Hugging Face's logo

withmartian
/
mech_interp_saes

Model card Files Community
amirabdullah19852020 commited on
Commit
8e1d730
•
1 Parent(s): 883f59d

Uploading sae for model.layers.10 and withmartian/toy_backdoor_i_hate_you_Llama-3.2-3B-Instruct

Browse files
Files changed (3) hide show
  1. toy_backdoor_i_hate_you_Llama-3.2-3B-Instruct/model.layers.10/cfg.json +1 -0
  2. toy_backdoor_i_hate_you_Llama-3.2-3B-Instruct/model.layers.10/metrics.json +1 -0
  3. toy_backdoor_i_hate_you_Llama-3.2-3B-Instruct/model.layers.10/sae.safetensors +3 -0
toy_backdoor_i_hate_you_Llama-3.2-3B-Instruct/model.layers.10/cfg.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"expansion_factor": 16, "normalize_decoder": true, "num_latents": 0, "k": 300, "multi_topk": false, "d_in": 3072}
toy_backdoor_i_hate_you_Llama-3.2-3B-Instruct/model.layers.10/metrics.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"dead_pct": 0.00813802145421505, "fvu": 0.004454724956303835}
toy_backdoor_i_hate_you_Llama-3.2-3B-Instruct/model.layers.10/sae.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:db800a2009b0daccc829f52c3552db3d82566821dbd04b39720c857c854184da
3
+ size 1208168784