# Qwen publishes residual-stream SAE for Qwen3-8B-Base (64K width, L0=100)

> Qwen published a TopK sparse autoencoder trained on the residual stream of Qwen3-8B-Base, with 64K width and L0 sparsity of 100, for mechanistic interpretability research. Published on Hugging Face under the Qwen organisation.

| | |
|---|---|
| **Tool** | Qwen |
| **Version** | — |
| **Kind** | release |
| **Published** | 2026-04-27 |
| **Observed** | 2026-08-11 |
| **Significance** | 1/5 |
| **Breaking** | no |
| **Categories** | feature, model-support, capability |

> **Note:** this classification is below our confidence threshold and is pending human review.

## What changed


- Repository: `Qwen/SAE-Res-Qwen3-8B-Base-W64K-L0_100`
- Tags: topk_sae, sparse-autoencoder, sae, mechanistic-interpretability, qwen-scope
- Associated paper: arxiv:2605.11887
- License: other (not specified)


## Sources

- [hf_model](https://huggingface.co/Qwen/SAE-Res-Qwen3-8B-Base-W64K-L0_100) — retrieved 2026-08-11


## Community

_No curated reactions recorded. Facts and community takes are kept in separate layers
and never blended._

---
Canonical: https://changelogs.info/qwen/qwen-publishes-residual-stream-sae-for-qwen3-8b-base-64k-width-l0-100
Entity: https://changelogs.info/qwen
Event ID: `evt_2026-04-27_qwen_qwen-sae-res-qwen3-8b-base-w64k-l0-100`
Licence: event synthesis © changelogs.info, CC BY 4.0. Linked sources belong to their vendors.