Instructions to use seonglae/gemma-2-2b-sae with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- SAELens
How to use seonglae/gemma-2-2b-sae 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 gemma-2-2b_blocks.20.hook_resid_pre_18432_topk_64_0.0003_49_faithful-gemma2-2b_1024
Browse files
gemma-2-2b_blocks.20.hook_resid_pre_18432_topk_64_0.0003_49_faithful-gemma2-2b_1024/cfg.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"architecture": "topk", "d_in": 2304, "d_sae": 18432, "dtype": "torch.bfloat16", "device": "cuda", "model_name": "google/gemma-2-2b-topk-sae-18432-faithful-gemma2-2b", "hook_name": "blocks.20.hook_resid_pre", "hook_layer": 20, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 64}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": "5.5.2", "prepend_bos": true, "dataset_path": "seonglae/faithful-gemma2-2b", "dataset_trust_remote_code": true, "context_size": 1024, "normalize_activations": "none", "neuronpedia_id": null, "model_from_pretrained_kwargs": {}, "seqpos_slice": [null]}
|
gemma-2-2b_blocks.20.hook_resid_pre_18432_topk_64_0.0003_49_faithful-gemma2-2b_1024/sae_weights.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c8ede56d091a351afaf14cbf5688bd8d64adc9632cdbed7ba97a55e314cf1aa2
|
| 3 |
+
size 169911104
|