Upload Meta-Llama-3.1-8B-L16-mu2.1e-02-lr1e-04-local-shuffling-CrosscoderLoss dictionary model
Browse files- trainer_config.json +21 -0
trainer_config.json
ADDED
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"trainer": {
|
3 |
+
"dict_class": "CrossCoder",
|
4 |
+
"trainer_class": "CrossCoderTrainer",
|
5 |
+
"activation_dim": 4096,
|
6 |
+
"dict_size": 131072,
|
7 |
+
"lr": 0.0001,
|
8 |
+
"l1_penalty": 0.021,
|
9 |
+
"warmup_steps": 1000,
|
10 |
+
"resample_steps": null,
|
11 |
+
"device": "cuda",
|
12 |
+
"layer": 16,
|
13 |
+
"lm_name": "meta-llama/Meta-Llama-3.1-8B-Instruct-meta-llama/Meta-Llama-3.1-8B",
|
14 |
+
"wandb_name": "Meta-Llama-3.1-8B-L16-mu2.1e-02-lr1e-04-local-shuffling-CrosscoderLoss",
|
15 |
+
"submodule_name": null,
|
16 |
+
"use_mse_loss": false,
|
17 |
+
"code_normalization": "CROSSCODER",
|
18 |
+
"code_normalization_alpha_sae": 1.0,
|
19 |
+
"code_normalization_alpha_cc": 0.1
|
20 |
+
}
|
21 |
+
}
|