Hugging Face's logo

adamkarvonen
/
mistral_test

Model card Files Community
adamkarvonen commited on
Commit
74b77f6
·
verified ·
1 Parent(s): d45135c

Upload folder using huggingface_hub

Browse files
Files changed (8) hide show
  1. mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_0/ae.pt +3 -0
  2. mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_0/config.json +23 -0
  3. mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_1/ae.pt +3 -0
  4. mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_1/config.json +23 -0
  5. mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_2/ae.pt +3 -0
  6. mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_2/config.json +23 -0
  7. mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_3/ae.pt +3 -0
  8. mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_3/config.json +23 -0
mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_0/ae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8250e990916617702fb11e91c949ca365b79a4c8b11e8dd71ee9a9422d3a8ee8
3
+ size 536954902
mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_0/config.json ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "trainer": {
3
+ "trainer_class": "BatchTopKTrainer",
4
+ "dict_class": "BatchTopKSAE",
5
+ "lr": 5e-05,
6
+ "steps": 61035,
7
+ "auxk_alpha": 0.03125,
8
+ "warmup_steps": 10,
9
+ "decay_start": 48828,
10
+ "threshold_beta": 0.999,
11
+ "threshold_start_step": 1000,
12
+ "top_k_aux": 2048,
13
+ "seed": 0,
14
+ "activation_dim": 4096,
15
+ "dict_size": 16384,
16
+ "k": 80,
17
+ "device": "cuda:7",
18
+ "layer": 9,
19
+ "lm_name": "mistralai/Ministral-8B-Instruct-2410",
20
+ "wandb_name": "BatchTopKTrainer-mistralai/Ministral-8B-Instruct-2410-resid_post_layer_9_trainer_0",
21
+ "submodule_name": "resid_post_layer_9"
22
+ }
23
+ }
mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_1/ae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b578b99727b42e867a99c4291084cee71ff2c7bf993eb0fe78dff814c974f2cf
3
+ size 536954902
mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_1/config.json ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "trainer": {
3
+ "trainer_class": "BatchTopKTrainer",
4
+ "dict_class": "BatchTopKSAE",
5
+ "lr": 5e-05,
6
+ "steps": 61035,
7
+ "auxk_alpha": 0.03125,
8
+ "warmup_steps": 10,
9
+ "decay_start": 48828,
10
+ "threshold_beta": 0.999,
11
+ "threshold_start_step": 1000,
12
+ "top_k_aux": 2048,
13
+ "seed": 0,
14
+ "activation_dim": 4096,
15
+ "dict_size": 16384,
16
+ "k": 160,
17
+ "device": "cuda:7",
18
+ "layer": 9,
19
+ "lm_name": "mistralai/Ministral-8B-Instruct-2410",
20
+ "wandb_name": "BatchTopKTrainer-mistralai/Ministral-8B-Instruct-2410-resid_post_layer_9_trainer_1",
21
+ "submodule_name": "resid_post_layer_9"
22
+ }
23
+ }
mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_2/ae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:786ca5f8625e8644ae7e0d04d37aa3e5d216a06b1ab35508630fbcb2c7cf9bdc
3
+ size 2147764246
mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_2/config.json ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "trainer": {
3
+ "trainer_class": "BatchTopKTrainer",
4
+ "dict_class": "BatchTopKSAE",
5
+ "lr": 5e-05,
6
+ "steps": 61035,
7
+ "auxk_alpha": 0.03125,
8
+ "warmup_steps": 10,
9
+ "decay_start": 48828,
10
+ "threshold_beta": 0.999,
11
+ "threshold_start_step": 1000,
12
+ "top_k_aux": 2048,
13
+ "seed": 0,
14
+ "activation_dim": 4096,
15
+ "dict_size": 65536,
16
+ "k": 80,
17
+ "device": "cuda:7",
18
+ "layer": 9,
19
+ "lm_name": "mistralai/Ministral-8B-Instruct-2410",
20
+ "wandb_name": "BatchTopKTrainer-mistralai/Ministral-8B-Instruct-2410-resid_post_layer_9_trainer_2",
21
+ "submodule_name": "resid_post_layer_9"
22
+ }
23
+ }
mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_3/ae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6656fbe86ba19e6ef5ddf63fefd2e0a0a5085d87401b8ece5ad81263968ddd68
3
+ size 2147764246
mistral_8b_short_mistralai_Ministral-8B-Instruct-2410_batch_top_k/resid_post_layer_9/trainer_3/config.json ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "trainer": {
3
+ "trainer_class": "BatchTopKTrainer",
4
+ "dict_class": "BatchTopKSAE",
5
+ "lr": 5e-05,
6
+ "steps": 61035,
7
+ "auxk_alpha": 0.03125,
8
+ "warmup_steps": 10,
9
+ "decay_start": 48828,
10
+ "threshold_beta": 0.999,
11
+ "threshold_start_step": 1000,
12
+ "top_k_aux": 2048,
13
+ "seed": 0,
14
+ "activation_dim": 4096,
15
+ "dict_size": 65536,
16
+ "k": 160,
17
+ "device": "cuda:7",
18
+ "layer": 9,
19
+ "lm_name": "mistralai/Ministral-8B-Instruct-2410",
20
+ "wandb_name": "BatchTopKTrainer-mistralai/Ministral-8B-Instruct-2410-resid_post_layer_9_trainer_3",
21
+ "submodule_name": "resid_post_layer_9"
22
+ }
23
+ }