Hugging Face's logo

adamkarvonen
/
kl_finetunes

Model card Files Community
adamkarvonen commited on
Commit
14c522c
·
verified ·
1 Parent(s): 3a50c77

Add files using upload-large-folder tool

Browse files
Files changed (6) hide show
  1. from_scratch_2pow16/trainer_80_checkpoints_475000/ae.pt +3 -0
  2. from_scratch_2pow16/trainer_80_checkpoints_475000/config.json +31 -0
  3. from_scratch_2pow16/trainer_80_peft_sae_from_scratch_0.0/ae.pt +3 -0
  4. from_scratch_2pow16/trainer_80_peft_sae_from_scratch_0.0/config.json +31 -0
  5. from_scratch_2pow16/trainer_80_peft_sae_from_scratch_95.0/ae.pt +3 -0
  6. from_scratch_2pow16/trainer_80_peft_sae_from_scratch_95.0/config.json +31 -0
from_scratch_2pow16/trainer_80_checkpoints_475000/ae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:94252e55ab0ffff8ddd915920a0b2c9b9e105278614abadf6ea5890f7d496ab6
3
+ size 1208232799
from_scratch_2pow16/trainer_80_checkpoints_475000/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": 65536,
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_20",
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
+ }
from_scratch_2pow16/trainer_80_peft_sae_from_scratch_0.0/ae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c284bb7787d24049dd76f4a03b6a6a3531e061417e366f3eb11ac6b39f1fc4a5
3
+ size 1208232799
from_scratch_2pow16/trainer_80_peft_sae_from_scratch_0.0/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": 65536,
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_20",
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
+ }
from_scratch_2pow16/trainer_80_peft_sae_from_scratch_95.0/ae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a7272bd476d8af5248d3fba2977abd2618637838a362c57a436c01960c5bb819
3
+ size 1208232799
from_scratch_2pow16/trainer_80_peft_sae_from_scratch_95.0/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": 65536,
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_20",
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
+ }