add samelayer_reduce qwen3-4b-ppo
Browse files- v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-ppo/L20-22/activations/activations.pt +3 -0
- v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-ppo/L20-22/checkpoints/final.pt +3 -0
- v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-ppo/L20-22/metrics/training_metrics.json +80 -0
- v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-ppo/L20-22/run_meta.json +34 -0
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-ppo/L20-22/activations/activations.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6ffa60ecc373390ea4057432e5f65dedda9b161e4f4ba7d9f40039af71a38fc7
|
| 3 |
+
size 1252425669
|
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-ppo/L20-22/checkpoints/final.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3919a45f75534a48a7dcbf9e019cf4a219bbbd979a10305e6b12f45bdf1bc8a0
|
| 3 |
+
size 2517091162
|
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-ppo/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 |
+
3.136068232770698,
|
| 10 |
+
1.5158578307567392,
|
| 11 |
+
0.814828442556428,
|
| 12 |
+
0.6112140031565161
|
| 13 |
+
],
|
| 14 |
+
"val_loss": [
|
| 15 |
+
2.468959703183299,
|
| 16 |
+
1.4073136433256859,
|
| 17 |
+
0.7347762211767167,
|
| 18 |
+
0.6545009833043782
|
| 19 |
+
],
|
| 20 |
+
"train_fve_base": [
|
| 21 |
+
0.7624281235278453,
|
| 22 |
+
0.8785512962678129,
|
| 23 |
+
0.895928901569647,
|
| 24 |
+
0.9048944492473703
|
| 25 |
+
],
|
| 26 |
+
"train_fve_aligned": [
|
| 27 |
+
0.7587978607092555,
|
| 28 |
+
0.8767403288675343,
|
| 29 |
+
0.8946501876412952,
|
| 30 |
+
0.9037233011735091
|
| 31 |
+
],
|
| 32 |
+
"val_fve_base": [
|
| 33 |
+
0.864462177166764,
|
| 34 |
+
0.8849002241464186,
|
| 35 |
+
0.8930270331692322,
|
| 36 |
+
0.8946845200673448
|
| 37 |
+
],
|
| 38 |
+
"val_fve_aligned": [
|
| 39 |
+
0.8621421634214711,
|
| 40 |
+
0.883700428832888,
|
| 41 |
+
0.8918584332416195,
|
| 42 |
+
0.893530086385018
|
| 43 |
+
],
|
| 44 |
+
"dead_neurons": [
|
| 45 |
+
0.930931616650641,
|
| 46 |
+
0.9643700179201917,
|
| 47 |
+
0.9637729063134268,
|
| 48 |
+
0.963201164398984
|
| 49 |
+
],
|
| 50 |
+
"l0_base": [
|
| 51 |
+
229.05992775831874,
|
| 52 |
+
216.43217308814945,
|
| 53 |
+
213.35496935201402,
|
| 54 |
+
212.23425642148277
|
| 55 |
+
],
|
| 56 |
+
"l0_aligned": [
|
| 57 |
+
228.941860040864,
|
| 58 |
+
216.36730151780503,
|
| 59 |
+
213.20942790426153,
|
| 60 |
+
212.077057793345
|
| 61 |
+
],
|
| 62 |
+
"self_recon": [
|
| 63 |
+
2.163471583935926,
|
| 64 |
+
1.0430334997643382,
|
| 65 |
+
0.5359563333679902,
|
| 66 |
+
0.3895886031877069
|
| 67 |
+
],
|
| 68 |
+
"cross_recon": [
|
| 69 |
+
2.145910840509791,
|
| 70 |
+
1.0457443478019768,
|
| 71 |
+
0.5518866882502371,
|
| 72 |
+
0.40367618451950776
|
| 73 |
+
],
|
| 74 |
+
"sparsity": [
|
| 75 |
+
0.11423225729778992,
|
| 76 |
+
0.05452661513760294,
|
| 77 |
+
0.05811742652518598,
|
| 78 |
+
0.06015492027317719
|
| 79 |
+
]
|
| 80 |
+
}
|
v3-samelayer-seed-99/samelayer_reduce/qwen3-4b-ppo/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-ppo",
|
| 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-ppo/activations/activations.pt",
|
| 7 |
+
"base_activations_path": "/home/igarg/research/interp-alignment/interp_utils/crosscoder/results-multi-v1/v1/assembled_activations/qwen3-4b-ppo/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 |
+
}
|