Instructions to use pratikdoshi/sparse-autoencoders-1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- SAELens
How to use pratikdoshi/sparse-autoencoders-1 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
Upload SAE blocks.0.hook_mlp_out
Browse files
blocks.0.hook_mlp_out/cfg.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"architecture": "standard", "d_in": 1024, "d_sae": 16384, "dtype": "float32", "device": "cuda", "model_name": "tiny-stories-1L-21M", "hook_name": "blocks.0.hook_mlp_out", "hook_layer": 0, "hook_head_index": null, "activation_fn_str": "relu", "activation_fn_kwargs": {}, "apply_b_dec_to_input": false, "finetuning_scaling_factor": false, "sae_lens_training_version": "3.17.1", "prepend_bos": true, "dataset_path": "apollo-research/roneneldan-TinyStories-tokenizer-gpt2", "dataset_trust_remote_code": true, "context_size": 512, "normalize_activations": "none", "l1_coefficient": 5, "lp_norm": 1.0, "use_ghost_grads": false, "normalize_sae_decoder": false, "noise_scale": 0.0, "decoder_orthogonal_init": false, "init_encoder_as_decoder_transpose": true, "mse_loss_normalization": null, "decoder_heuristic_init": true, "scale_sparsity_penalty_by_decoder_norm": false}
|
blocks.0.hook_mlp_out/sae_weights.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:493ba14e572993588c7c15e154c22bb6723d8c5b93b9839a3911dd8fa49c9309
|
| 3 |
+
size 134287680
|