Downloads · 30 days
4
12% of all-time downloads
sae_qwen3_8b_L24_d65k_k80
asher577/sae_qwen3_8b_L24_d65k_k80
sae_qwen3_8b_L24_d65k_k80 is a machine learning model from asher577. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
BatchTopK rank-1 SAE, Qwen3-8B layer 24. d=65536, k=80, lr 1e-4, ~100M FineFineWeb tokens (Karvonen-matched recipe). Autointerp: detection 0.724 / fuzzing 0.786. FVE ~0.804 — our best rank-1 SAE reconstruction.
All-time downloads
34
Public
Repo size
2.1 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt2.1 GB · 100%
At a glance
- Access
- Public
- Created
- Jul 19, 2026
- Updated
- Jul 19, 2026
- SHA
- 986d3602
Model card
From the Hugging Face model README
sae_qwen3_8b_L24_d65k_k80
BatchTopK rank-1 SAE, Qwen3-8B layer 24. d=65536, k=80, lr 1e-4, ~100M FineFineWeb tokens (Karvonen-matched recipe). Autointerp: detection 0.724 / fuzzing 0.786. FVE ~0.804 — our best rank-1 SAE reconstruction.
Standard dictionary_learning BatchTopK SAE (EleutherAI). Load with
dictionary_learning.trainers.batch_top_k.BatchTopKSAE.from_pretrained(ae.pt).
Config in config.json. Trained 2026-07 on the NLA project's L24 activation stream.
- Created
- Jul 19, 2026
- Updated
- Jul 19, 2026