Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,4 @@
|
|
|
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
datasets:
|
|
@@ -29,17 +30,30 @@ Directory structure:
|
|
| 29 |
|
| 30 |
```
|
| 31 |
.
|
| 32 |
-
βββ 2b_l10
|
| 33 |
βΒ Β βββ outputs_add_free
|
| 34 |
βΒ Β βββ all -- full-sequence intervention
|
| 35 |
-
βΒ Β βΒ Β βββ lang -- Lang. objective
|
| 36 |
-
βΒ Β β β βββ 0 -- concept 0
|
| 37 |
-
βΒ Β β β βββ 1 -- concept 1
|
| 38 |
βΒ Β β β ...
|
| 39 |
βΒ Β βΒ Β βββ simpo -- SimPO objective
|
| 40 |
-
βΒ Β βββ f2+l2 -- prompt-only intervention (2 prefix tokens, 2 suffix tokens)
|
| 41 |
βΒ Β βββ lang
|
| 42 |
βΒ Β βββ simpo
|
|
|
|
| 43 |
```
|
| 44 |
|
|
|
|
|
|
|
|
|
|
| 45 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
---
|
| 3 |
license: mit
|
| 4 |
datasets:
|
|
|
|
| 30 |
|
| 31 |
```
|
| 32 |
.
|
| 33 |
+
βββ 2b_l10 -- setup
|
| 34 |
βΒ Β βββ outputs_add_free
|
| 35 |
βΒ Β βββ all -- full-sequence intervention
|
| 36 |
+
βΒ Β βΒ Β βββ lang ---- Lang. objective
|
| 37 |
+
βΒ Β β β βββ 0 ------ concept 0
|
| 38 |
+
βΒ Β β β βββ 1 ------ concept 1
|
| 39 |
βΒ Β β β ...
|
| 40 |
βΒ Β βΒ Β βββ simpo -- SimPO objective
|
| 41 |
+
βΒ Β βββ f2+l2 ---- prompt-only intervention (2 prefix tokens, 2 suffix tokens)
|
| 42 |
βΒ Β βββ lang
|
| 43 |
βΒ Β βββ simpo
|
| 44 |
+
...
|
| 45 |
```
|
| 46 |
|
| 47 |
+
## Citation
|
| 48 |
+
|
| 49 |
+
If you find our work useful, please cite:
|
| 50 |
|
| 51 |
+
```bibtex
|
| 52 |
+
@inproceedings{bao2026towards,
|
| 53 |
+
title = {Towards Steering without Sacrifice: Principled Training of Steering Vectors for Prompt-only Interventions},
|
| 54 |
+
author = {Bao, Yuntai and Li, Qinfeng and Yu, Xinyan and Zhang, Xuhong and Su, Ge and Zhang, Wenqi and Yan, Liu and Weng, Haiqin and Yin, Jianwei},
|
| 55 |
+
booktitle = {Forty-third International Conference on Machine Learning},
|
| 56 |
+
year = {2026},
|
| 57 |
+
url = {https://openreview.net/forum?id=AaT3liS5PE},
|
| 58 |
+
}
|
| 59 |
+
```
|