Upload folder using huggingface_hub
Browse files
README.md
CHANGED
|
@@ -8,7 +8,7 @@ tags:
|
|
| 8 |
|
| 9 |
# Sparse Autoencoders for Qwen/Qwen2.5-7B-Instruct
|
| 10 |
|
| 11 |
-
This repository contains
|
| 12 |
|
| 13 |
## Model Details
|
| 14 |
|
|
@@ -25,13 +25,6 @@ This repository contains 8 Sparse Autoencoder(s) (SAE) trained using [SAELens](h
|
|
| 25 |
| Hook Point |
|
| 26 |
|------------|
|
| 27 |
| `blocks.11.hook_resid_post` |
|
| 28 |
-
| `blocks.0.hook_resid_post` |
|
| 29 |
-
| `blocks.3.hook_resid_post` |
|
| 30 |
-
| `blocks.7.hook_resid_post` |
|
| 31 |
-
| `blocks.15.hook_resid_post` |
|
| 32 |
-
| `blocks.19.hook_resid_post` |
|
| 33 |
-
| `blocks.23.hook_resid_post` |
|
| 34 |
-
| `blocks.27.hook_resid_post` |
|
| 35 |
|
| 36 |
## Usage
|
| 37 |
|
|
@@ -60,27 +53,6 @@ features = sae.encode(activations)
|
|
| 60 |
- `blocks.11.hook_resid_post/cfg.json` - SAE configuration
|
| 61 |
- `blocks.11.hook_resid_post/sae_weights.safetensors` - Model weights
|
| 62 |
- `blocks.11.hook_resid_post/sparsity.safetensors` - Feature sparsity statistics
|
| 63 |
-
- `blocks.0.hook_resid_post/cfg.json` - SAE configuration
|
| 64 |
-
- `blocks.0.hook_resid_post/sae_weights.safetensors` - Model weights
|
| 65 |
-
- `blocks.0.hook_resid_post/sparsity.safetensors` - Feature sparsity statistics
|
| 66 |
-
- `blocks.3.hook_resid_post/cfg.json` - SAE configuration
|
| 67 |
-
- `blocks.3.hook_resid_post/sae_weights.safetensors` - Model weights
|
| 68 |
-
- `blocks.3.hook_resid_post/sparsity.safetensors` - Feature sparsity statistics
|
| 69 |
-
- `blocks.7.hook_resid_post/cfg.json` - SAE configuration
|
| 70 |
-
- `blocks.7.hook_resid_post/sae_weights.safetensors` - Model weights
|
| 71 |
-
- `blocks.7.hook_resid_post/sparsity.safetensors` - Feature sparsity statistics
|
| 72 |
-
- `blocks.15.hook_resid_post/cfg.json` - SAE configuration
|
| 73 |
-
- `blocks.15.hook_resid_post/sae_weights.safetensors` - Model weights
|
| 74 |
-
- `blocks.15.hook_resid_post/sparsity.safetensors` - Feature sparsity statistics
|
| 75 |
-
- `blocks.19.hook_resid_post/cfg.json` - SAE configuration
|
| 76 |
-
- `blocks.19.hook_resid_post/sae_weights.safetensors` - Model weights
|
| 77 |
-
- `blocks.19.hook_resid_post/sparsity.safetensors` - Feature sparsity statistics
|
| 78 |
-
- `blocks.23.hook_resid_post/cfg.json` - SAE configuration
|
| 79 |
-
- `blocks.23.hook_resid_post/sae_weights.safetensors` - Model weights
|
| 80 |
-
- `blocks.23.hook_resid_post/sparsity.safetensors` - Feature sparsity statistics
|
| 81 |
-
- `blocks.27.hook_resid_post/cfg.json` - SAE configuration
|
| 82 |
-
- `blocks.27.hook_resid_post/sae_weights.safetensors` - Model weights
|
| 83 |
-
- `blocks.27.hook_resid_post/sparsity.safetensors` - Feature sparsity statistics
|
| 84 |
|
| 85 |
## Training
|
| 86 |
|
|
|
|
| 8 |
|
| 9 |
# Sparse Autoencoders for Qwen/Qwen2.5-7B-Instruct
|
| 10 |
|
| 11 |
+
This repository contains 1 Sparse Autoencoder(s) (SAE) trained using [SAELens](https://github.com/jbloomAus/SAELens).
|
| 12 |
|
| 13 |
## Model Details
|
| 14 |
|
|
|
|
| 25 |
| Hook Point |
|
| 26 |
|------------|
|
| 27 |
| `blocks.11.hook_resid_post` |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
|
| 29 |
## Usage
|
| 30 |
|
|
|
|
| 53 |
- `blocks.11.hook_resid_post/cfg.json` - SAE configuration
|
| 54 |
- `blocks.11.hook_resid_post/sae_weights.safetensors` - Model weights
|
| 55 |
- `blocks.11.hook_resid_post/sparsity.safetensors` - Feature sparsity statistics
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 56 |
|
| 57 |
## Training
|
| 58 |
|
blocks.11.hook_resid_post/cfg.json
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
{"d_in": 3584, "d_sae": 16384, "dtype": "float32", "device": "cuda", "apply_b_dec_to_input": true, "normalize_activations": "none", "reshape_activations": "none", "metadata": {"sae_lens_version": "6.26.2", "sae_lens_training_version": "6.26.2", "dataset_path": "TQRG/DeltaSecommits_qwen-2.5-7b-instruct_tokenized_v2_vulnerable", "hook_name": "blocks.11.hook_resid_post", "model_name": "Qwen/Qwen2.5-7B-Instruct", "model_class_name": "HookedTransformer", "hook_head_index": null, "context_size": 128, "seqpos_slice": [null, null], "model_from_pretrained_kwargs": {}, "prepend_bos": true, "exclude_special_tokens": false, "sequence_separator_token": "bos", "disable_concat_sequences":
|
|
|
|
| 1 |
+
{"d_in": 3584, "d_sae": 16384, "dtype": "float32", "device": "cuda", "apply_b_dec_to_input": true, "normalize_activations": "none", "reshape_activations": "none", "metadata": {"sae_lens_version": "6.26.2", "sae_lens_training_version": "6.26.2", "dataset_path": "TQRG/DeltaSecommits_qwen-2.5-7b-instruct_tokenized_v2_vulnerable", "hook_name": "blocks.11.hook_resid_post", "model_name": "Qwen/Qwen2.5-7B-Instruct", "model_class_name": "HookedTransformer", "hook_head_index": null, "context_size": 128, "seqpos_slice": [null, null], "model_from_pretrained_kwargs": {}, "prepend_bos": true, "exclude_special_tokens": false, "sequence_separator_token": "bos", "disable_concat_sequences": true}, "decoder_init_norm": 0.1, "l1_coefficient": 1.0, "lp_norm": 1.0, "l1_warm_up_steps": 0, "architecture": "standard"}
|
blocks.11.hook_resid_post/sae_weights.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 469842240
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8313929119bfed6a3bca3417b6e7ee581babc4a3adbf3ccfcfc9cf09ca15a708
|
| 3 |
size 469842240
|