Downloads · 30 days
0
gLM2_650M_sae
tattabio/gLM2_650M_sae
gLM2_650M_sae is a machine learning model from tattabio. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
A BatchTopK sparse autoencoder trained on the residual stream of tattabio/gLM2650M layer 24. Full 4096-token contigs are embedded, but only intergenic (nucleotide) positions are trained on.
Access
Public
Updated Sep 30, 2026
Repo size
168 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors168 MB · 100%
At a glance
- Access
- Public
- Created
- Aug 26, 2026
- Updated
- Sep 30, 2026
- SHA
- f210c86c
Model card
From the Hugging Face model README
SAE on gLM2_650M layer 24 (intergenic tokens)
A BatchTopK sparse autoencoder trained on the residual stream of
tattabio/gLM2_650M layer 24. Full 4096-token
contigs are embedded, but only intergenic (nucleotide) positions are
trained on.
Configuration
| architecture | BatchTopK |
| k | 8 |
| d_in | 1280 |
| d_sae | 16384 (expansion 12x) |
| hook | glm2.encoder.layers.24 |
| base model | tattabio/gLM2_650M |
| context size | 4096 |
| training tokens | 1000.0M |
| trained on | intergenic (nucleotide) tokens |
- Created
- Aug 26, 2026
- Updated
- Sep 30, 2026