Hugging Face's logo

Datasets:
MInAlA
/
crosscoder-multilayer-split-activations

Tags:
mechanistic-interpretability
crosscoder
activations
sparse-autoencoder
License:
Dataset card Files
xet
 Community
1
ishangarg183 commited on
Commit
ee51ec3
·
verified ·
1 Parent(s): d920543

add samelayer_reduce qwen3-4b-orpo

Browse files
Files changed (4) hide show
  1. v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-orpo/L21-23/activations/activations.pt +3 -0
  2. v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-orpo/L21-23/checkpoints/final.pt +3 -0
  3. v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-orpo/L21-23/metrics/training_metrics.json +80 -0
  4. v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-orpo/L21-23/run_meta.json +34 -0
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-orpo/L21-23/activations/activations.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:26a458a7898e5f2a7a3b5314b190bf2cb9fec6feeb33a9b9cc497aae55b4f124
3
+ size 1252425669
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-orpo/L21-23/checkpoints/final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1b8429998011e2bc2ac03b7fac6e38aa0abf251494906fd505c8cf1a9785aa70
3
+ size 2517091162
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-orpo/L21-23/metrics/training_metrics.json ADDED
@@ -0,0 +1,80 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epochs": [
3
+ 1,
4
+ 2,
5
+ 3,
6
+ 4
7
+ ],
8
+ "train_loss": [
9
+ 4.194132217975367,
10
+ 2.697435090921806,
11
+ 1.9153036690565555,
12
+ 1.6338303976078445
13
+ ],
14
+ "val_loss": [
15
+ 6.552748238541069,
16
+ 5.720973900163361,
17
+ 4.673560012385483,
18
+ 4.536974454737458
19
+ ],
20
+ "train_fve_base": [
21
+ 0.7607689940017771,
22
+ 0.8761505458832344,
23
+ 0.8943001572945212,
24
+ 0.9039191275191321
25
+ ],
26
+ "train_fve_aligned": [
27
+ 0.7798895121177455,
28
+ 0.8863331410387978,
29
+ 0.9039092412206491,
30
+ 0.9132240051006756
31
+ ],
32
+ "val_fve_base": [
33
+ 0.8612724739219506,
34
+ 0.8818735948407838,
35
+ 0.8898067249677568,
36
+ 0.8913905779728715
37
+ ],
38
+ "val_fve_aligned": [
39
+ 0.8725164941468164,
40
+ 0.8914449130677428,
41
+ 0.8992658504016736,
42
+ 0.9009928768842008
43
+ ],
44
+ "dead_neurons": [
45
+ 0.9227553431360052,
46
+ 0.9545448757164393,
47
+ 0.9549381798356162,
48
+ 0.9546533922876312
49
+ ],
50
+ "l0_base": [
51
+ 231.68498248686515,
52
+ 223.71030356100408,
53
+ 223.64962784588442,
54
+ 223.7739528604787
55
+ ],
56
+ "l0_aligned": [
57
+ 230.3627408056042,
58
+ 224.9152072387624,
59
+ 224.89919001751312,
60
+ 224.9386675423234
61
+ ],
62
+ "self_recon": [
63
+ 2.683090607307838,
64
+ 1.5408020906293873,
65
+ 0.9012226588323808,
66
+ 0.6601085132558793
67
+ ],
68
+ "cross_recon": [
69
+ 3.322288798576061,
70
+ 2.6166024815249846,
71
+ 2.2491691697082254,
72
+ 2.142143298729915
73
+ ],
74
+ "sparsity": [
75
+ 0.18212609443219865,
76
+ 0.10999198875542766,
77
+ 0.1144133195394114,
78
+ 0.11686457124670833
79
+ ]
80
+ }
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-orpo/L21-23/run_meta.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "base_model": "Qwen/Qwen3-4B-Instruct-2507",
3
+ "aligned_run_id": "qwen3-4b-orpo",
4
+ "position": "last_prompt",
5
+ "multi_type": "reduce",
6
+ "source_activations": "/home/igarg/research/interp-alignment/interp_utils/crosscoder/results-multi-v1/v1/assembled_activations/qwen3-4b-orpo/activations/activations.pt",
7
+ "base_activations_path": "/home/igarg/research/interp-alignment/interp_utils/crosscoder/results-multi-v1/v1/assembled_activations/qwen3-4b-orpo/activations/activations.pt",
8
+ "source_layers": [
9
+ 21,
10
+ 22,
11
+ 23
12
+ ],
13
+ "base_source_layers": [
14
+ 21,
15
+ 22,
16
+ 23
17
+ ],
18
+ "layers": [
19
+ 21,
20
+ 22,
21
+ 23
22
+ ],
23
+ "base_layers": [
24
+ 21,
25
+ 22,
26
+ 23
27
+ ],
28
+ "aligned_layers": [
29
+ 21,
30
+ 22,
31
+ 23
32
+ ],
33
+ "layer_reduction": "mean"
34
+ }