Hugging Face's logo

PKU-Alignment
/
SAE-V

Model card Files
xet
Community
htlou commited on
Commit
37cd7c5
·
verified ·
1 Parent(s): 298c4a7

Upload folder using huggingface_hub

Browse files
Files changed (27) hide show
  1. .gitattributes +24 -0
  2. README.md +145 -0
  3. SAE-LlaVA-NeXT-7b-Pile/.gitattributes +58 -0
  4. SAE-LlaVA-NeXT-7b-Pile/README.md +2 -0
  5. SAE-LlaVA-NeXT-7b-Pile/cfg.json +1 -0
  6. SAE-LlaVA-NeXT-7b-Pile/sae_weights.safetensors +3 -0
  7. SAE-LlaVA-NeXT-7b-Pile/sparsity.safetensors +3 -0
  8. SAE-Mistral-7b-v0.2-Pile/.gitattributes +58 -0
  9. SAE-Mistral-7b-v0.2-Pile/README.md +2 -0
  10. SAE-Mistral-7b-v0.2-Pile/cfg.json +1 -0
  11. SAE-Mistral-7b-v0.2-Pile/sae_weights.safetensors +3 -0
  12. SAE-Mistral-7b-v0.2-Pile/sparsity.safetensors +3 -0
  13. SAEV_Anole-7b_OBELICS/.gitattributes +35 -0
  14. SAEV_Anole-7b_OBELICS/README.md +2 -0
  15. SAEV_Anole-7b_OBELICS/cfg.json +1 -0
  16. SAEV_Anole-7b_OBELICS/sae_weights.safetensors +3 -0
  17. SAEV_Anole-7b_OBELICS/sparsity.safetensors +3 -0
  18. SAEV_Chameleon-7b_OBELICS/.gitattributes +35 -0
  19. SAEV_Chameleon-7b_OBELICS/README.md +2 -0
  20. SAEV_Chameleon-7b_OBELICS/cfg.json +1 -0
  21. SAEV_Chameleon-7b_OBELICS/sae_weights.safetensors +3 -0
  22. SAEV_Chameleon-7b_OBELICS/sparsity.safetensors +3 -0
  23. SAEV_LLaVA_NeXT-7b_OBELICS/.gitattributes +58 -0
  24. SAEV_LLaVA_NeXT-7b_OBELICS/README.md +2 -0
  25. SAEV_LLaVA_NeXT-7b_OBELICS/cfg.json +1 -0
  26. SAEV_LLaVA_NeXT-7b_OBELICS/sae_weights.safetensors +3 -0
  27. SAEV_LLaVA_NeXT-7b_OBELICS/sparsity.safetensors +3 -0
.gitattributes CHANGED
@@ -8,6 +8,8 @@
8
  *.h5 filter=lfs diff=lfs merge=lfs -text
9
  *.joblib filter=lfs diff=lfs merge=lfs -text
10
  *.lfs.* filter=lfs diff=lfs merge=lfs -text
 
 
11
  *.mlmodel filter=lfs diff=lfs merge=lfs -text
12
  *.model filter=lfs diff=lfs merge=lfs -text
13
  *.msgpack filter=lfs diff=lfs merge=lfs -text
@@ -33,3 +35,25 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8
  *.h5 filter=lfs diff=lfs merge=lfs -text
9
  *.joblib filter=lfs diff=lfs merge=lfs -text
10
  *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.lz4 filter=lfs diff=lfs merge=lfs -text
12
+ *.mds filter=lfs diff=lfs merge=lfs -text
13
  *.mlmodel filter=lfs diff=lfs merge=lfs -text
14
  *.model filter=lfs diff=lfs merge=lfs -text
15
  *.msgpack filter=lfs diff=lfs merge=lfs -text
 
35
  *.zip filter=lfs diff=lfs merge=lfs -text
36
  *.zst filter=lfs diff=lfs merge=lfs -text
37
  *tfevents* filter=lfs diff=lfs merge=lfs -text
38
+ # Audio files - uncompressed
39
+ *.pcm filter=lfs diff=lfs merge=lfs -text
40
+ *.sam filter=lfs diff=lfs merge=lfs -text
41
+ *.raw filter=lfs diff=lfs merge=lfs -text
42
+ # Audio files - compressed
43
+ *.aac filter=lfs diff=lfs merge=lfs -text
44
+ *.flac filter=lfs diff=lfs merge=lfs -text
45
+ *.mp3 filter=lfs diff=lfs merge=lfs -text
46
+ *.ogg filter=lfs diff=lfs merge=lfs -text
47
+ *.wav filter=lfs diff=lfs merge=lfs -text
48
+ # Image files - uncompressed
49
+ *.bmp filter=lfs diff=lfs merge=lfs -text
50
+ *.gif filter=lfs diff=lfs merge=lfs -text
51
+ *.png filter=lfs diff=lfs merge=lfs -text
52
+ *.tiff filter=lfs diff=lfs merge=lfs -text
53
+ # Image files - compressed
54
+ *.jpg filter=lfs diff=lfs merge=lfs -text
55
+ *.jpeg filter=lfs diff=lfs merge=lfs -text
56
+ *.webp filter=lfs diff=lfs merge=lfs -text
57
+ # Video files - compressed
58
+ *.mp4 filter=lfs diff=lfs merge=lfs -text
59
+ *.webm filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,145 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # SAE-V
2
+ Repository for SAE-V, including 2 sparse autoencoder (SAE) and 3 sparse autoencoder with Vision (SAE-V). See each model folders for more information.
3
+
4
+ ## 1.Training Parameter
5
+
6
+ All 5 models training paramters are list below:
7
+
8
+ <table border="1" style="border-collapse: collapse;">
9
+ <thead>
10
+ <tr>
11
+ <th><strong>Hyper-parameters</strong></th>
12
+ <th><strong>SAE and SAE-V of LLaVA-NeXT/Mistral</strong></th>
13
+ <th><strong>SAE and SAE-V of Chameleon/Anole</strong></th>
14
+ </tr>
15
+ </thead>
16
+ <tbody>
17
+ <tr>
18
+ <td colspan="3" style="text-align: center; border-left: none; border-right: none;"><strong>Training Parameters</strong></td>
19
+ </tr>
20
+ <tr>
21
+ <td>total training steps</td>
22
+ <td>30000</td>
23
+ <td>30000</td>
24
+ </tr>
25
+ <tr>
26
+ <td>batch size</td>
27
+ <td>4096</td>
28
+ <td>4096</td>
29
+ </tr>
30
+ <tr>
31
+ <td>LR</td>
32
+ <td>5e-5</td>
33
+ <td>5e-5</td>
34
+ </tr>
35
+ <tr>
36
+ <td>LR warmup steps</td>
37
+ <td>1500</td>
38
+ <td>1500</td>
39
+ </tr>
40
+ <tr>
41
+ <td>LR decay steps</td>
42
+ <td>6000</td>
43
+ <td>6000</td>
44
+ </tr>
45
+ <tr>
46
+ <td>adam beta1</td>
47
+ <td>0.9</td>
48
+ <td>0.9</td>
49
+ </tr>
50
+ <tr>
51
+ <td>adam beta2</td>
52
+ <td>0.999</td>
53
+ <td>0.999</td>
54
+ </tr>
55
+ <tr>
56
+ <td>LR scheduler name</td>
57
+ <td>constant</td>
58
+ <td>constant</td>
59
+ </tr>
60
+ <tr>
61
+ <td>LR coefficient</td>
62
+ <td>5</td>
63
+ <td>5</td>
64
+ </tr>
65
+ <tr>
66
+ <td>seed</td>
67
+ <td>42</td>
68
+ <td>42</td>
69
+ </tr>
70
+ <tr>
71
+ <td>dtype</td>
72
+ <td>float32</td>
73
+ <td>float32</td>
74
+ </tr>
75
+ <tr>
76
+ <td>buffer batches num</td>
77
+ <td>32</td>
78
+ <td>64</td>
79
+ </tr>
80
+ <tr>
81
+ <td>store batch size prompts</td>
82
+ <td>4</td>
83
+ <td>16</td>
84
+ </tr>
85
+ <tr>
86
+ <td>feature sampling window</td>
87
+ <td>1000</td>
88
+ <td>1000</td>
89
+ </tr>
90
+ <tr>
91
+ <td>dead feature window</td>
92
+ <td>1000</td>
93
+ <td>1000</td>
94
+ </tr>
95
+ <tr>
96
+ <td>dead feature threshold</td>
97
+ <td>1e-4</td>
98
+ <td>1e-4</td>
99
+ </tr>
100
+ <!-- "SAE and SAE-V Parameters" row without vertical lines between columns -->
101
+ <tr>
102
+ <td colspan="3" style="text-align: center; border-left: none; border-right: none;"><strong>Model Parameters</strong></td>
103
+ </tr>
104
+ <tr>
105
+ <td>hook layer</td>
106
+ <td>16</td>
107
+ <td>8</td>
108
+ </tr>
109
+ <tr>
110
+ <td>input dimension</td>
111
+ <td>4096</td>
112
+ <td>4096</td>
113
+ </tr>
114
+ <tr>
115
+ <td>expansion factor</td>
116
+ <td>16</td>
117
+ <td>32</td>
118
+ </tr>
119
+ <tr>
120
+ <td>feature number</td>
121
+ <td>65536</td>
122
+ <td>131072</td>
123
+ </tr>
124
+ <tr>
125
+ <td>context size</td>
126
+ <td>4096</td>
127
+ <td>2048</td>
128
+ </tr>
129
+ </tbody>
130
+ </table>
131
+
132
+ The differences in training parameters arise because the LLaVA-NeXT-7B model requires more GPU memory to handle vision input, so fewer batches can be cached. For the SAE and SAE-V parameters, we set different hook layers and context sizes based on the distinct architectures of the two models. We also experimented with different feature numbers on both models, but found that only around 30,000 features are actually activated during training. All training runs were conducted until convergence. All SAE and SAE-V training is performed on 8xA800 GPUs. We ensured that the variations in the parameters did not affect the experiment results.
133
+
134
+ ## 2. Quickstart
135
+
136
+ The SAE and SAE-V is developed based on [SAELens-V](https://github.com/saev-2025/SAELens-V). The loading example is as follow:
137
+
138
+ ```python
139
+ from saev_lens import SAE
140
+ sae = SAE.load_from_pretrained(
141
+ path = "./SAEV_LLaVA_NeXT-7b_OBELICS",
142
+ device ="cuda:0"
143
+ )
144
+ ```
145
+ More using tutorial is presented in [SAELens-V](https://github.com/saev-2025/SAELens-V).
SAE-LlaVA-NeXT-7b-Pile/.gitattributes ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
5
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.lz4 filter=lfs diff=lfs merge=lfs -text
12
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
13
+ *.model filter=lfs diff=lfs merge=lfs -text
14
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
15
+ *.npy filter=lfs diff=lfs merge=lfs -text
16
+ *.npz filter=lfs diff=lfs merge=lfs -text
17
+ *.onnx filter=lfs diff=lfs merge=lfs -text
18
+ *.ot filter=lfs diff=lfs merge=lfs -text
19
+ *.parquet filter=lfs diff=lfs merge=lfs -text
20
+ *.pb filter=lfs diff=lfs merge=lfs -text
21
+ *.pickle filter=lfs diff=lfs merge=lfs -text
22
+ *.pkl filter=lfs diff=lfs merge=lfs -text
23
+ *.pt filter=lfs diff=lfs merge=lfs -text
24
+ *.pth filter=lfs diff=lfs merge=lfs -text
25
+ *.rar filter=lfs diff=lfs merge=lfs -text
26
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
27
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
28
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
29
+ *.tar filter=lfs diff=lfs merge=lfs -text
30
+ *.tflite filter=lfs diff=lfs merge=lfs -text
31
+ *.tgz filter=lfs diff=lfs merge=lfs -text
32
+ *.wasm filter=lfs diff=lfs merge=lfs -text
33
+ *.xz filter=lfs diff=lfs merge=lfs -text
34
+ *.zip filter=lfs diff=lfs merge=lfs -text
35
+ *.zst filter=lfs diff=lfs merge=lfs -text
36
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
37
+ # Audio files - uncompressed
38
+ *.pcm filter=lfs diff=lfs merge=lfs -text
39
+ *.sam filter=lfs diff=lfs merge=lfs -text
40
+ *.raw filter=lfs diff=lfs merge=lfs -text
41
+ # Audio files - compressed
42
+ *.aac filter=lfs diff=lfs merge=lfs -text
43
+ *.flac filter=lfs diff=lfs merge=lfs -text
44
+ *.mp3 filter=lfs diff=lfs merge=lfs -text
45
+ *.ogg filter=lfs diff=lfs merge=lfs -text
46
+ *.wav filter=lfs diff=lfs merge=lfs -text
47
+ # Image files - uncompressed
48
+ *.bmp filter=lfs diff=lfs merge=lfs -text
49
+ *.gif filter=lfs diff=lfs merge=lfs -text
50
+ *.png filter=lfs diff=lfs merge=lfs -text
51
+ *.tiff filter=lfs diff=lfs merge=lfs -text
52
+ # Image files - compressed
53
+ *.jpg filter=lfs diff=lfs merge=lfs -text
54
+ *.jpeg filter=lfs diff=lfs merge=lfs -text
55
+ *.webp filter=lfs diff=lfs merge=lfs -text
56
+ # Video files - compressed
57
+ *.mp4 filter=lfs diff=lfs merge=lfs -text
58
+ *.webm filter=lfs diff=lfs merge=lfs -text
SAE-LlaVA-NeXT-7b-Pile/README.md ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ # SAE-LlaVA-NeXT-7b-Pile
2
+ SAE-LlaVA-NeXt-Mistral-Pile is a sparse autoencoder (SAE) based on [LLaVA-NeXT](https://huggingface.co/llava-hf/llava-v1.6-mistral-7b-hf) model and trained on [The Pile](https://huggingface.co/datasets/EleutherAI/pile) dataset.
SAE-LlaVA-NeXT-7b-Pile/cfg.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"model_name": "llava-hf/llava-v1.6-mistral-7b-hf", "local_model_path": "", "model_class_name": "HookedLlava", "hook_name": "blocks.16.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 16, "hook_head_index": null, "dataset_path": "", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 2048, "use_cached_activations": false, "cached_activations_path": null, "architecture": "standard", "d_in": 4096, "d_sae": 65536, "b_dec_init_method": "zeros", "expansion_factor": 16, "activation_fn": "relu", "activation_fn_kwargs": {}, "normalize_sae_decoder": false, "noise_scale": 0.0, "from_pretrained_path": null, "apply_b_dec_to_input": false, "decoder_orthogonal_init": false, "decoder_heuristic_init": true, "init_encoder_as_decoder_transpose": true, "n_batches_in_buffer": 64, "training_tokens": 163840000, "finetuning_tokens": 0, "store_batch_size_prompts": 4, "train_batch_size_tokens": 4096, "normalize_activations": "expected_average_only_in", "device": "cuda:3", "act_store_device": "cuda:3", "seed": 42, "dtype": "float32", "prepend_bos": true, "autocast": false, "autocast_lm": false, "compile_llm": false, "llm_compilation_mode": null, "compile_sae": false, "sae_compilation_mode": null, "adam_beta1": 0.9, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 5, "lp_norm": 1.0, "scale_sparsity_penalty_by_decoder_norm": true, "l1_warm_up_steps": 2000, "lr": 5e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 5e-06, "lr_decay_steps": 8000, "n_restart_cycles": 1, "finetuning_method": null, "use_ghost_grads": false, "feature_sampling_window": 1000, "dead_feature_window": 1000, "dead_feature_threshold": 0.0001, "n_eval_batches": 10, "eval_batch_size_prompts": null, "log_to_wandb": true, "log_activations_store_to_wandb": false, "log_optimizer_state_to_wandb": false, "wandb_project": "interp", "wandb_id": null, "run_name": "65536-L1-5-LR-5e-05-Tokens-1.638e+08", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 20, "resume": false, "n_checkpoints": 20, "checkpoint_path": "checkpoints/xepk4xea", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {"n_devices": 3}, "sae_lens_version": "3.20.0", "sae_lens_training_version": "3.20.0", "vision": false, "tokens_per_buffer": 536870912}
SAE-LlaVA-NeXT-7b-Pile/sae_weights.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5d8b1d505384f6bf74113dde3e5506468d366320ca90932c08efa0ac68dde4a1
3
+ size 2147762504
SAE-LlaVA-NeXT-7b-Pile/sparsity.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9c55970c8e5a624d294e7cbe7a777e9a1c3ee18b25ae5ac2ba4b7b7aa40573ee
3
+ size 262224
SAE-Mistral-7b-v0.2-Pile/.gitattributes ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
5
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.lz4 filter=lfs diff=lfs merge=lfs -text
12
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
13
+ *.model filter=lfs diff=lfs merge=lfs -text
14
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
15
+ *.npy filter=lfs diff=lfs merge=lfs -text
16
+ *.npz filter=lfs diff=lfs merge=lfs -text
17
+ *.onnx filter=lfs diff=lfs merge=lfs -text
18
+ *.ot filter=lfs diff=lfs merge=lfs -text
19
+ *.parquet filter=lfs diff=lfs merge=lfs -text
20
+ *.pb filter=lfs diff=lfs merge=lfs -text
21
+ *.pickle filter=lfs diff=lfs merge=lfs -text
22
+ *.pkl filter=lfs diff=lfs merge=lfs -text
23
+ *.pt filter=lfs diff=lfs merge=lfs -text
24
+ *.pth filter=lfs diff=lfs merge=lfs -text
25
+ *.rar filter=lfs diff=lfs merge=lfs -text
26
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
27
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
28
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
29
+ *.tar filter=lfs diff=lfs merge=lfs -text
30
+ *.tflite filter=lfs diff=lfs merge=lfs -text
31
+ *.tgz filter=lfs diff=lfs merge=lfs -text
32
+ *.wasm filter=lfs diff=lfs merge=lfs -text
33
+ *.xz filter=lfs diff=lfs merge=lfs -text
34
+ *.zip filter=lfs diff=lfs merge=lfs -text
35
+ *.zst filter=lfs diff=lfs merge=lfs -text
36
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
37
+ # Audio files - uncompressed
38
+ *.pcm filter=lfs diff=lfs merge=lfs -text
39
+ *.sam filter=lfs diff=lfs merge=lfs -text
40
+ *.raw filter=lfs diff=lfs merge=lfs -text
41
+ # Audio files - compressed
42
+ *.aac filter=lfs diff=lfs merge=lfs -text
43
+ *.flac filter=lfs diff=lfs merge=lfs -text
44
+ *.mp3 filter=lfs diff=lfs merge=lfs -text
45
+ *.ogg filter=lfs diff=lfs merge=lfs -text
46
+ *.wav filter=lfs diff=lfs merge=lfs -text
47
+ # Image files - uncompressed
48
+ *.bmp filter=lfs diff=lfs merge=lfs -text
49
+ *.gif filter=lfs diff=lfs merge=lfs -text
50
+ *.png filter=lfs diff=lfs merge=lfs -text
51
+ *.tiff filter=lfs diff=lfs merge=lfs -text
52
+ # Image files - compressed
53
+ *.jpg filter=lfs diff=lfs merge=lfs -text
54
+ *.jpeg filter=lfs diff=lfs merge=lfs -text
55
+ *.webp filter=lfs diff=lfs merge=lfs -text
56
+ # Video files - compressed
57
+ *.mp4 filter=lfs diff=lfs merge=lfs -text
58
+ *.webm filter=lfs diff=lfs merge=lfs -text
SAE-Mistral-7b-v0.2-Pile/README.md ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ # SAE-Mistral-7b-v0.2-Pile
2
+ SAE-Mistral-7b-v0.2-Pile is a sparse autoencoder (SAE) based on [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2) model and trained on [The Pile](https://huggingface.co/datasets/EleutherAI/pile) dataset.
SAE-Mistral-7b-v0.2-Pile/cfg.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"model_name": "mistralai/Mistral-7B-Instruct-v0.2", "local_model_path": "", "model_class_name": "HookedLlava", "hook_name": "blocks.16.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 16, "hook_head_index": null, "dataset_path": "", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 4096, "use_cached_activations": false, "cached_activations_path": null, "architecture": "standard", "d_in": 4096, "d_sae": 65536, "b_dec_init_method": "zeros", "expansion_factor": 16, "activation_fn": "relu", "activation_fn_kwargs": {}, "normalize_sae_decoder": false, "noise_scale": 0.0, "from_pretrained_path": null, "apply_b_dec_to_input": false, "decoder_orthogonal_init": false, "decoder_heuristic_init": true, "init_encoder_as_decoder_transpose": true, "n_batches_in_buffer": 32, "training_tokens": 122880000, "finetuning_tokens": 0, "store_batch_size_prompts": 4, "train_batch_size_tokens": 4096, "normalize_activations": "expected_average_only_in", "device": "cuda:2", "act_store_device": "cuda:2", "seed": 42, "dtype": "float32", "prepend_bos": true, "autocast": false, "autocast_lm": false, "compile_llm": false, "llm_compilation_mode": null, "compile_sae": false, "sae_compilation_mode": null, "adam_beta1": 0.9, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 5, "lp_norm": 1.0, "scale_sparsity_penalty_by_decoder_norm": true, "l1_warm_up_steps": 1500, "lr": 5e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 5e-06, "lr_decay_steps": 6000, "n_restart_cycles": 1, "finetuning_method": null, "use_ghost_grads": false, "feature_sampling_window": 1000, "dead_feature_window": 1000, "dead_feature_threshold": 0.0001, "n_eval_batches": 10, "eval_batch_size_prompts": null, "log_to_wandb": true, "log_activations_store_to_wandb": false, "log_optimizer_state_to_wandb": false, "wandb_project": "interp-M", "wandb_id": null, "run_name": "65536-L1-5-LR-5e-05-Tokens-1.229e+08", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 20, "resume": false, "n_checkpoints": 20, "checkpoint_path": "checkpoints-M/am4f73zh", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {"n_devices": 2}, "sae_lens_version": "3.20.0", "sae_lens_training_version": "3.20.0", "vision": false, "tokens_per_buffer": 536870912}
SAE-Mistral-7b-v0.2-Pile/sae_weights.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c5bde7e1c003a106cf21333583f3c031bf19c0f371c3a8dfeecc596e0030bfd1
3
+ size 2147762504
SAE-Mistral-7b-v0.2-Pile/sparsity.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c641290c877f4770106cfae1f18778d8a4a500835306f79ec577b25ab349bcd7
3
+ size 262224
SAEV_Anole-7b_OBELICS/.gitattributes ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
5
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
12
+ *.model filter=lfs diff=lfs merge=lfs -text
13
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
14
+ *.npy filter=lfs diff=lfs merge=lfs -text
15
+ *.npz filter=lfs diff=lfs merge=lfs -text
16
+ *.onnx filter=lfs diff=lfs merge=lfs -text
17
+ *.ot filter=lfs diff=lfs merge=lfs -text
18
+ *.parquet filter=lfs diff=lfs merge=lfs -text
19
+ *.pb filter=lfs diff=lfs merge=lfs -text
20
+ *.pickle filter=lfs diff=lfs merge=lfs -text
21
+ *.pkl filter=lfs diff=lfs merge=lfs -text
22
+ *.pt filter=lfs diff=lfs merge=lfs -text
23
+ *.pth filter=lfs diff=lfs merge=lfs -text
24
+ *.rar filter=lfs diff=lfs merge=lfs -text
25
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
26
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
27
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
28
+ *.tar filter=lfs diff=lfs merge=lfs -text
29
+ *.tflite filter=lfs diff=lfs merge=lfs -text
30
+ *.tgz filter=lfs diff=lfs merge=lfs -text
31
+ *.wasm filter=lfs diff=lfs merge=lfs -text
32
+ *.xz filter=lfs diff=lfs merge=lfs -text
33
+ *.zip filter=lfs diff=lfs merge=lfs -text
34
+ *.zst filter=lfs diff=lfs merge=lfs -text
35
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
SAEV_Anole-7b_OBELICS/README.md ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ # SAEV_Anole-7b_OBELICS
2
+ SAEV_Anole-7b_OBELICS is a sparse autoencoder with Vision (SAE-V) based on [Anole-7b-v0.1](https://huggingface.co/leloy/Anole-7b-v0.1-hf) model and trained on [OBELICS](https://github.com/huggingface/OBELICS) dataset.
SAEV_Anole-7b_OBELICS/cfg.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"model_name": "", "local_model_path": "", "model_class_name": "HookedChameleon", "hook_name": "blocks.16.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 16, "hook_head_index": null, "dataset_path": "", "dataset_trust_remote_code": true, "streaming": false, "is_dataset_tokenized": true, "context_size": 2048, "use_cached_activations": false, "cached_activations_path": null, "architecture": "standard", "d_in": 4096, "d_sae": 131072, "b_dec_init_method": "zeros", "expansion_factor": 32, "activation_fn": "relu", "activation_fn_kwargs": {}, "normalize_sae_decoder": false, "noise_scale": 0.0, "from_pretrained_path": null, "apply_b_dec_to_input": false, "decoder_orthogonal_init": false, "decoder_heuristic_init": true, "init_encoder_as_decoder_transpose": true, "n_batches_in_buffer": 64, "training_tokens": 122880000, "finetuning_tokens": 0, "store_batch_size_prompts": 4, "train_batch_size_tokens": 4096, "normalize_activations": "expected_average_only_in", "device": "cuda:3", "act_store_device": "cpu", "seed": 42, "dtype": "float32", "prepend_bos": true, "autocast": false, "autocast_lm": false, "compile_llm": false, "llm_compilation_mode": null, "compile_sae": false, "sae_compilation_mode": null, "adam_beta1": 0.9, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 5, "lp_norm": 1.0, "scale_sparsity_penalty_by_decoder_norm": true, "l1_warm_up_steps": 1500, "lr": 5e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 5e-06, "lr_decay_steps": 6000, "n_restart_cycles": 1, "finetuning_method": null, "use_ghost_grads": false, "feature_sampling_window": 1000, "dead_feature_window": 1000, "dead_feature_threshold": 0.0001, "n_eval_batches": 10, "eval_batch_size_prompts": null, "log_to_wandb": true, "log_activations_store_to_wandb": false, "log_optimizer_state_to_wandb": false, "wandb_project": "interp", "wandb_id": null, "run_name": "131072-L1-5-LR-5e-05-Tokens-1.229e+08", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 20, "resume": false, "n_checkpoints": 8, "checkpoint_path": "checkpoints/9ujb3fly", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {"n_devices": 3}, "sae_lens_version": "3.20.0", "sae_lens_training_version": "3.20.0", "tokens_per_buffer": 536870912}
SAEV_Anole-7b_OBELICS/sae_weights.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e929ac1c198534535751aea816d57ca96e3f51c81c61753da018d64e803c690f
3
+ size 4295508296
SAEV_Anole-7b_OBELICS/sparsity.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ce7b78c9e222c461a1e349eca69880e8b8df9c2b21011d8a6ef54b8ba8e6afe4
3
+ size 524368
SAEV_Chameleon-7b_OBELICS/.gitattributes ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
5
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
12
+ *.model filter=lfs diff=lfs merge=lfs -text
13
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
14
+ *.npy filter=lfs diff=lfs merge=lfs -text
15
+ *.npz filter=lfs diff=lfs merge=lfs -text
16
+ *.onnx filter=lfs diff=lfs merge=lfs -text
17
+ *.ot filter=lfs diff=lfs merge=lfs -text
18
+ *.parquet filter=lfs diff=lfs merge=lfs -text
19
+ *.pb filter=lfs diff=lfs merge=lfs -text
20
+ *.pickle filter=lfs diff=lfs merge=lfs -text
21
+ *.pkl filter=lfs diff=lfs merge=lfs -text
22
+ *.pt filter=lfs diff=lfs merge=lfs -text
23
+ *.pth filter=lfs diff=lfs merge=lfs -text
24
+ *.rar filter=lfs diff=lfs merge=lfs -text
25
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
26
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
27
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
28
+ *.tar filter=lfs diff=lfs merge=lfs -text
29
+ *.tflite filter=lfs diff=lfs merge=lfs -text
30
+ *.tgz filter=lfs diff=lfs merge=lfs -text
31
+ *.wasm filter=lfs diff=lfs merge=lfs -text
32
+ *.xz filter=lfs diff=lfs merge=lfs -text
33
+ *.zip filter=lfs diff=lfs merge=lfs -text
34
+ *.zst filter=lfs diff=lfs merge=lfs -text
35
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
SAEV_Chameleon-7b_OBELICS/README.md ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ # SAEV_Chameleon-7b_OBELICS
2
+ SAEV_Anole-7b_OBELICS is a sparse autoencoder with Vision (SAE-V) based on [Chameleon](https://huggingface.co/facebook/chameleon-7b) model and trained on [OBELICS](https://github.com/huggingface/OBELICS) dataset.
SAEV_Chameleon-7b_OBELICS/cfg.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"model_name": "", "local_model_path": "", "model_class_name": "HookedChameleon", "hook_name": "blocks.16.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 16, "hook_head_index": null, "dataset_path": "", "dataset_trust_remote_code": true, "streaming": false, "is_dataset_tokenized": true, "context_size": 2048, "use_cached_activations": false, "cached_activations_path": null, "architecture": "standard", "d_in": 4096, "d_sae": 131072, "b_dec_init_method": "zeros", "expansion_factor": 32, "activation_fn": "relu", "activation_fn_kwargs": {}, "normalize_sae_decoder": false, "noise_scale": 0.0, "from_pretrained_path": null, "apply_b_dec_to_input": false, "decoder_orthogonal_init": false, "decoder_heuristic_init": true, "init_encoder_as_decoder_transpose": true, "n_batches_in_buffer": 64, "training_tokens": 122880000, "finetuning_tokens": 0, "store_batch_size_prompts": 4, "train_batch_size_tokens": 4096, "normalize_activations": "expected_average_only_in", "device": "cuda:1", "act_store_device": "cpu", "seed": 42, "dtype": "float32", "prepend_bos": true, "autocast": false, "autocast_lm": false, "compile_llm": false, "llm_compilation_mode": null, "compile_sae": false, "sae_compilation_mode": null, "adam_beta1": 0.9, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 5, "lp_norm": 1.0, "scale_sparsity_penalty_by_decoder_norm": true, "l1_warm_up_steps": 1500, "lr": 5e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 5e-06, "lr_decay_steps": 6000, "n_restart_cycles": 1, "finetuning_method": null, "use_ghost_grads": false, "feature_sampling_window": 1000, "dead_feature_window": 1000, "dead_feature_threshold": 0.0001, "n_eval_batches": 10, "eval_batch_size_prompts": null, "log_to_wandb": true, "log_activations_store_to_wandb": false, "log_optimizer_state_to_wandb": false, "wandb_project": "interp", "wandb_id": null, "run_name": "131072-L1-5-LR-5e-05-Tokens-1.229e+08", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 20, "resume": false, "n_checkpoints": 8, "checkpoint_path": "checkpoints/d099y7cn", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {"n_devices": 1}, "sae_lens_version": "3.20.0", "sae_lens_training_version": "3.20.0", "tokens_per_buffer": 536870912}
SAEV_Chameleon-7b_OBELICS/sae_weights.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dabe5ffd54c5c9f62da0ac003f3fc13ec72dc1e144d9d62de3f03c2203cd5e9a
3
+ size 4295508296
SAEV_Chameleon-7b_OBELICS/sparsity.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2d7093a37ff0a0ad7c0e15f87b8a0510455ceb7ff0dfccb3c50a60587150f06a
3
+ size 524368
SAEV_LLaVA_NeXT-7b_OBELICS/.gitattributes ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
5
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.lz4 filter=lfs diff=lfs merge=lfs -text
12
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
13
+ *.model filter=lfs diff=lfs merge=lfs -text
14
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
15
+ *.npy filter=lfs diff=lfs merge=lfs -text
16
+ *.npz filter=lfs diff=lfs merge=lfs -text
17
+ *.onnx filter=lfs diff=lfs merge=lfs -text
18
+ *.ot filter=lfs diff=lfs merge=lfs -text
19
+ *.parquet filter=lfs diff=lfs merge=lfs -text
20
+ *.pb filter=lfs diff=lfs merge=lfs -text
21
+ *.pickle filter=lfs diff=lfs merge=lfs -text
22
+ *.pkl filter=lfs diff=lfs merge=lfs -text
23
+ *.pt filter=lfs diff=lfs merge=lfs -text
24
+ *.pth filter=lfs diff=lfs merge=lfs -text
25
+ *.rar filter=lfs diff=lfs merge=lfs -text
26
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
27
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
28
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
29
+ *.tar filter=lfs diff=lfs merge=lfs -text
30
+ *.tflite filter=lfs diff=lfs merge=lfs -text
31
+ *.tgz filter=lfs diff=lfs merge=lfs -text
32
+ *.wasm filter=lfs diff=lfs merge=lfs -text
33
+ *.xz filter=lfs diff=lfs merge=lfs -text
34
+ *.zip filter=lfs diff=lfs merge=lfs -text
35
+ *.zst filter=lfs diff=lfs merge=lfs -text
36
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
37
+ # Audio files - uncompressed
38
+ *.pcm filter=lfs diff=lfs merge=lfs -text
39
+ *.sam filter=lfs diff=lfs merge=lfs -text
40
+ *.raw filter=lfs diff=lfs merge=lfs -text
41
+ # Audio files - compressed
42
+ *.aac filter=lfs diff=lfs merge=lfs -text
43
+ *.flac filter=lfs diff=lfs merge=lfs -text
44
+ *.mp3 filter=lfs diff=lfs merge=lfs -text
45
+ *.ogg filter=lfs diff=lfs merge=lfs -text
46
+ *.wav filter=lfs diff=lfs merge=lfs -text
47
+ # Image files - uncompressed
48
+ *.bmp filter=lfs diff=lfs merge=lfs -text
49
+ *.gif filter=lfs diff=lfs merge=lfs -text
50
+ *.png filter=lfs diff=lfs merge=lfs -text
51
+ *.tiff filter=lfs diff=lfs merge=lfs -text
52
+ # Image files - compressed
53
+ *.jpg filter=lfs diff=lfs merge=lfs -text
54
+ *.jpeg filter=lfs diff=lfs merge=lfs -text
55
+ *.webp filter=lfs diff=lfs merge=lfs -text
56
+ # Video files - compressed
57
+ *.mp4 filter=lfs diff=lfs merge=lfs -text
58
+ *.webm filter=lfs diff=lfs merge=lfs -text
SAEV_LLaVA_NeXT-7b_OBELICS/README.md ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ # SAEV_LLaVA_NeXT-7b_OBELICS
2
+ SAEV_Anole-7b_OBELICS is a sparse autoencoder with Vision (SAE-V) based on [LLaVA-NeXT](https://huggingface.co/llava-hf/llava-v1.6-mistral-7b-hf) model and trained on [OBELICS](https://github.com/huggingface/OBELICS) dataset.
SAEV_LLaVA_NeXT-7b_OBELICS/cfg.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"model_name": "llava-hf/llava-v1.6-mistral-7b-hf", "local_model_path": "", "model_class_name": "HookedLlava", "hook_name": "blocks.16.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 16, "hook_head_index": null, "dataset_path": "", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 4096, "use_cached_activations": false, "cached_activations_path": null, "architecture": "standard", "d_in": 4096, "d_sae": 65536, "b_dec_init_method": "zeros", "expansion_factor": 16, "activation_fn": "relu", "activation_fn_kwargs": {}, "normalize_sae_decoder": false, "noise_scale": 0.0, "from_pretrained_path": null, "apply_b_dec_to_input": false, "decoder_orthogonal_init": false, "decoder_heuristic_init": true, "init_encoder_as_decoder_transpose": true, "n_batches_in_buffer": 32, "training_tokens": 122880000, "finetuning_tokens": 0, "store_batch_size_prompts": 1, "train_batch_size_tokens": 4096, "normalize_activations": "expected_average_only_in", "device": "cuda:3", "act_store_device": "cuda:3", "seed": 42, "dtype": "float32", "prepend_bos": true, "autocast": false, "autocast_lm": false, "compile_llm": false, "llm_compilation_mode": null, "compile_sae": false, "sae_compilation_mode": null, "adam_beta1": 0.9, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 5, "lp_norm": 1.0, "scale_sparsity_penalty_by_decoder_norm": true, "l1_warm_up_steps": 1500, "lr": 5e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 5e-06, "lr_decay_steps": 6000, "n_restart_cycles": 1, "finetuning_method": null, "use_ghost_grads": false, "feature_sampling_window": 1000, "dead_feature_window": 1000, "dead_feature_threshold": 0.0001, "n_eval_batches": 10, "eval_batch_size_prompts": null, "log_to_wandb": true, "log_activations_store_to_wandb": false, "log_optimizer_state_to_wandb": false, "wandb_project": "interp-V", "wandb_id": null, "run_name": "65536-L1-5-LR-5e-05-Tokens-1.229e+08", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 20, "resume": false, "n_checkpoints": 20, "checkpoint_path": "checkpoints-V/kxpk98cr", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {"n_devices": 3}, "sae_lens_version": "3.20.0", "sae_lens_training_version": "3.20.0", "vision": false, "tokens_per_buffer": 536870912}
SAEV_LLaVA_NeXT-7b_OBELICS/sae_weights.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2e4f97c02ca09cb9a680ab6306ca65a4d9b65a3eee77fe2fcd151e4f1aa4a850
3
+ size 2147762504
SAEV_LLaVA_NeXT-7b_OBELICS/sparsity.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ccf1916cb6dd3e88553f51f2a5c27b0f9fae9e2d4bee34cd9c1b2eb724b48919
3
+ size 262224