Push model using huggingface_hub.
Browse files- config.json +3 -24
config.json
CHANGED
@@ -1,26 +1,5 @@
|
|
1 |
{
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
"lr": 0.0001,
|
6 |
-
"steps": 97656,
|
7 |
-
"auxk_alpha": 0.03125,
|
8 |
-
"warmup_steps": 1000,
|
9 |
-
"decay_start": null,
|
10 |
-
"threshold_beta": 0.999,
|
11 |
-
"threshold_start_step": 1000,
|
12 |
-
"top_k_aux": 1152,
|
13 |
-
"seed": null,
|
14 |
-
"activation_dim": 2304,
|
15 |
-
"dict_size": 73728,
|
16 |
-
"k": 100,
|
17 |
-
"sparsity_loss_type": "LossType.CROSSCODER",
|
18 |
-
"sparsity_loss_alpha_sae": 1.0,
|
19 |
-
"sparsity_loss_alpha_cc": 0.1,
|
20 |
-
"device": "cuda",
|
21 |
-
"layer": 13,
|
22 |
-
"lm_name": "google/gemma-2-2b-it-google/gemma-2-2b",
|
23 |
-
"wandb_name": "gemma-2-2b-L13-k100-lr1e-04-local-shuffling-CCLoss",
|
24 |
-
"submodule_name": null
|
25 |
-
}
|
26 |
}
|
|
|
1 |
{
|
2 |
+
"activation_dim": 2304,
|
3 |
+
"dict_size": 73728,
|
4 |
+
"num_layers": 2
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
}
|