Commit
e9a118f
1 Parent(s): 25442a5

Adding Evaluation Results (#3)

Browse files

- Adding Evaluation Results (092dbe7d32fb4d51d9dfc514878f8138162d4ade)


Co-authored-by: Open LLM Leaderboard PR Bot <leaderboard-pr-bot@users.noreply.huggingface.co>

Files changed (1) hide show
  1. README.md +121 -5
README.md CHANGED
@@ -1,13 +1,116 @@
1
  ---
2
  license: other
3
- license_name: yi-license
4
- license_link: LICENSE
5
- datasets:
6
- - adamo1139/AEZAKMI_v2
7
  tags:
8
  - llm
9
  - fine-tune
10
  - yi
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
11
  ---
12
 
13
  ## Model description
@@ -115,4 +218,17 @@ There is also some issue with handling long system messages for RP, I was planni
115
  I will probably be working on de-contaminating base Yi-34B model now. \
116
  My second run of AEZAKMI v2 fine-tune was just 0.15 epochs and I really like how natural this model is and how rich is it's vocabulary. I will try to train less to hit the sweetspot. \
117
  I will be uploading LoRA adapter for that second run that was just 0.15 epochs. \
118
- I believe that I might have gotten what I want if I would have stopped training sooner. I don't have checkpoints older than 1500 steps back so I would need to re-run training to get it back.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: other
 
 
 
 
3
  tags:
4
  - llm
5
  - fine-tune
6
  - yi
7
+ datasets:
8
+ - adamo1139/AEZAKMI_v2
9
+ license_name: yi-license
10
+ license_link: LICENSE
11
+ model-index:
12
+ - name: Yi-34B-200K-AEZAKMI-v2
13
+ results:
14
+ - task:
15
+ type: text-generation
16
+ name: Text Generation
17
+ dataset:
18
+ name: AI2 Reasoning Challenge (25-Shot)
19
+ type: ai2_arc
20
+ config: ARC-Challenge
21
+ split: test
22
+ args:
23
+ num_few_shot: 25
24
+ metrics:
25
+ - type: acc_norm
26
+ value: 67.92
27
+ name: normalized accuracy
28
+ source:
29
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=adamo1139/Yi-34B-200K-AEZAKMI-v2
30
+ name: Open LLM Leaderboard
31
+ - task:
32
+ type: text-generation
33
+ name: Text Generation
34
+ dataset:
35
+ name: HellaSwag (10-Shot)
36
+ type: hellaswag
37
+ split: validation
38
+ args:
39
+ num_few_shot: 10
40
+ metrics:
41
+ - type: acc_norm
42
+ value: 85.61
43
+ name: normalized accuracy
44
+ source:
45
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=adamo1139/Yi-34B-200K-AEZAKMI-v2
46
+ name: Open LLM Leaderboard
47
+ - task:
48
+ type: text-generation
49
+ name: Text Generation
50
+ dataset:
51
+ name: MMLU (5-Shot)
52
+ type: cais/mmlu
53
+ config: all
54
+ split: test
55
+ args:
56
+ num_few_shot: 5
57
+ metrics:
58
+ - type: acc
59
+ value: 75.22
60
+ name: accuracy
61
+ source:
62
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=adamo1139/Yi-34B-200K-AEZAKMI-v2
63
+ name: Open LLM Leaderboard
64
+ - task:
65
+ type: text-generation
66
+ name: Text Generation
67
+ dataset:
68
+ name: TruthfulQA (0-shot)
69
+ type: truthful_qa
70
+ config: multiple_choice
71
+ split: validation
72
+ args:
73
+ num_few_shot: 0
74
+ metrics:
75
+ - type: mc2
76
+ value: 56.74
77
+ source:
78
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=adamo1139/Yi-34B-200K-AEZAKMI-v2
79
+ name: Open LLM Leaderboard
80
+ - task:
81
+ type: text-generation
82
+ name: Text Generation
83
+ dataset:
84
+ name: Winogrande (5-shot)
85
+ type: winogrande
86
+ config: winogrande_xl
87
+ split: validation
88
+ args:
89
+ num_few_shot: 5
90
+ metrics:
91
+ - type: acc
92
+ value: 81.61
93
+ name: accuracy
94
+ source:
95
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=adamo1139/Yi-34B-200K-AEZAKMI-v2
96
+ name: Open LLM Leaderboard
97
+ - task:
98
+ type: text-generation
99
+ name: Text Generation
100
+ dataset:
101
+ name: GSM8k (5-shot)
102
+ type: gsm8k
103
+ config: main
104
+ split: test
105
+ args:
106
+ num_few_shot: 5
107
+ metrics:
108
+ - type: acc
109
+ value: 58.91
110
+ name: accuracy
111
+ source:
112
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=adamo1139/Yi-34B-200K-AEZAKMI-v2
113
+ name: Open LLM Leaderboard
114
  ---
115
 
116
  ## Model description
 
218
  I will probably be working on de-contaminating base Yi-34B model now. \
219
  My second run of AEZAKMI v2 fine-tune was just 0.15 epochs and I really like how natural this model is and how rich is it's vocabulary. I will try to train less to hit the sweetspot. \
220
  I will be uploading LoRA adapter for that second run that was just 0.15 epochs. \
221
+ I believe that I might have gotten what I want if I would have stopped training sooner. I don't have checkpoints older than 1500 steps back so I would need to re-run training to get it back.
222
+ # [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
223
+ Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_adamo1139__Yi-34B-200K-AEZAKMI-v2)
224
+
225
+ | Metric |Value|
226
+ |---------------------------------|----:|
227
+ |Avg. |71.00|
228
+ |AI2 Reasoning Challenge (25-Shot)|67.92|
229
+ |HellaSwag (10-Shot) |85.61|
230
+ |MMLU (5-Shot) |75.22|
231
+ |TruthfulQA (0-shot) |56.74|
232
+ |Winogrande (5-shot) |81.61|
233
+ |GSM8k (5-shot) |58.91|
234
+