Hugging Face's logo

cesun
/
ThinkEdit-deepseek-llama3-8b

Text Generation
Transformers
Safetensors
llama
conversational
text-generation-inference
Model card Files
xet
Community
2

Add link to project page

#2
by nielsr HF Staff - opened
base: refs/heads/main
←
from: refs/pr/2
Discussion Files changed
+35
-33
Files changed (1) hide show
  1. README.md +35 -33
README.md CHANGED
@@ -1,10 +1,10 @@
1
  ---
2
- library_name: transformers
3
- tags: []
4
- pipeline_tag: text-generation
5
- license: mit
6
  base_model:
7
  - deepseek-ai/DeepSeek-R1-Distill-Llama-8B
 
 
 
 
8
  ---
9
 
10
  **Repository for:**
@@ -13,11 +13,13 @@ base_model:
13
 
14
  (We also release ThinkEdit versions for ThinkEdit-deepseek-qwen-1.5b, ThinkEdit-deepseek-qwen-14b, and ThinkEdit-deepseek-qwen-32b.)
15
 
16
- **Authors**: Chung-En Sun, Ge Yan, Tsui-Wei Weng
17
  **Paper**: [ThinkEdit: Interpretable Weight Editing to Mitigate Overly Short Thinking in Reasoning Models](https://arxiv.org/abs/2503.22048)
18
 
19
  Github: https://github.com/Trustworthy-ML-Lab/ThinkEdit
20
 
 
 
21
  ---
22
 
23
  ## Introduction
@@ -26,10 +28,10 @@ Reasoning-augmented models sometimes fail by generating **overly short**, abstra
26
 
27
  **ThinkEdit** is a lightweight weight-editing method that:
28
 
29
- - Identifies ~4% of "short reasoning" attention heads
30
- - Edits only ~0.2% of total parameters
31
- - Removes the "short reasoning" direction from their output
32
- - Boosts performance, especially on cases with short reasoning traces
33
 
34
  ---
35
 
@@ -37,31 +39,31 @@ Reasoning-augmented models sometimes fail by generating **overly short**, abstra
37
 
38
  ### 1. Overall Accuracy
39
 
40
- | Model | GSM8K | MMLU Elementary Math | MATH-Level1 | MATH-Level5 | MATH-500 |
41
  |---------------------------------|---------------------|----------------------|---------------------|---------------------|---------------------|
42
- | deepseek-qwen-32b | 92.97 ± 0.39 | 95.93 ± 0.83 | **96.41 ± 0.45** | 91.27 ± 0.53 | **91.62 ± 0.58** |
43
- | **ThinkEdit-deepseek-qwen-32b** | **95.25 ± 0.25** | **98.02 ± 0.31** | 96.02 ± 0.42 | **91.31 ± 0.50** | 91.60 ± 0.65 |
44
- | deepseek-qwen-14b | 90.80 ± 0.36 | 95.08 ± 0.65 | 96.32 ± 0.35 | 90.25 ± 0.72 | 91.48 ± 0.55 |
45
- | **ThinkEdit-deepseek-qwen-14b** | **93.78 ± 0.50** | **96.56 ± 0.84** | **96.38 ± 0.52** | **91.03 ± 0.44** | **91.92 ± 0.63** |
46
- | deepseek-llama3-8b | 82.26 ± 0.91 | 96.01 ± 0.62 | 93.46 ± 0.84 | 85.49 ± 0.83 | 87.26 ± 1.16 |
47
- | **ThinkEdit-deepseek-llama3-8b**| **89.44 ± 0.55** | **96.19 ± 0.73** | **94.44 ± 0.31** | **86.49 ± 0.54** | **88.06 ± 1.09** |
48
- | deepseek-qwen-1.5b | 79.15 ± 1.08 | 68.52 ± 1.56 | 93.00 ± 0.33 | **75.48 ± 0.90** | 82.22 ± 1.29 |
49
- | **ThinkEdit-deepseek-qwen-1.5b**| **84.56 ± 0.79** | **90.66 ± 0.97** | **93.66 ± 0.62** | 75.05 ± 0.82 | **82.24 ± 0.89** |
50
 
51
  ---
52
 
53
  ### 2. Accuracy on Short Reasoning Cases (Top 5% / 10% / 20%)
54
 
55
- | Model | GSM8K | MMLU Elementary Math | MATH-Level1 | MATH-Level5 | MATH-500 |
56
  |---------------------------------|---------------------------------|----------------------------------|----------------------------------|----------------------------------|----------------------------------|
57
- | deepseek-qwen-32b | 98.31 / 97.18 / 96.20 | 97.78 / 97.03 / 95.87 | 100.00 / 100.00 / **98.97** | 93.03 / 96.36 / 97.35 | 86.40 / 92.00 / 94.00 |
58
- | **ThinkEdit-deepseek-qwen-32b** | **98.92** / **97.71** / **97.83** | **97.78** / **97.57** / **97.20** | **100.00** / **100.00** / 98.74 | **98.03** / **98.64** / **97.99** | **92.00** / **94.40** / **95.80** |
59
- | deepseek-qwen-14b | **96.31** / 95.65 / 92.93 | 93.89 / **96.22** / 95.60 | 99.52 / **99.30** / 97.70 | 89.39 / 94.32 / 96.25 | 86.40 / 91.40 / 93.50 |
60
- | **ThinkEdit-deepseek-qwen-14b** | **96.31** / **96.18** / **96.77** | **97.78** / 95.14 / **96.53** | **99.53** / 98.62 / **98.67** | **96.67** / **97.88** / **98.11** | **91.20** / **93.20** / **95.00** |
61
- | deepseek-llama3-8b | 88.92 / 87.18 / 85.82 | 97.22 / 96.49 / 96.80 | 97.14 / 94.88 / 94.83 | 78.64 / 88.79 / 93.41 | 82.00 / 81.40 / 88.30 |
62
  | **ThinkEdit-deepseek-llama3-8b**| **97.08** / **95.27** / **93.95** | **97.78** / **98.65** / **97.87** | **100.00** / **99.30** / **98.62** | **95.61** / **96.89** / **97.12** | **92.80** / **93.60** / **94.40** |
63
- | deepseek-qwen-1.5b | 88.46 / 87.48 / 85.02 | 62.78 / 62.16 / 60.53 | **97.62** / 95.12 / 93.91 | 91.52 / 95.00 / 95.72 | 82.40 / 89.80 / 93.40 |
64
- | **ThinkEdit-deepseek-qwen-1.5b**| **92.62** / **92.90** / **92.32** | **87.78** / **88.11** / **88.67** | 95.71 / **95.58** / **96.44** | **95.15** / **96.59** / **97.27** | **90.80** / **92.00** / **94.20** |
65
 
66
  ---
67
 
@@ -73,11 +75,11 @@ The usage of ThinkEdit models is exactly the same as the original deepseek-disti
73
 
74
  ```bibtex
75
  @misc{sun2025thinkedit,
76
- title={ThinkEdit: Interpretable Weight Editing to Mitigate Overly Short Thinking in Reasoning Models},
77
- author={Chung-En Sun and Ge Yan and Tsui-Wei Weng},
78
- year={2025},
79
- eprint={2503.22048},
80
- archivePrefix={arXiv},
81
- primaryClass={cs.CL},
82
- url={https://arxiv.org/abs/2503.22048},
83
  }
 
 
1
  ---
 
 
 
 
2
  base_model:
3
  - deepseek-ai/DeepSeek-R1-Distill-Llama-8B
4
+ library_name: transformers
5
+ license: mit
6
+ pipeline_tag: text-generation
7
+ tags: []
8
  ---
9
 
10
  **Repository for:**
 
13
 
14
  (We also release ThinkEdit versions for ThinkEdit-deepseek-qwen-1.5b, ThinkEdit-deepseek-qwen-14b, and ThinkEdit-deepseek-qwen-32b.)
15
 
16
+ **Authors**: Chung-En Sun, Ge Yan, Tsui-Wei Weng
17
  **Paper**: [ThinkEdit: Interpretable Weight Editing to Mitigate Overly Short Thinking in Reasoning Models](https://arxiv.org/abs/2503.22048)
18
 
19
  Github: https://github.com/Trustworthy-ML-Lab/ThinkEdit
20
 
21
+ Project page: https://lilywenglab.github.io/ThinkEdit/
22
+
23
  ---
24
 
25
  ## Introduction
 
28
 
29
  **ThinkEdit** is a lightweight weight-editing method that:
30
 
31
+ - Identifies ~4% of "short reasoning" attention heads
32
+ - Edits only ~0.2% of total parameters
33
+ - Removes the "short reasoning" direction from their output
34
+ - Boosts performance, especially on cases with short reasoning traces
35
 
36
  ---
37
 
 
39
 
40
  ### 1. Overall Accuracy
41
 
42
+ | Model | GSM8K | MMLU Elementary Math | MATH-Level1 | MATH-Level5 | MATH-500 |
43
  |---------------------------------|---------------------|----------------------|---------------------|---------------------|---------------------|
44
+ | deepseek-qwen-32b | 92.97 ± 0.39 | 95.93 ± 0.83 | **96.41 ± 0.45** | 91.27 ± 0.53 | **91.62 ± 0.58** |
45
+ | **ThinkEdit-deepseek-qwen-32b** | **95.25 ± 0.25** | **98.02 ± 0.31** | 96.02 ± 0.42 | **91.31 ± 0.50** | 91.60 ± 0.65 |
46
+ | deepseek-qwen-14b | 90.80 ± 0.36 | 95.08 ± 0.65 | 96.32 ± 0.35 | 90.25 ± 0.72 | 91.48 ± 0.55 |
47
+ | **ThinkEdit-deepseek-qwen-14b** | **93.78 ± 0.50** | **96.56 ± 0.84** | **96.38 ± 0.52** | **91.03 ± 0.44** | **91.92 ± 0.63** |
48
+ | deepseek-llama3-8b | 82.26 ± 0.91 | 96.01 ± 0.62 | 93.46 ± 0.84 | 85.49 ± 0.83 | 87.26 ± 1.16 |
49
+ | **ThinkEdit-deepseek-llama3-8b**| **89.44 ± 0.55** | **96.19 ± 0.73** | **94.44 ± 0.31** | **86.49 ± 0.54** | **88.06 ± 1.09** |
50
+ | deepseek-qwen-1.5b | 79.15 ± 1.08 | 68.52 ± 1.56 | 93.00 ± 0.33 | **75.48 ± 0.90** | 82.22 ± 1.29 |
51
+ | **ThinkEdit-deepseek-qwen-1.5b**| **84.56 ± 0.79** | **90.66 ± 0.97** | **93.66 ± 0.62** | 75.05 ± 0.82 | **82.24 ± 0.89** |
52
 
53
  ---
54
 
55
  ### 2. Accuracy on Short Reasoning Cases (Top 5% / 10% / 20%)
56
 
57
+ | Model | GSM8K | MMLU Elementary Math | MATH-Level1 | MATH-Level5 | MATH-500 |
58
  |---------------------------------|---------------------------------|----------------------------------|----------------------------------|----------------------------------|----------------------------------|
59
+ | deepseek-qwen-32b | 98.31 / 97.18 / 96.20 | 97.78 / 97.03 / 95.87 | 100.00 / 100.00 / **98.97** | 93.03 / 96.36 / 97.35 | 86.40 / 92.00 / 94.00 |
60
+ | **ThinkEdit-deepseek-qwen-32b** | **98.92** / **97.71** / **97.83** | **97.78** / **97.57** / **97.20** | **100.00** / **100.00** / 98.74 | **98.03** / **98.64** / **97.99** | **92.00** / **94.40** / **95.80** |
61
+ | deepseek-qwen-14b | **96.31** / 95.65 / 92.93 | 93.89 / **96.22** / 95.60 | 99.52 / **99.30** / 97.70 | 89.39 / 94.32 / 96.25 | 86.40 / 91.40 / 93.50 |
62
+ | **ThinkEdit-deepseek-qwen-14b** | **96.31** / **96.18** / **96.77** | **97.78** / 95.14 / **96.53** | **99.53** / 98.62 / **98.67** | **96.67** / **97.88** / **98.11** | **91.20** / **93.20** / **95.00** |
63
+ | deepseek-llama3-8b | 88.92 / 87.18 / 85.82 | 97.22 / 96.49 / 96.80 | 97.14 / 94.88 / 94.83 | 78.64 / 88.79 / 93.41 | 82.00 / 81.40 / 88.30 |
64
  | **ThinkEdit-deepseek-llama3-8b**| **97.08** / **95.27** / **93.95** | **97.78** / **98.65** / **97.87** | **100.00** / **99.30** / **98.62** | **95.61** / **96.89** / **97.12** | **92.80** / **93.60** / **94.40** |
65
+ | deepseek-qwen-1.5b | 88.46 / 87.48 / 85.02 | 62.78 / 62.16 / 60.53 | **97.62** / 95.12 / 93.91 | 91.52 / 95.00 / 95.72 | 82.40 / 89.80 / 93.40 |
66
+ | **ThinkEdit-deepseek-qwen-1.5b**| **92.62** / **92.90** / **92.32** | **87.78** / **88.11** / **88.67** | 95.71 / **95.58** / **96.44** | **95.15** / **96.59** / **97.27** | **90.80** / **92.00** / **94.20** |
67
 
68
  ---
69
 
 
75
 
76
  ```bibtex
77
  @misc{sun2025thinkedit,
78
+ title={ThinkEdit: Interpretable Weight Editing to Mitigate Overly Short Thinking in Reasoning Models},
79
+ author={Chung-En Sun and Ge Yan and Tsui-Wei Weng},
80
+ year={2025},
81
+ eprint={2503.22048},
82
+ archivePrefix={arXiv},
83
+ url={https://arxiv.org/abs/2503.22048},
 
84
  }
85
+ ```