Instructions to use chanind/synth-sae-bench-variations with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- SAELens
How to use chanind/synth-sae-bench-variations with SAELens:
# pip install sae-lens from sae_lens import SAE sae, cfg_dict, sparsity = SAE.from_pretrained( release = "RELEASE_ID", # e.g., "gpt2-small-res-jb". See other options in https://github.com/jbloomAus/SAELens/blob/main/sae_lens/pretrained_saes.yaml sae_id = "SAE_ID", # e.g., "blocks.8.hook_resid_pre". Won't always be a hook point ) - Notebooks
- Google Colab
- Kaggle
Delete misc/rand-hierarchy-4-64-l0-142-4kl0-112
Browse files
misc/rand-hierarchy-4-64-l0-142-4kl0-112/hierarchy.json
DELETED
|
The diff for this file is too large to render.
See raw diff
|
|
|
misc/rand-hierarchy-4-64-l0-142-4kl0-112/synthetic_model.safetensors
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:ef0267c2a4f85921a7eec7a3442ed32a26c05f1fab06bda71e5a5c0bf7a5be81
|
| 3 |
-
size 52104608
|
|
|
|
|
|
|
|
|
|
|
|
misc/rand-hierarchy-4-64-l0-142-4kl0-112/synthetic_model_config.json
DELETED
|
@@ -1,47 +0,0 @@
|
|
| 1 |
-
{
|
| 2 |
-
"num_features": 16384,
|
| 3 |
-
"hidden_dim": 768,
|
| 4 |
-
"firing_probability": {
|
| 5 |
-
"max_prob": 0.2,
|
| 6 |
-
"min_prob": 0.05,
|
| 7 |
-
"generator_name": "linear"
|
| 8 |
-
},
|
| 9 |
-
"hierarchy": {
|
| 10 |
-
"total_root_nodes": 128,
|
| 11 |
-
"branching_factor": [
|
| 12 |
-
4,
|
| 13 |
-
64
|
| 14 |
-
],
|
| 15 |
-
"max_depth": 2,
|
| 16 |
-
"mutually_exclusive_portion": 0.5,
|
| 17 |
-
"mutually_exclusive_min_depth": 0,
|
| 18 |
-
"mutually_exclusive_max_depth": null,
|
| 19 |
-
"compensate_probabilities": false,
|
| 20 |
-
"scale_children_by_parent": false
|
| 21 |
-
},
|
| 22 |
-
"orthogonalization": {
|
| 23 |
-
"num_steps": 100,
|
| 24 |
-
"lr": 0.0003,
|
| 25 |
-
"chunk_size": 1024
|
| 26 |
-
},
|
| 27 |
-
"correlation": {
|
| 28 |
-
"rank": 25,
|
| 29 |
-
"correlation_scale": 0.1,
|
| 30 |
-
"seed": null
|
| 31 |
-
},
|
| 32 |
-
"std_firing_magnitudes": {
|
| 33 |
-
"mean": 0.5,
|
| 34 |
-
"std": 0.5,
|
| 35 |
-
"min_value": null,
|
| 36 |
-
"max_value": null,
|
| 37 |
-
"generator_name": "folded_normal"
|
| 38 |
-
},
|
| 39 |
-
"mean_firing_magnitudes": {
|
| 40 |
-
"start": 5.0,
|
| 41 |
-
"end": 4.0,
|
| 42 |
-
"generator_name": "linear"
|
| 43 |
-
},
|
| 44 |
-
"bias": 0.5,
|
| 45 |
-
"dtype": "float32",
|
| 46 |
-
"seed": 42
|
| 47 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|