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
276d670
·
verified ·
1 Parent(s): c1aa918

add samelayer_reduce qwen3-4b-grpo

Browse files
Files changed (4) hide show
  1. v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-grpo/L19-21/activations/activations.pt +3 -0
  2. v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-grpo/L19-21/checkpoints/final.pt +3 -0
  3. v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-grpo/L19-21/metrics/training_metrics.json +80 -0
  4. v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-grpo/L19-21/run_meta.json +34 -0
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-grpo/L19-21/activations/activations.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7cb5bb6f5019107a6573b62d665fb9f86312582be10a6b103e8976d1ba742682
3
+ size 1252425669
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-grpo/L19-21/checkpoints/final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3e6329138acd56722a9f8c5174ca91a86e476e74204657ca249b43718fe1c686
3
+ size 2517091162
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-grpo/L19-21/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
+ 3.28041413243723,
10
+ 1.6552151844331693,
11
+ 0.9234442875807005,
12
+ 0.7004394293944876
13
+ ],
14
+ "val_loss": [
15
+ 2.3091629891807495,
16
+ 1.338037808372088,
17
+ 0.6499913561094494,
18
+ 0.5677431450464339
19
+ ],
20
+ "train_fve_base": [
21
+ 0.7628979059039538,
22
+ 0.8780185144962237,
23
+ 0.8950777703116529,
24
+ 0.9036649709129222
25
+ ],
26
+ "train_fve_aligned": [
27
+ 0.7594711212685981,
28
+ 0.8765863077861833,
29
+ 0.8941733665238051,
30
+ 0.9029274943691906
31
+ ],
32
+ "val_fve_base": [
33
+ 0.8643374268297126,
34
+ 0.8847131482593676,
35
+ 0.8926589595085663,
36
+ 0.8942377729565686
37
+ ],
38
+ "val_fve_aligned": [
39
+ 0.8624677842200115,
40
+ 0.8838734558115455,
41
+ 0.8918608779682539,
42
+ 0.8934841577295234
43
+ ],
44
+ "dead_neurons": [
45
+ 0.9321396926725669,
46
+ 0.9649079253780724,
47
+ 0.9640836046929256,
48
+ 0.9634101306981534
49
+ ],
50
+ "l0_base": [
51
+ 228.82328152364272,
52
+ 216.67478473438413,
53
+ 213.72033712784588,
54
+ 212.66486062463514
55
+ ],
56
+ "l0_aligned": [
57
+ 228.24255691768826,
58
+ 216.36724678925862,
59
+ 213.39906231757152,
60
+ 212.3208917104495
61
+ ],
62
+ "self_recon": [
63
+ 2.2322836759420284,
64
+ 1.0755863555136718,
65
+ 0.5284681794974655,
66
+ 0.3477131219649273
67
+ ],
68
+ "cross_recon": [
69
+ 2.366490118853613,
70
+ 1.3315063260465216,
71
+ 0.8616131008001913,
72
+ 0.7516340549167526
73
+ ],
74
+ "sparsity": [
75
+ 0.10153437312964243,
76
+ 0.047026311686198335,
77
+ 0.050330877578091494,
78
+ 0.0520726869603694
79
+ ]
80
+ }
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-grpo/L19-21/run_meta.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "base_model": "Qwen/Qwen3-4B-Instruct-2507",
3
+ "aligned_run_id": "qwen3-4b-grpo",
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-grpo/activations/activations.pt",
7
+ "base_activations_path": "/home/igarg/research/interp-alignment/interp_utils/crosscoder/results-multi-v1/v1/assembled_activations/qwen3-4b-grpo/activations/activations.pt",
8
+ "source_layers": [
9
+ 19,
10
+ 20,
11
+ 21
12
+ ],
13
+ "base_source_layers": [
14
+ 19,
15
+ 20,
16
+ 21
17
+ ],
18
+ "layers": [
19
+ 19,
20
+ 20,
21
+ 21
22
+ ],
23
+ "base_layers": [
24
+ 19,
25
+ 20,
26
+ 21
27
+ ],
28
+ "aligned_layers": [
29
+ 19,
30
+ 20,
31
+ 21
32
+ ],
33
+ "layer_reduction": "mean"
34
+ }