Files changed (1) hide show
  1. README.md +117 -0
README.md CHANGED
@@ -4,6 +4,109 @@ tags:
4
  - moe
5
  - DPO
6
  - RL-TUNED
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7
  ---
8
 
9
  * [DPO Trainer](https://huggingface.co/docs/trl/main/en/dpo_trainer) with dataset kyujinpy/orca_math_dpo to improve [yunconglong/MoE_13B_DPO]
@@ -11,3 +114,17 @@ tags:
11
  DPO Trainer
12
  TRL supports the DPO Trainer for training language models from preference data, as described in the paper Direct Preference Optimization: Your Language Model is Secretly a Reward Model by Rafailov et al., 2023.
13
  ```
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4
  - moe
5
  - DPO
6
  - RL-TUNED
7
+ model-index:
8
+ - name: 13B_MATH_DPO
9
+ results:
10
+ - task:
11
+ type: text-generation
12
+ name: Text Generation
13
+ dataset:
14
+ name: AI2 Reasoning Challenge (25-Shot)
15
+ type: ai2_arc
16
+ config: ARC-Challenge
17
+ split: test
18
+ args:
19
+ num_few_shot: 25
20
+ metrics:
21
+ - type: acc_norm
22
+ value: 74.66
23
+ name: normalized accuracy
24
+ source:
25
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=yunconglong/13B_MATH_DPO
26
+ name: Open LLM Leaderboard
27
+ - task:
28
+ type: text-generation
29
+ name: Text Generation
30
+ dataset:
31
+ name: HellaSwag (10-Shot)
32
+ type: hellaswag
33
+ split: validation
34
+ args:
35
+ num_few_shot: 10
36
+ metrics:
37
+ - type: acc_norm
38
+ value: 89.51
39
+ name: normalized accuracy
40
+ source:
41
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=yunconglong/13B_MATH_DPO
42
+ name: Open LLM Leaderboard
43
+ - task:
44
+ type: text-generation
45
+ name: Text Generation
46
+ dataset:
47
+ name: MMLU (5-Shot)
48
+ type: cais/mmlu
49
+ config: all
50
+ split: test
51
+ args:
52
+ num_few_shot: 5
53
+ metrics:
54
+ - type: acc
55
+ value: 64.53
56
+ name: accuracy
57
+ source:
58
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=yunconglong/13B_MATH_DPO
59
+ name: Open LLM Leaderboard
60
+ - task:
61
+ type: text-generation
62
+ name: Text Generation
63
+ dataset:
64
+ name: TruthfulQA (0-shot)
65
+ type: truthful_qa
66
+ config: multiple_choice
67
+ split: validation
68
+ args:
69
+ num_few_shot: 0
70
+ metrics:
71
+ - type: mc2
72
+ value: 78.63
73
+ source:
74
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=yunconglong/13B_MATH_DPO
75
+ name: Open LLM Leaderboard
76
+ - task:
77
+ type: text-generation
78
+ name: Text Generation
79
+ dataset:
80
+ name: Winogrande (5-shot)
81
+ type: winogrande
82
+ config: winogrande_xl
83
+ split: validation
84
+ args:
85
+ num_few_shot: 5
86
+ metrics:
87
+ - type: acc
88
+ value: 88.08
89
+ name: accuracy
90
+ source:
91
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=yunconglong/13B_MATH_DPO
92
+ name: Open LLM Leaderboard
93
+ - task:
94
+ type: text-generation
95
+ name: Text Generation
96
+ dataset:
97
+ name: GSM8k (5-shot)
98
+ type: gsm8k
99
+ config: main
100
+ split: test
101
+ args:
102
+ num_few_shot: 5
103
+ metrics:
104
+ - type: acc
105
+ value: 67.1
106
+ name: accuracy
107
+ source:
108
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=yunconglong/13B_MATH_DPO
109
+ name: Open LLM Leaderboard
110
  ---
111
 
112
  * [DPO Trainer](https://huggingface.co/docs/trl/main/en/dpo_trainer) with dataset kyujinpy/orca_math_dpo to improve [yunconglong/MoE_13B_DPO]
 
114
  DPO Trainer
115
  TRL supports the DPO Trainer for training language models from preference data, as described in the paper Direct Preference Optimization: Your Language Model is Secretly a Reward Model by Rafailov et al., 2023.
116
  ```
117
+
118
+ # [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
119
+ Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_yunconglong__13B_MATH_DPO)
120
+
121
+ | Metric |Value|
122
+ |---------------------------------|----:|
123
+ |Avg. |77.08|
124
+ |AI2 Reasoning Challenge (25-Shot)|74.66|
125
+ |HellaSwag (10-Shot) |89.51|
126
+ |MMLU (5-Shot) |64.53|
127
+ |TruthfulQA (0-shot) |78.63|
128
+ |Winogrande (5-shot) |88.08|
129
+ |GSM8k (5-shot) |67.10|
130
+