eoinf commited on
Commit
3546345
·
verified ·
1 Parent(s): c43e460

Initial commit

Browse files
.gitattributes CHANGED
@@ -35,3 +35,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
  pythia-2.8b-deduped_resid_pre_layer_11 filter=lfs diff=lfs merge=lfs -text
37
  pythia-2.8b-deduped_resid_pre_layer_24 filter=lfs diff=lfs merge=lfs -text
 
 
 
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
  pythia-2.8b-deduped_resid_pre_layer_11 filter=lfs diff=lfs merge=lfs -text
37
  pythia-2.8b-deduped_resid_pre_layer_24 filter=lfs diff=lfs merge=lfs -text
38
+ devout-surf-31/hyperparameters.yaml filter=lfs diff=lfs merge=lfs -text
39
+ giddy-morning-30/hyperparameters.yaml filter=lfs diff=lfs merge=lfs -text
chocolate-snow-25/hyperparameters.yaml ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ adam_beta1: 0.9
2
+ adam_beta2: 0.999
3
+ b_dec_init_method: zeros
4
+ cached_activations_path: null
5
+ checkpoint_path: ../outputs/checkpoints
6
+ clip_grad_norm: false
7
+ context_size: 256
8
+ custom_loss: null
9
+ d_in: 2560
10
+ d_out: null
11
+ dataset: Skylion007/openwebtext
12
+ dense_loss_coefficient: 0
13
+ device: cuda
14
+ different_output: false
15
+ dtype: float32
16
+ epsilon_l0_approx: 0.2
17
+ eval_frequency: 500
18
+ expansion_factor: 8
19
+ feature_reinit_scale: 0.2
20
+ feature_resampling_method: null
21
+ flatten_activations_over_layer: false
22
+ flatten_activations_over_layer_output: false
23
+ from_pretrained_path: null
24
+ hook_point: blocks.11.hook_resid_pre
25
+ hook_point_head_index: null
26
+ hook_point_head_index_output: null
27
+ hook_point_layer: 11
28
+ hook_point_layer_output: null
29
+ hook_point_output: null
30
+ initial_decoder_norm: 0.1
31
+ initialise_encoder_to_decoder_transpose: false
32
+ is_dataset_tokenized: false
33
+ l0_coefficient: 9.0e-05
34
+ l0_warmup: false
35
+ l0_warmup_steps: 1000
36
+ l1_coefficient: 0
37
+ l1_warmup: false
38
+ l1_warmup_steps: 1000
39
+ log_to_wandb: true
40
+ loop_dataset: false
41
+ lr: 0.0004
42
+ lr_scheduler_name: constant
43
+ lr_warm_up_steps: 500
44
+ max_resample_step: 100000
45
+ max_sparsity_target: 1
46
+ min_sparsity_for_resample: 0
47
+ min_sparsity_target: 8.0e-06
48
+ model_name: EleutherAI/pythia-2.8b-deduped
49
+ mse_loss_coefficient: 1
50
+ mse_loss_type: centered
51
+ multiple_runs: false
52
+ n_batches_in_store_buffer: 128
53
+ n_checkpoints: 80
54
+ n_running_sparsity: 300
55
+ normalise_initial_decoder_weights: false
56
+ normalise_w_dec: true
57
+ resample_batches: 128
58
+ resample_frequency: 25000
59
+ scale_input_norm: false
60
+ seed: 42
61
+ sparse_loss_coefficient: 1.0e-06
62
+ sparsity_log_frequency: 5000
63
+ store_batch_size: 8
64
+ subtract_b_dec_from_inputs: false
65
+ total_training_steps: 200000
66
+ train_batch_size: 4096
67
+ use_cached_activations: false
68
+ use_gated_sparse_autoencoder: false
69
+ wandb_log_frequency: 10
70
+ wandb_project: test_gemma_2b
71
+ weight_l1_by_decoder_norms: false
devout-surf-31/hyperparameters.yaml ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bec3cd5394d19ad235f88ce3825b5ab897b9d1658fdb5bc7a6c91be67846c962
3
+ size 27635242
devout-surf-31/sparse_autoencoder_EleutherAI/pythia-2.8b-deduped_blocks.11.hook_resid_pre_s20480_0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8f8fe015f01a26d8d42c9e9e5b68837a55ecae1a1011954d04d6b381c5fd2b88
3
+ size 438472512
devout-surf-31/sparse_autoencoder_EleutherAI/pythia-2.8b-deduped_blocks.11.hook_resid_pre_s20480_0_log_feature_sparsity.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d39e928e0a196df623d04e8e33489e53389f7abbcbc14e16dad56088326f68e2
3
+ size 83755
devout-surf-31/sparse_autoencoder_EleutherAI/pythia-2.8b-deduped_blocks.11.hook_resid_pre_s20480_10240000.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d0f81d154e481d330ee7decaaa149b413d4870e0c6c96351bccd8d10d31d48eb
3
+ size 438472632
devout-surf-31/sparse_autoencoder_EleutherAI/pythia-2.8b-deduped_blocks.11.hook_resid_pre_s20480_20480000.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a41f70df434f664b7eaba3895e6a2549e72b868fd94548b45cc558d1a05a6183
3
+ size 438472632
devout-surf-31/sparse_autoencoder_EleutherAI/pythia-2.8b-deduped_blocks.11.hook_resid_pre_s20480_20480000_log_feature_sparsity.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2fab99adf5ad7d7a43ac05e4f3eddddf9254fc57f118513820ce467da36aed71
3
+ size 83790
devout-surf-31/sparse_autoencoder_EleutherAI/pythia-2.8b-deduped_blocks.11.hook_resid_pre_s20480_30720000.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:79b9abb147dd062cc7fdb8fe4608d112b5557a6ee83c55162eb2f5d5c8ec8904
3
+ size 438472632
devout-surf-31/sparse_autoencoder_EleutherAI/pythia-2.8b-deduped_blocks.11.hook_resid_pre_s20480_40960000.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bb46e4364ecdf4ce0380f25ec26170f44180b7c3161b7de1f586cf31a291ce6f
3
+ size 438472632
devout-surf-31/sparse_autoencoder_EleutherAI/pythia-2.8b-deduped_blocks.11.hook_resid_pre_s20480_40960000_log_feature_sparsity.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f9ed250b3bce8607715d178a4075c38a2b611261095d6c9963a531749d593953
3
+ size 83790
giddy-morning-30/hyperparameters.yaml ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bec3cd5394d19ad235f88ce3825b5ab897b9d1658fdb5bc7a6c91be67846c962
3
+ size 27635242