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
a27110b
·
verified ·
1 Parent(s): 9c05935

add samelayer_reduce llama32-3b-orpo

Browse files
Files changed (4) hide show
  1. v3-samelayer-seed-99/samelayer_reduce/llama32-3b-orpo/L24-26/activations/activations.pt +3 -0
  2. v3-samelayer-seed-99/samelayer_reduce/llama32-3b-orpo/L24-26/checkpoints/final.pt +3 -0
  3. v3-samelayer-seed-99/samelayer_reduce/llama32-3b-orpo/L24-26/metrics/training_metrics.json +80 -0
  4. v3-samelayer-seed-99/samelayer_reduce/llama32-3b-orpo/L24-26/run_meta.json +34 -0
v3-samelayer-seed-99/samelayer_reduce/llama32-3b-orpo/L24-26/activations/activations.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:df578b47e33b322e8d39ff0a476746a1b682eb14a97c6c9744e609ab58b09f7d
3
+ size 1501941701
v3-samelayer-seed-99/samelayer_reduce/llama32-3b-orpo/L24-26/checkpoints/final.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f364254e23e4530b4b4f4987538ff984fe64d20b83d865d164fbcd103d801fc6
3
+ size 3624487706
v3-samelayer-seed-99/samelayer_reduce/llama32-3b-orpo/L24-26/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
+ 0.5415019516839107,
10
+ 0.3949927885631198,
11
+ 0.36957866738083345,
12
+ 0.3555426879612521
13
+ ],
14
+ "val_loss": [
15
+ 0.412341258132645,
16
+ 0.38601018714655133,
17
+ 0.3743657700678441,
18
+ 0.3718731281645011
19
+ ],
20
+ "train_fve_base": [
21
+ 0.5477810111647103,
22
+ 0.6718696302930584,
23
+ 0.699596045174518,
24
+ 0.7139567328835535
25
+ ],
26
+ "train_fve_aligned": [
27
+ 0.6926669120718983,
28
+ 0.78910743191586,
29
+ 0.8089638271320653,
30
+ 0.8199342274554487
31
+ ],
32
+ "val_fve_base": [
33
+ 0.6485626429163348,
34
+ 0.6784370173334451,
35
+ 0.691269757859994,
36
+ 0.6938060943368842
37
+ ],
38
+ "val_fve_aligned": [
39
+ 0.7797627723653903,
40
+ 0.80022475201422,
41
+ 0.8100096031009215,
42
+ 0.8128517457327917
43
+ ],
44
+ "dead_neurons": [
45
+ 0.9234839297496152,
46
+ 0.9583472019736696,
47
+ 0.9583595293485415,
48
+ 0.9576163236258396
49
+ ],
50
+ "l0_base": [
51
+ 222.56941403969643,
52
+ 214.87538309982486,
53
+ 218.1674146234676,
54
+ 220.14238543490953
55
+ ],
56
+ "l0_aligned": [
57
+ 196.93272037361353,
58
+ 199.17004159369526,
59
+ 210.8103108581436,
60
+ 215.25631202568593
61
+ ],
62
+ "self_recon": [
63
+ 0.3197431033004186,
64
+ 0.22575095002754209,
65
+ 0.2056193405156656,
66
+ 0.19483781643799014
67
+ ],
68
+ "cross_recon": [
69
+ 0.3844892715092507,
70
+ 0.305981599111969,
71
+ 0.2877394074907932,
72
+ 0.2772268031468812
73
+ ],
74
+ "sparsity": [
75
+ 0.06796313556301545,
76
+ 0.046849196221107496,
77
+ 0.04886356190334129,
78
+ 0.04981414866087316
79
+ ]
80
+ }
v3-samelayer-seed-99/samelayer_reduce/llama32-3b-orpo/L24-26/run_meta.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "base_model": "meta-llama/Llama-3.2-3B-Instruct",
3
+ "aligned_run_id": "llama32-3b-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/llama32-3b-orpo/activations/activations.pt",
7
+ "base_activations_path": "/home/igarg/research/interp-alignment/interp_utils/crosscoder/results-multi-v1/v1/assembled_activations/llama32-3b-orpo/activations/activations.pt",
8
+ "source_layers": [
9
+ 24,
10
+ 25,
11
+ 26
12
+ ],
13
+ "base_source_layers": [
14
+ 24,
15
+ 25,
16
+ 26
17
+ ],
18
+ "layers": [
19
+ 24,
20
+ 25,
21
+ 26
22
+ ],
23
+ "base_layers": [
24
+ 24,
25
+ 25,
26
+ 26
27
+ ],
28
+ "aligned_layers": [
29
+ 24,
30
+ 25,
31
+ 26
32
+ ],
33
+ "layer_reduction": "mean"
34
+ }