tim-lawson commited on
Commit
a279b0c
·
verified ·
1 Parent(s): a0163a0

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +37 -14
README.md CHANGED
@@ -3,22 +3,45 @@ language: en
3
  library_name: mlsae
4
  license: mit
5
  tags:
6
- - model_hub_mixin
7
- - pytorch_model_hub_mixin
8
- datasets:
9
- - monology/pile-uncopyrighted
10
  ---
11
 
12
- # mlsae-pythia-70m-deduped-x4-k32
13
 
14
- A Multi-Layer Sparse Autoencoder (MLSAE) trained on the residual stream
15
- activation vectors from every layer of
16
- [EleutherAI/pythia-70m-deduped](https://huggingface.co/EleutherAI/pythia-70m-deduped)
17
- with an expansion factor of 4 and k = 32, over 1 billion tokens from
18
- [monology/pile-uncopyrighted](https://huggingface.co/datasets/monology/pile-uncopyrighted).
19
 
20
- For more details, see:
21
 
22
- - Paper: <https://arxiv.org/abs/2409.04185>
23
- - GitHub repository: <https://github.com/tim-lawson/mlsae>
24
- - Weights & Biases project: <https://wandb.ai/timlawson-/mlsae>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  library_name: mlsae
4
  license: mit
5
  tags:
6
+ - arxiv:2409.04185
7
+ - model_hub_mixin
8
+ - pytorch_model_hub_mixin
 
9
  ---
10
 
11
+ # Model Card for tim-lawson/mlsae-pythia-70m-deduped-x4-k32
12
 
13
+ A Multi-Layer Sparse Autoencoder (MLSAE) trained on the residual stream activation
14
+ vectors from [EleutherAI/pythia-70m-deduped](https://huggingface.co/EleutherAI/pythia-70m-deduped) with an
15
+ expansion factor of R = 4 and sparsity k = 32, over 1 billion
16
+ tokens from [monology/pile-uncopyrighted](https://huggingface.co/datasets/monology/pile-uncopyrighted).
 
17
 
 
18
 
19
+ This model is a PyTorch TopKSAE module, which does not include the underlying
20
+ transformer.
21
+
22
+
23
+ ### Model Sources
24
+
25
+ - **Repository:** <https://github.com/tim-lawson/mlsae>
26
+ - **Paper:** <https://arxiv.org/abs/2409.04185>
27
+ - **Weights & Biases:** <https://wandb.ai/timlawson-/mlsae>
28
+
29
+ ## Citation
30
+
31
+ **BibTeX:**
32
+
33
+ ```bibtex
34
+ @misc{lawson_residual_2024,
35
+ title = {Residual {{Stream Analysis}} with {{Multi-Layer SAEs}}},
36
+ author = {Lawson, Tim and Farnik, Lucy and Houghton, Conor and Aitchison, Laurence},
37
+ year = {2024},
38
+ month = oct,
39
+ number = {arXiv:2409.04185},
40
+ eprint = {2409.04185},
41
+ primaryclass = {cs},
42
+ publisher = {arXiv},
43
+ doi = {10.48550/arXiv.2409.04185},
44
+ urldate = {2024-10-08},
45
+ archiveprefix = {arXiv}
46
+ }
47
+ ```