VoyagerXHF commited on
Commit
8ac456b
·
verified ·
1 Parent(s): 7d92fd7

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +8 -6
README.md CHANGED
@@ -17,7 +17,7 @@ base_model: Qwen/Qwen3.5-9B
17
 
18
  ![Overview](https://qianwen-res.oss-cn-beijing.aliyuncs.com/qwen-scope/Figures/overview.png)
19
 
20
- We are excited to introduce Qwen-Scope, an interpretability module trained on the Qwen3 and Qwen3.5 series models. Specifically, we integrated and trained Sparse Autoencoders (SAEs) within Qwen’s hidden layers. By implementing sparsity constraints, we can automatically extract data features that are highly decoupled, low-redundancy, and significantly more interpretable. Qwen-Scope can be used not only to analyze the internal mechanisms of Qwen’s behavior but also holds immense potential for model optimization. Application scenarios include steerable inference control, evaluation sample distribution analysis and comparison, data classification and synthesis, and model training and optimization.
21
 
22
  ## Model Details
23
 
@@ -135,11 +135,13 @@ It is strictly prohibited to use interpretability tools for non-scientific resea
135
 
136
  If you use these SAEs in your research, please cite:
137
 
 
138
  ```bibtex
139
- @misc{qwenscope,
140
- title = {{Qwen-Scope}: Turning Sparse Features into Development Tools for Large Language Models},
141
- author = {{Qwen Team}},
142
- month = {April},
143
- year = {2026}
 
144
  }
145
  ```
 
17
 
18
  ![Overview](https://qianwen-res.oss-cn-beijing.aliyuncs.com/qwen-scope/Figures/overview.png)
19
 
20
+ We are excited to introduce Qwen-Scope, an interpretability module trained on the Qwen3 and Qwen3.5 series models. Specifically, we integrated and trained Sparse Autoencoders (SAEs) within Qwen’s hidden layers. By implementing sparsity constraints, we can automatically extract data features that are highly decoupled, low-redundancy, and significantly more interpretable. Qwen-Scope can be used not only to analyze the internal mechanisms of Qwen’s behavior but also holds immense potential for model optimization. Application scenarios include steerable inference control, evaluation sample distribution analysis and comparison, data classification and synthesis, and model training and optimization. See our [technical report](https://qianwen-res.oss-accelerate.aliyuncs.com/qwen-scope/Qwen_Scope.pdf) for more details.
21
 
22
  ## Model Details
23
 
 
135
 
136
  If you use these SAEs in your research, please cite:
137
 
138
+
139
  ```bibtex
140
+ @misc{qwen_scope,
141
+ title = {{Qwen-Scope}: Turning Sparse Features into Development Tools for Large Language Models},
142
+ url = {https://qianwen-res.oss-accelerate.aliyuncs.com/qwen-scope/Qwen_Scope.pdf},
143
+ author = {{Qwen Team}},
144
+ month = {April},
145
+ year = {2026}
146
  }
147
  ```