Inigomf's picture
Initial upload: model weights + mechanistic interpretability analysis + neuron surgery PoC
b0f2899 verified
Raw
History Blame Contribute Delete
2.3 kB
{
"investment_top_layers": [
[
29,
2.232006788253784
],
[
30,
2.114232063293457
],
[
28,
2.0922117233276367
],
[
27,
1.9785888195037842
],
[
26,
1.9130127429962158
],
[
25,
1.8515747785568237
],
[
24,
1.7547423839569092
],
[
23,
1.676640272140503
],
[
22,
1.5357362031936646
],
[
31,
1.5167540311813354
]
],
"caution_top_layers": [
[
28,
2.30419921875
],
[
27,
2.1999988555908203
],
[
26,
2.1429691314697266
],
[
29,
2.142333984375
],
[
25,
2.114550828933716
],
[
24,
1.923681616783142
],
[
23,
1.8480467796325684
],
[
30,
1.75537109375
],
[
22,
1.690255880355835
],
[
1,
1.5375022888183594
]
],
"top_investment_attention_heads": [
[
[
0,
11
],
0.1900270339101553
],
[
[
16,
22
],
0.16320469602942467
],
[
[
0,
2
],
0.13410883583128452
],
[
[
31,
14
],
0.12042155489325523
],
[
[
12,
24
],
0.11490468680858612
],
[
[
15,
0
],
0.10889454744756222
],
[
[
31,
25
],
0.1000001635402441
],
[
[
10,
24
],
0.09960965998470783
],
[
[
15,
26
],
0.09792683459818363
],
[
[
0,
23
],
0.0974971316754818
],
[
[
12,
27
],
0.0947993928566575
],
[
[
12,
25
],
0.09411307144910097
],
[
[
31,
23
],
0.09340370818972588
],
[
[
8,
5
],
0.08770010154694319
],
[
[
26,
27
],
0.0853622518479824
]
],
"model_info": {
"n_layers": 32,
"n_heads": 32,
"d_model": 4096
}
}