add samelayer_reduce qwen3-4b-dpo
Browse files- v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-dpo/L21-23/activations/activations.pt +3 -0
- v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-dpo/L21-23/checkpoints/final.pt +3 -0
- v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-dpo/L21-23/metrics/training_metrics.json +80 -0
- v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-dpo/L21-23/run_meta.json +34 -0
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-dpo/L21-23/activations/activations.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:348fe9b98db50c005cc6abe9a2eaa5db2f1b849ab66d22f26f2c4f8650cea0e2
|
| 3 |
+
size 1252425669
|
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-dpo/L21-23/checkpoints/final.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:be34b9505cc512da40c9b8276ea66a5d23053371c1382732b7d70c7718e8109a
|
| 3 |
+
size 2517091162
|
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-dpo/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 |
+
3.3375215266645823,
|
| 10 |
+
1.5776150459722684,
|
| 11 |
+
0.8992629916919858,
|
| 12 |
+
0.7114237641912416
|
| 13 |
+
],
|
| 14 |
+
"val_loss": [
|
| 15 |
+
2.578335719582922,
|
| 16 |
+
1.4854628687129594,
|
| 17 |
+
0.8326438353323812,
|
| 18 |
+
0.7736427375159338
|
| 19 |
+
],
|
| 20 |
+
"train_fve_base": [
|
| 21 |
+
0.7638574011201964,
|
| 22 |
+
0.8803312151110527,
|
| 23 |
+
0.8981487480994211,
|
| 24 |
+
0.9074702429548732
|
| 25 |
+
],
|
| 26 |
+
"train_fve_aligned": [
|
| 27 |
+
0.7611979500842386,
|
| 28 |
+
0.8791301332575079,
|
| 29 |
+
0.8974397545729335,
|
| 30 |
+
0.9067315829127279
|
| 31 |
+
],
|
| 32 |
+
"val_fve_base": [
|
| 33 |
+
0.8655412387473421,
|
| 34 |
+
0.8861626695587997,
|
| 35 |
+
0.8942326291693443,
|
| 36 |
+
0.8958897094451944
|
| 37 |
+
],
|
| 38 |
+
"val_fve_aligned": [
|
| 39 |
+
0.8639179383897032,
|
| 40 |
+
0.885466780650054,
|
| 41 |
+
0.8935956602321246,
|
| 42 |
+
0.8952798618696123
|
| 43 |
+
],
|
| 44 |
+
"dead_neurons": [
|
| 45 |
+
0.9294303459377924,
|
| 46 |
+
0.9635147986116983,
|
| 47 |
+
0.9633944812603186,
|
| 48 |
+
0.9630551361856722
|
| 49 |
+
],
|
| 50 |
+
"l0_base": [
|
| 51 |
+
228.7855917980152,
|
| 52 |
+
215.67310639229422,
|
| 53 |
+
212.6975518096906,
|
| 54 |
+
211.5966323701109
|
| 55 |
+
],
|
| 56 |
+
"l0_aligned": [
|
| 57 |
+
228.45552393461762,
|
| 58 |
+
215.50246278458843,
|
| 59 |
+
212.50091214244017,
|
| 60 |
+
211.3654589900759
|
| 61 |
+
],
|
| 62 |
+
"self_recon": [
|
| 63 |
+
2.2875756650762953,
|
| 64 |
+
1.0798564990117963,
|
| 65 |
+
0.5877871077901358,
|
| 66 |
+
0.4515890134438672
|
| 67 |
+
],
|
| 68 |
+
"cross_recon": [
|
| 69 |
+
2.286910035285905,
|
| 70 |
+
1.0785740389854535,
|
| 71 |
+
0.601522126012703,
|
| 72 |
+
0.46665952415126705
|
| 73 |
+
],
|
| 74 |
+
"sparsity": [
|
| 75 |
+
0.1351818600620572,
|
| 76 |
+
0.06632892658713263,
|
| 77 |
+
0.07086702316381815,
|
| 78 |
+
0.0731709359754127
|
| 79 |
+
]
|
| 80 |
+
}
|
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-dpo/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-dpo",
|
| 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-dpo/activations/activations.pt",
|
| 7 |
+
"base_activations_path": "/home/igarg/research/interp-alignment/interp_utils/crosscoder/results-multi-v1/v1/assembled_activations/qwen3-4b-dpo/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 |
+
}
|