865, 0.3127, 0.2601, 0.4496, 0.4288, 0.0151, 0.2038, 0.1711, 0.3475, 0.2855, 0.3196 ]
74
induction_heads_induction
512
32
0.6058
0.8631
7.35
14.19
1.93
6.9
0.965
4.615
3.004
[ 0.1639, 0.1711, 0.0214, 0.3083, 0.0785, 0.011, 0.3481, 0.2372, 0.2754, 0.0424, 0.1925, 0.3231, 0.1049, 1.5347, 0.8568, 0.0512, 0.0315, 0.7113, 0.119, 0.0309, 0.4661, 0.4186, 0.1552, 0.269, 0.098, 0.2095, 0.246, 0.2146, 0.8095, 0.0215, 0.0747, 0.195 ]
75
noise_filtering_retrieval
4,096
8
0.8977
0.8978
57.84
626.33
10.83
37.7
0.9787
3.686
4.405
[ 0.1311, 0.1737, 0.2544, 0.1609, 0.3349, 0.1034, 0.8059, 0.091, 0.1187, 0.6564, 0.2082, 0.2541, 0.6433, 0.9424, 0.4424, 0.0959, 0.1882, 0.0899, 0.9266, 0.0815, 0.1055, 0.6339, 0.0827, 0.6265, 0.6895, 0.0038, 1.1187, 0.1085, 0.1864, 0.2226, 0.5416, 0.0669 ...
76
selective_associative_recall
512
16
0.9395
0.8299
7.03
12.43
1.77
6.9
0.9842
3.502
2.87
[ 0.0326, 0.6511, 0.0191, 0.0178, 0.3799, 0.1867, 0.6395, 0.4644, 0.5869, 0.3602, 0.675, 0.1638, 0.1724, 0.0345, 0.3443, 0.0854, 0.0145, 0.6016, 0.1826, 0.0565, 0.1134, 1.0002, 0.0376, 0.1433, 0.0084, 0.1237, 0.2054, 0.1787, 0.419, 0.0744, 0.0579, 0.0063 ]
77
variable_length_copying
4,096
16
0.6498
0.7639
58.07
625.48
10.77
37.7
0.9955
3.741
3.27
[ 0.0861, 0.0992, 1.1003, 0.1202, 0.3015, 0.9278, 0.3438, 0.1807, 0.2384, 0.919, 1.3236, 0.0199, 0.7951, 0.9126, 0.9825, 1.4547, 0.1953, 0.4811, 0.1249, 0.2835, 0.1113, 0.1082, 0.2511, 0.1914, 0.0012, 0.0692, 0.2067, 0.1094, 0.445, 0.3692, 0.7402, 0.7044 ]
78
induction_heads_induction
1,024
32
0.6645
0.7814
14.16
44.5
3.14
11.3
0.965
5.417
3.439
[ 0.1838, 0.1938, 0.0535, 0.0266, 0.0674, 0.131, 0.0117, 0.1417, 0.639, 0.1961, 0.3529, 0.5839, 0.0545, 0.2569, 0.2856, 0.1136, 0.2817, 0.0006, 1.6714, 0.1731, 0.1879, 0.163, 0.1724, 1.3001, 0.5246, 0.4237, 1.2466, 0.1744, 0.0122, 0.309, 0.1302, 0.5053 ]
79
noise_filtering_retrieval
512
8
0.7129
0.9106
7.28
12.43
1.71
6.9
0.9851
4.534
1.934
[ 0.4904, 0.3449, 0.0492, 0.013, 0.1532, 0.1223, 0.1646, 0.1703, 0.21, 0.4155, 0.1852, 0.3341, 0.197, 0.2624, 0.3651, 0.022, 0.2254, 0.0884, 0.2799, 0.2385, 0.0923, 0.0409, 1.1335, 0.4088, 0.1471, 0.2552, 0.1443, 0.2287, 0.0368, 0.412, 0.1846, 0.0041 ]
80
selective_associative_recall
4,096
32
0.6373
0.8691
57.93
626.61
10.82
37.7
0.9878
3.556
2.004
[ 0.2524, 0.1566, 0.7603, 0.3596, 0.3426, 0.3669, 0.5214, 0.1326, 0.4973, 0.2686, 0.4626, 0.0301, 0.5885, 0.7878, 0.1598, 0.2315, 0.2147, 0.6716, 0.6396, 0.3989, 0.9146, 0.0286, 0.9456, 0.3452, 0.225, 0.3572, 0.3275, 0.1905, 0.565, 0.3283, 0.398, 0.595 ]
81
variable_length_copying
512
16
0.9676
0.8577
7.16
14.54
2.03
6.9
0.9847
4.578
2.815
[ 0.1295, 0.5892, 0.0039, 0.3057, 0.4158, 0.2229, 0.0481, 0.5645, 0.3432, 0.5629, 0.2033, 0.235, 0.1967, 0.1044, 0.5137, 0.0483, 0.1221, 0.2585, 0.0597, 0.0546, 0.6239, 0.1778, 0.2694, 0.0194, 0.4766, 0.0011, 0.224, 0.8014, 0.1173, 0.4619, 0.5243, 0.7316 ]
82
induction_heads_induction
256
8
0.8437
0.9015
3.43
4.8
1.4
4.7
0.9672
5.237
2.275
[ 0.0269, 0.0034, 0.6208, 0.0278, 0.5168, 0.3922, 0.0611, 0.0108, 0.1264, 0.1932, 0.6116, 0.7093, 0.2038, 0.1951, 0.0704, 0.4723, 0.1743, 0.1389, 0.3646, 0.1733, 0.4485, 0.2096, 0.231, 0.5389, 1.4449, 0.4676, 0.1302, 0.0955, 0.8496, 0.0534, 0.7411, 0.2557 ...
83
noise_filtering_retrieval
512
16
0.8782
0.8354
7.09
12.6
1.78
6.9
0.9949
4.86
1.903
[ 0.5884, 0.0929, 0.0521, 0.6554, 0.3838, 0.0391, 0.0655, 0.0072, 0.0666, 0.1604, 0.0544, 0.4427, 0.7661, 0.2397, 0.8386, 0.0127, 0.1058, 0.1011, 0.2927, 0.011, 0.1632, 0.3537, 0.1595, 0.3456, 0.0907, 0.2649, 0.17, 0.0104, 0.0112, 0.3848, 0.0612, 1.7578 ]
84
selective_associative_recall
256
16
0.8201
0.9122
4.14
4
0.97
4.7
0.9737
3.804
2.48
[ 0.1162, 0.1831, 0.3505, 0.4009, 0.1851, 0.0624, 0.2549, 0.1874, 0.954, 0.7025, 0.7878, 0.4241, 0.6067, 0.1666, 0.1947, 0.2223, 0.1076, 0.1833, 0.322, 0.282, 0.1429, 0.0774, 0.1967, 0.3594, 0.1486, 0.0526, 0.14, 0.3644, 0.3656, 0.2911, 0.5375, 0.2559 ]
85
variable_length_copying
2,048
32
0.7987
0.7579
28.37
163.16
5.75
20.1
0.9782
4.505
3.855
[ 0.1934, 0.8096, 0.0706, 0.0655, 0.1292, 0.8779, 1.594, 1.2975, 0.2363, 0.0378, 0.9291, 0.1885, 0.0644, 1.2537, 0.4225, 0.0995, 0.5465, 0.0271, 0.2374, 0.1639, 0.2869, 0.7841, 0.0237, 0.0259, 0.0852, 0.0737, 0.1399, 0.3159, 0.0057, 0.293, 0.3411, 0.0417 ]
86
induction_heads_induction
256
32
0.7358
0.8176
3.23
4
1.24
4.7
0.9809
5.52
3.324
[ 0.0982, 0.1966, 0.5464, 0.2595, 0.0064, 1.0394, 0.4982, 0.4454, 0.6378, 0.491, 0.3175, 0.0396, 0.5159, 0.9466, 0.0253, 0.1447, 0.11, 0.3066, 0.5607, 0.1677, 0.4845, 0.1863, 0.9884, 0.3205, 0.353, 0.3969, 0.2162, 0.0433, 0.1591, 1.2372, 0.7038, 0.333 ]
87
noise_filtering_retrieval
2,048
32
0.8558
0.9012
28.91
163.81
5.67
20.1
0.965
3.265
1.372
[ 0.2227, 0.2271, 0.6457, 0.2375, 0.0177, 0.0621, 2.0072, 0.1947, 0.8338, 0.5428, 0.6771, 0.8351, 0.3937, 0.4546, 0.015, 0.6441, 0.1024, 0.7066, 0.0112, 0.2004, 1.5978, 0.8844, 1.1382, 0.6915, 0.1368, 0.3769, 0.7789, 0.8216, 0.2441, 0.1258, 0.0261, 0.3936 ...
88
selective_associative_recall
2,048
32
0.8364
0.7909
28.8
162.77
5.65
20.1
0.9951
3.803
1.285
[ 0.6051, 0.153, 0.1886, 0.3184, 0.0477, 0.3996, 1.2853, 0.2379, 0.1579, 0.1851, 0.7236, 0.0207, 0.2979, 0.1827, 1.3844, 0.2998, 0.285, 0.3224, 0.0174, 0.067, 0.0372, 0.6443, 0.3948, 0.1399, 0.1396, 0.1953, 0.0257, 0.1102, 0.2473, 0.1811, 0.4411, 0.4057 ]
89
variable_length_copying
2,048
32
0.9218
0.8709
28.3
161.27
5.7
20.1
0.9715
5.164
4.201
[ 0.23, 0.9229, 0.4644, 1.0915, 0.1941, 0.9383, 0.6371, 0.0171, 0.9479, 0.0415, 0.2904, 0.0618, 0.7407, 0.182, 0.047, 0.2857, 0.3797, 0.2361, 0.9357, 0.1354, 0.1384, 0.1859, 0.0295, 0.2262, 0.3213, 0.4037, 1.3372, 0.0914, 0.9692, 0.2999, 0.0484, 0.3027 ]
90
induction_heads_induction
4,096
16
0.8777
0.8023
57.88
630.34
10.89
37.7
0.9924
3.695
2.206
[ 0.3612, 0.5438, 0.0349, 0.0847, 0.3513, 0.5931, 0.6046, 0.202, 0.1229, 0.1573, 0.0727, 0.4775, 0.0428, 0.3244, 0.1434, 0.5074, 0.3109, 0.1755, 0.244, 0.8141, 0.1689, 0.6664, 0.0656, 0.2638, 0.5124, 1.1265, 0.2921, 0.1075, 0.1091, 0.2198, 0.1243, 0.0197 ]
91
noise_filtering_retrieval
4,096
8
0.8383
0.7965
57.69
627.45
10.88
37.7
0.9978
4.793
2.249
[ 0.4304, 0.1318, 0.5872, 0.0462, 0.4154, 0.4068, 0.6287, 0.4144, 0.3571, 0.5369, 0.2195, 0.1997, 0.5002, 1.4636, 0.0138, 0.4789, 0.1116, 0.0748, 0.7023, 0.0331, 0.71, 0.1304, 0.6246, 0.1543, 0.109, 0.0055, 0.0123, 0.4617, 0.1448, 1.1063, 0.2658, 0.1451 ]
92
selective_associative_recall
4,096
32
0.6023
0.782
58.26
628.22
10.78
37.7
0.9867
4.642
4.004
[ 0.1234, 0.4553, 0.1515, 0.0285, 0.0649, 0.5795, 0.1848, 0.0438, 0.0045, 0.5607, 0.4235, 0.2133, 0.8963, 0.7676, 0.9704, 0.176, 0.7131, 0.3091, 0.6746, 0.624, 0.4872, 0.0279, 0.2992, 0.1898, 0.6992, 0.2772, 0.4091, 0.1074, 0.2217, 0.4545, 0.9961, 0.0172 ]
93
variable_length_copying
2,048
32
0.6066
0.9161
29.57
162.19
5.48
20.1
0.9832
3.421
1.797
[ 1.0692, 0.2861, 0.1353, 0.2886, 0.4632, 0.5195, 0.2009, 0.0404, 0.5196, 0.0581, 1.0109, 0.2501, 0.3232, 0.1751, 0.335, 0.4367, 0.2153, 0.4269, 0.1826, 0.1988, 0.2829, 0.0137, 0.1636, 0.0057, 0.1047, 0.1374, 0.4628, 0.3434, 0.185, 0.1368, 0.2068, 0.0644 ]
94
induction_heads_induction
512
32
0.9843
0.8649
7.7
11.17
1.45
6.9
0.9961
3.376
1.449
[ 0.01, 0.1615, 0.7561, 0.4246, 0.0236, 0.3046, 0.5196, 0.5561, 0.8472, 0.0252, 0.3031, 0.107, 0.6683, 0.3823, 0.4108, 0.2894, 0.0288, 0.3086, 0.0689, 0.1047, 0.5586, 0.1032, 0.2276, 0.0345, 0.2155, 0.0022, 0.0308, 0.008, 0.3348, 0.1207, 0.5141, 0.0163 ]
95
noise_filtering_retrieval
2,048
8
0.9505
0.8805
28.71
162.37
5.66
20.1
0.9705
5.33
2.372
[ 0.3256, 0.0305, 0.3681, 0.091, 0.3612, 0.1168, 0.0672, 0.3145, 0.2487, 0.1156, 0.3575, 0.1585, 0.1792, 0.1345, 0.1541, 0.074, 0.1467, 0.576, 0.3804, 0.9791, 0.0775, 0.2188, 0.1364, 0.309, 0.2727, 0.2403, 0.122, 0.7623, 0.065, 0.3108, 0.0471, 0.1561 ]
96
selective_associative_recall
1,024
8
0.9496
0.917
14.78
41.53
2.81
11.3
0.9668
4.089
2.149
[ 1.1324, 0.1907, 0.5349, 0.6151, 0.3772, 0.0055, 0.0153, 0.6922, 1.0603, 0.0853, 0.0799, 0.3506, 0.0837, 0.1784, 0.2097, 0.1914, 0.4843, 0.1559, 0.0768, 0.4178, 0.0143, 0.2084, 0.0214, 0.5251, 0.0499, 0.6948, 0.5493, 0.1028, 0.5789, 0.2894, 0.3712, 0.3514...
97
variable_length_copying
256
32
0.8082
0.8554
3.68
4
1.09
4.7
0.9795
4.716
3.889
[ 0.0815, 0.3398, 0.1158, 0.0386, 0.0978, 0.0024, 0.4663, 0.5546, 0.5629, 0.0569, 0.3735, 0.1979, 0.0144, 1.6429, 1.1032, 0.092, 0.0867, 0.3121, 0.4878, 0.29, 0.1177, 0.1104, 0.4966, 0.2474, 0.1741, 0.23, 0.0038, 0.6756, 0.0512, 0.7202, 0.0604, 0.7268 ]
98
induction_heads_induction
2,048
8
0.8397
0.7533
29.11
161.08
5.53
20.1
0.9871
4.728
2.68
[ 0.1232, 0.2153, 0.284, 0.0156, 0.1273, 0.3863, 0.0989, 0.8997, 0.1288, 0.0477, 0.0536, 0.1468, 0.1217, 0.0124, 0.4949, 0.5356, 0.8162, 0.4385, 0.27, 0.6364, 1.128, 0.2172, 0.0483, 0.0577, 0.4135, 0.1222, 0.2511, 0.1506, 0.2275, 0.3773, 0.2766, 0.2302 ]
99
noise_filtering_retrieval
2,048
8
1.139
0.7977
28.69
161.36
5.62
20.1
0.965
5.672
2.939
[ 0.1317, 0.105, 0.0486, 0.5347, 0.0859, 0.2678, 0.0232, 0.0579, 0.1781, 0.2494, 0.3572, 0.5541, 0.3135, 0.0731, 0.0394, 0.0092, 0.1858, 0.4045, 0.0007, 0.4332, 0.1246, 0.2438, 0.5101, 0.6431, 0.1825, 0.0583, 0.1229, 0.2452, 0.0393, 0.1364, 0.3828, 0.1906 ...
End of preview. Expand in Data Studio

Mamba S6 Long-Context Recall Benchmarks (mamba-long-context-recall-benchmarks)

A benchmark dataset featuring 1,000 algorithmic long-context sequence evaluation trajectories comparing Mamba Selective State Space Models (S6) against standard Quadratic Attention Transformers across sequence lengths up to 4,096 tokens.

Dataset Structure

The benchmark evaluates 4 canonical algorithmic capabilities:

  1. selective_associative_recall: Extracting target key-value associations across variable distractor noise spans.
  2. variable_length_copying: Remembering and faithfully reproducing prefix sequence tokens after long delays.
  3. induction_heads_induction: Detecting and extrapolating repetitive pattern structures across long contexts ($A \dots B \dots A \to B$).
  4. noise_filtering_retrieval: Selectively gating out uninformative background tokens while preserving memory states.

Features Schema

Column Name Type Description
sample_id int64 Unique sequence evaluation index $[0, 999]$
task_type string Benchmark task category
sequence_length int64 Context length $L \in [256, 4096]$
state_dimension int64 Latent SSM state dimension $d_{\text{state}} \in [8, 16, 32]$
mean_delta_gate float64 Average input-dependent discretization step size $\Delta_t$
selective_sparsity_ratio float64 Proportion of tokens dynamically gated/suppressed
mamba_latency_ms float64 Wall-clock latency for linear Mamba S6 forward pass (ms)
transformer_latency_ms float64 Wall-clock latency for quadratic self-attention (ms)
speedup_factor float64 Empirical acceleration ratio ($t_{\text{attn}} / t_{\text{mamba}}$)
vram_reduction_factor float64 Memory footprint reduction factor
task_accuracy float64 Token retrieval / prediction accuracy
initial_token_entropy float64 Shannon entropy of input sequence distribution
state_norm_final float64 Frobenius norm of final hidden memory state $|h_L|_F$
token_activation_profile list<float64> 32-element profile of activation intensities

Quickstart with Pandas / PyArrow

import pandas as pd

df = pd.read_parquet("train.parquet")
print(f"Loaded {len(df)} sequences across context lengths {df['sequence_length'].unique()}.")
print(df.groupby("sequence_length")[["mamba_latency_ms", "transformer_latency_ms", "speedup_factor"]].mean())

Attribution & Links

Downloads last month
-

Collection including At1uri/mamba-long-context-recall-benchmarks