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
328536c
·
verified ·
1 Parent(s): af8d5b3

add samelayer_reduce qwen3-4b-simpo

Browse files
Files changed (4) hide show
  1. v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-simpo/L20-22/activations/activations.pt +3 -0
  2. v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-simpo/L20-22/checkpoints/final.pt +3 -0
  3. v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-simpo/L20-22/metrics/training_metrics.json +80 -0
  4. v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-simpo/L20-22/run_meta.json +34 -0
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-simpo/L20-22/activations/activations.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4cdd7e5ea5f02a5e461423ab6faf7214261272a887d032195664bed025a3dc22
3
+ size 1252425669
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-simpo/L20-22/checkpoints/final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8cb4edde0353befdd9553196d506cc54c2761daac9dfbe51a11fd67c92334511
3
+ size 2517091162
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-simpo/L20-22/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
+ 3.1704801654370294,
10
+ 1.505906887817856,
11
+ 0.7899075483482481,
12
+ 0.5979754329945564
13
+ ],
14
+ "val_loss": [
15
+ 2.4776868242867955,
16
+ 1.368216071453394,
17
+ 0.7155951383226204,
18
+ 0.6421655895510269
19
+ ],
20
+ "train_fve_base": [
21
+ 0.7624973607425166,
22
+ 0.8788041012090894,
23
+ 0.8962035433145135,
24
+ 0.9051208808557095
25
+ ],
26
+ "train_fve_aligned": [
27
+ 0.7596464340476633,
28
+ 0.8774470295115514,
29
+ 0.8953212007958501,
30
+ 0.9042756406104293
31
+ ],
32
+ "val_fve_base": [
33
+ 0.8645774328895889,
34
+ 0.8853567888599416,
35
+ 0.8934072493258571,
36
+ 0.8950707244623394
37
+ ],
38
+ "val_fve_aligned": [
39
+ 0.86279947720273,
40
+ 0.8844324169358658,
41
+ 0.8925797324530117,
42
+ 0.8942323158548765
43
+ ],
44
+ "dead_neurons": [
45
+ 0.9311422362219014,
46
+ 0.9644511706705199,
47
+ 0.9639129778549675,
48
+ 0.9634123828003299
49
+ ],
50
+ "l0_base": [
51
+ 229.00693228254525,
52
+ 216.254360040864,
53
+ 213.1434982486865,
54
+ 212.03787215411558
55
+ ],
56
+ "l0_aligned": [
57
+ 228.69122154115587,
58
+ 215.95630837711616,
59
+ 212.80567717454758,
60
+ 211.68290280210158
61
+ ],
62
+ "self_recon": [
63
+ 2.1896952121353817,
64
+ 1.0390876749420612,
65
+ 0.5203630114969147,
66
+ 0.3820335275585444
67
+ ],
68
+ "cross_recon": [
69
+ 2.169264988285439,
70
+ 1.0334538519904408,
71
+ 0.5314814370153803,
72
+ 0.39264539375648955
73
+ ],
74
+ "sparsity": [
75
+ 0.11307895726155136,
76
+ 0.05343766945647484,
77
+ 0.056951950296054994,
78
+ 0.0588837432747373
79
+ ]
80
+ }
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-simpo/L20-22/run_meta.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "base_model": "Qwen/Qwen3-4B-Instruct-2507",
3
+ "aligned_run_id": "qwen3-4b-simpo",
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-simpo/activations/activations.pt",
7
+ "base_activations_path": "/home/igarg/research/interp-alignment/interp_utils/crosscoder/results-multi-v1/v1/assembled_activations/qwen3-4b-simpo/activations/activations.pt",
8
+ "source_layers": [
9
+ 20,
10
+ 21,
11
+ 22
12
+ ],
13
+ "base_source_layers": [
14
+ 20,
15
+ 21,
16
+ 22
17
+ ],
18
+ "layers": [
19
+ 20,
20
+ 21,
21
+ 22
22
+ ],
23
+ "base_layers": [
24
+ 20,
25
+ 21,
26
+ 22
27
+ ],
28
+ "aligned_layers": [
29
+ 20,
30
+ 21,
31
+ 22
32
+ ],
33
+ "layer_reduction": "mean"
34
+ }