Hugging Face's logo

adamkarvonen
/
kl_finetunes

Model card Files Community
adamkarvonen commited on
Commit
62a8d47
·
verified ·
1 Parent(s): e44b36d

Add files using upload-large-folder tool

Browse files
Files changed (12) hide show
  1. gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_0/ae.pt +3 -0
  2. gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_0/config.json +31 -0
  3. gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_1/ae.pt +3 -0
  4. gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_1/config.json +31 -0
  5. gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_2/ae.pt +3 -0
  6. gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_2/config.json +31 -0
  7. gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_3/ae.pt +3 -0
  8. gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_3/config.json +31 -0
  9. gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_4/ae.pt +3 -0
  10. gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_4/config.json +31 -0
  11. gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_5/ae.pt +3 -0
  12. gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_5/config.json +31 -0
gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_0/ae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0f630db2f0a617057a91825a2a27a4c30c067521f474a96bd113956b7fffbc58
3
+ size 302066710
gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_0/config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "trainer": {
3
+ "trainer_class": "TopKTrainer",
4
+ "dict_class": "AutoEncoderTopK",
5
+ "lr": 0.0003,
6
+ "steps": 244140,
7
+ "auxk_alpha": 0.03125,
8
+ "warmup_steps": 1000,
9
+ "decay_start": 195312,
10
+ "threshold_beta": 0.999,
11
+ "threshold_start_step": 1000,
12
+ "seed": 0,
13
+ "activation_dim": 2304,
14
+ "dict_size": 16384,
15
+ "k": 20,
16
+ "device": "cuda:0",
17
+ "layer": 12,
18
+ "lm_name": "google/gemma-2-2b",
19
+ "wandb_name": "TopKTrainer-google/gemma-2-2b-resid_post_layer_12_trainer_0",
20
+ "submodule_name": "resid_post_layer_12"
21
+ },
22
+ "buffer": {
23
+ "d_submodule": 2304,
24
+ "io": "out",
25
+ "n_ctxs": 244,
26
+ "ctx_len": 1024,
27
+ "refresh_batch_size": 4,
28
+ "out_batch_size": 2048,
29
+ "device": "cuda:0"
30
+ }
31
+ }
gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_1/ae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fc17cd87690337f4eafe6c52833193c223b843539c7b879c8941111c13d33948
3
+ size 302066710
gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_1/config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "trainer": {
3
+ "trainer_class": "TopKTrainer",
4
+ "dict_class": "AutoEncoderTopK",
5
+ "lr": 0.0003,
6
+ "steps": 244140,
7
+ "auxk_alpha": 0.03125,
8
+ "warmup_steps": 1000,
9
+ "decay_start": 195312,
10
+ "threshold_beta": 0.999,
11
+ "threshold_start_step": 1000,
12
+ "seed": 0,
13
+ "activation_dim": 2304,
14
+ "dict_size": 16384,
15
+ "k": 40,
16
+ "device": "cuda:0",
17
+ "layer": 12,
18
+ "lm_name": "google/gemma-2-2b",
19
+ "wandb_name": "TopKTrainer-google/gemma-2-2b-resid_post_layer_12_trainer_1",
20
+ "submodule_name": "resid_post_layer_12"
21
+ },
22
+ "buffer": {
23
+ "d_submodule": 2304,
24
+ "io": "out",
25
+ "n_ctxs": 244,
26
+ "ctx_len": 1024,
27
+ "refresh_batch_size": 4,
28
+ "out_batch_size": 2048,
29
+ "device": "cuda:0"
30
+ }
31
+ }
gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_2/ae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ac11c45b95a20ded698d3e0dacbcc94d7c172a14b35ee9a061232202b8ce8664
3
+ size 302066710
gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_2/config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "trainer": {
3
+ "trainer_class": "TopKTrainer",
4
+ "dict_class": "AutoEncoderTopK",
5
+ "lr": 0.0003,
6
+ "steps": 244140,
7
+ "auxk_alpha": 0.03125,
8
+ "warmup_steps": 1000,
9
+ "decay_start": 195312,
10
+ "threshold_beta": 0.999,
11
+ "threshold_start_step": 1000,
12
+ "seed": 0,
13
+ "activation_dim": 2304,
14
+ "dict_size": 16384,
15
+ "k": 80,
16
+ "device": "cuda:0",
17
+ "layer": 12,
18
+ "lm_name": "google/gemma-2-2b",
19
+ "wandb_name": "TopKTrainer-google/gemma-2-2b-resid_post_layer_12_trainer_2",
20
+ "submodule_name": "resid_post_layer_12"
21
+ },
22
+ "buffer": {
23
+ "d_submodule": 2304,
24
+ "io": "out",
25
+ "n_ctxs": 244,
26
+ "ctx_len": 1024,
27
+ "refresh_batch_size": 4,
28
+ "out_batch_size": 2048,
29
+ "device": "cuda:0"
30
+ }
31
+ }
gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_3/ae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b6dfd0dea92921e13cd7c2a9c595e579147cd5f1bdeaa656fba06f153b8eb49a
3
+ size 302066710
gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_3/config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "trainer": {
3
+ "trainer_class": "TopKTrainer",
4
+ "dict_class": "AutoEncoderTopK",
5
+ "lr": 0.0003,
6
+ "steps": 244140,
7
+ "auxk_alpha": 0.03125,
8
+ "warmup_steps": 1000,
9
+ "decay_start": 195312,
10
+ "threshold_beta": 0.999,
11
+ "threshold_start_step": 1000,
12
+ "seed": 0,
13
+ "activation_dim": 2304,
14
+ "dict_size": 16384,
15
+ "k": 160,
16
+ "device": "cuda:0",
17
+ "layer": 12,
18
+ "lm_name": "google/gemma-2-2b",
19
+ "wandb_name": "TopKTrainer-google/gemma-2-2b-resid_post_layer_12_trainer_3",
20
+ "submodule_name": "resid_post_layer_12"
21
+ },
22
+ "buffer": {
23
+ "d_submodule": 2304,
24
+ "io": "out",
25
+ "n_ctxs": 244,
26
+ "ctx_len": 1024,
27
+ "refresh_batch_size": 4,
28
+ "out_batch_size": 2048,
29
+ "device": "cuda:0"
30
+ }
31
+ }
gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_4/ae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:352e6ab2a8fd83decebd75620af74d9d5ff1418a83ed88aad9ea276c8e58cd21
3
+ size 302066710
gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_4/config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "trainer": {
3
+ "trainer_class": "TopKTrainer",
4
+ "dict_class": "AutoEncoderTopK",
5
+ "lr": 0.0003,
6
+ "steps": 244140,
7
+ "auxk_alpha": 0.03125,
8
+ "warmup_steps": 1000,
9
+ "decay_start": 195312,
10
+ "threshold_beta": 0.999,
11
+ "threshold_start_step": 1000,
12
+ "seed": 0,
13
+ "activation_dim": 2304,
14
+ "dict_size": 16384,
15
+ "k": 320,
16
+ "device": "cuda:0",
17
+ "layer": 12,
18
+ "lm_name": "google/gemma-2-2b",
19
+ "wandb_name": "TopKTrainer-google/gemma-2-2b-resid_post_layer_12_trainer_4",
20
+ "submodule_name": "resid_post_layer_12"
21
+ },
22
+ "buffer": {
23
+ "d_submodule": 2304,
24
+ "io": "out",
25
+ "n_ctxs": 244,
26
+ "ctx_len": 1024,
27
+ "refresh_batch_size": 4,
28
+ "out_batch_size": 2048,
29
+ "device": "cuda:0"
30
+ }
31
+ }
gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_5/ae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7802d39e58fdc375ffb3e062d3aed070542de2b0a03686916944c65da1472337
3
+ size 302066710
gemma-2-2b_top_k_width-2pow14_date-0107/resid_post_layer_12/trainer_5/config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "trainer": {
3
+ "trainer_class": "TopKTrainer",
4
+ "dict_class": "AutoEncoderTopK",
5
+ "lr": 0.0003,
6
+ "steps": 244140,
7
+ "auxk_alpha": 0.03125,
8
+ "warmup_steps": 1000,
9
+ "decay_start": 195312,
10
+ "threshold_beta": 0.999,
11
+ "threshold_start_step": 1000,
12
+ "seed": 0,
13
+ "activation_dim": 2304,
14
+ "dict_size": 16384,
15
+ "k": 640,
16
+ "device": "cuda:0",
17
+ "layer": 12,
18
+ "lm_name": "google/gemma-2-2b",
19
+ "wandb_name": "TopKTrainer-google/gemma-2-2b-resid_post_layer_12_trainer_5",
20
+ "submodule_name": "resid_post_layer_12"
21
+ },
22
+ "buffer": {
23
+ "d_submodule": 2304,
24
+ "io": "out",
25
+ "n_ctxs": 244,
26
+ "ctx_len": 1024,
27
+ "refresh_batch_size": 4,
28
+ "out_batch_size": 2048,
29
+ "device": "cuda:0"
30
+ }
31
+ }