Update README.md
Browse files
README.md
CHANGED
@@ -1,11 +1,12 @@
|
|
1 |
---
|
2 |
license: mit
|
3 |
---
|
4 |
-
This directory contains sparse autoencoders trained on activations at various points within gpt2-small using [Neel Nanda's open source code](https://github.com/neelnanda-io/1L-Sparse-Autoencoder)
|
5 |
-
|
6 |
Each autoencoder was trained on 1B tokens from OpenWebText.
|
|
|
|
|
7 |
|
8 |
-
The autoencoders are named "gpt2-small_{feature_dict_size}_{point}_{layer}.pt", where:
|
9 |
- "feature_dict_size" is the number of hidden neurons in the autoencoder
|
10 |
- "point" is either "mlp_out" or "resid_pre"
|
11 |
- "layer" is an integer from 0,...,11.
|
|
|
1 |
---
|
2 |
license: mit
|
3 |
---
|
4 |
+
This directory contains sparse autoencoders trained on activations at various points within gpt2-small using [Neel Nanda's open source code](https://github.com/neelnanda-io/1L-Sparse-Autoencoder).
|
|
|
5 |
Each autoencoder was trained on 1B tokens from OpenWebText.
|
6 |
+
A demo colab notebook is [here](https://colab.research.google.com/drive/1KeRGixXf_5GrG_7vQalG6UJQyhQd6byi?usp=sharing).
|
7 |
+
|
8 |
|
9 |
+
The autoencoders are named "gpt2-small_{feature_dict_size}_{point} _{layer}.pt", where:
|
10 |
- "feature_dict_size" is the number of hidden neurons in the autoencoder
|
11 |
- "point" is either "mlp_out" or "resid_pre"
|
12 |
- "layer" is an integer from 0,...,11.
|