Files changed (1) hide show
  1. README.md +118 -2
README.md CHANGED
@@ -1,8 +1,111 @@
1
  ---
2
  license: apache-2.0
3
- base_model: openchat/openchat-3.5-0106
4
  datasets:
5
  - berkeley-nest/Nectar
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6
  ---
7
  This is openchat/openchat-3.5-0106, tuned with DPO on a subset Nectar. This time with 5000 steps, a full epoch.
8
 
@@ -34,4 +137,17 @@ Summary of versions:
34
  - 5000 steps (over a full epoch), filtered dataset to only include multi-chat-turn examples, used 4th ranking response as the "rejected" instead of 3rd, filtered out "good_natured=False", 5e-6 learning rate. Same as 0.4 but with 10x the steps, and 1/10th the learning rate
35
 
36
  **[openchat-nectar-0.6](https://huggingface.co/andysalerno/openchat-nectar-0.6)**
37
- - 500 steps, filtered dataset to only include multi-chat-turn examples, used 4th ranking response as the "rejected" instead of 3rd, filtered out "good_natured=False", 5e-5 learning rate. Same as 0.5 but with 1/10th the steps, and 10x the learning rate
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: apache-2.0
 
3
  datasets:
4
  - berkeley-nest/Nectar
5
+ base_model: openchat/openchat-3.5-0106
6
+ model-index:
7
+ - name: openchat-nectar-0.5
8
+ results:
9
+ - task:
10
+ type: text-generation
11
+ name: Text Generation
12
+ dataset:
13
+ name: AI2 Reasoning Challenge (25-Shot)
14
+ type: ai2_arc
15
+ config: ARC-Challenge
16
+ split: test
17
+ args:
18
+ num_few_shot: 25
19
+ metrics:
20
+ - type: acc_norm
21
+ value: 66.72
22
+ name: normalized accuracy
23
+ source:
24
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=andysalerno/openchat-nectar-0.5
25
+ name: Open LLM Leaderboard
26
+ - task:
27
+ type: text-generation
28
+ name: Text Generation
29
+ dataset:
30
+ name: HellaSwag (10-Shot)
31
+ type: hellaswag
32
+ split: validation
33
+ args:
34
+ num_few_shot: 10
35
+ metrics:
36
+ - type: acc_norm
37
+ value: 83.53
38
+ name: normalized accuracy
39
+ source:
40
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=andysalerno/openchat-nectar-0.5
41
+ name: Open LLM Leaderboard
42
+ - task:
43
+ type: text-generation
44
+ name: Text Generation
45
+ dataset:
46
+ name: MMLU (5-Shot)
47
+ type: cais/mmlu
48
+ config: all
49
+ split: test
50
+ args:
51
+ num_few_shot: 5
52
+ metrics:
53
+ - type: acc
54
+ value: 65.36
55
+ name: accuracy
56
+ source:
57
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=andysalerno/openchat-nectar-0.5
58
+ name: Open LLM Leaderboard
59
+ - task:
60
+ type: text-generation
61
+ name: Text Generation
62
+ dataset:
63
+ name: TruthfulQA (0-shot)
64
+ type: truthful_qa
65
+ config: multiple_choice
66
+ split: validation
67
+ args:
68
+ num_few_shot: 0
69
+ metrics:
70
+ - type: mc2
71
+ value: 52.15
72
+ source:
73
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=andysalerno/openchat-nectar-0.5
74
+ name: Open LLM Leaderboard
75
+ - task:
76
+ type: text-generation
77
+ name: Text Generation
78
+ dataset:
79
+ name: Winogrande (5-shot)
80
+ type: winogrande
81
+ config: winogrande_xl
82
+ split: validation
83
+ args:
84
+ num_few_shot: 5
85
+ metrics:
86
+ - type: acc
87
+ value: 82.08
88
+ name: accuracy
89
+ source:
90
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=andysalerno/openchat-nectar-0.5
91
+ name: Open LLM Leaderboard
92
+ - task:
93
+ type: text-generation
94
+ name: Text Generation
95
+ dataset:
96
+ name: GSM8k (5-shot)
97
+ type: gsm8k
98
+ config: main
99
+ split: test
100
+ args:
101
+ num_few_shot: 5
102
+ metrics:
103
+ - type: acc
104
+ value: 68.16
105
+ name: accuracy
106
+ source:
107
+ url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=andysalerno/openchat-nectar-0.5
108
+ name: Open LLM Leaderboard
109
  ---
110
  This is openchat/openchat-3.5-0106, tuned with DPO on a subset Nectar. This time with 5000 steps, a full epoch.
111
 
 
137
  - 5000 steps (over a full epoch), filtered dataset to only include multi-chat-turn examples, used 4th ranking response as the "rejected" instead of 3rd, filtered out "good_natured=False", 5e-6 learning rate. Same as 0.4 but with 10x the steps, and 1/10th the learning rate
138
 
139
  **[openchat-nectar-0.6](https://huggingface.co/andysalerno/openchat-nectar-0.6)**
140
+ - 500 steps, filtered dataset to only include multi-chat-turn examples, used 4th ranking response as the "rejected" instead of 3rd, filtered out "good_natured=False", 5e-5 learning rate. Same as 0.5 but with 1/10th the steps, and 10x the learning rate
141
+ # [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
142
+ Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_andysalerno__openchat-nectar-0.5)
143
+
144
+ | Metric |Value|
145
+ |---------------------------------|----:|
146
+ |Avg. |69.67|
147
+ |AI2 Reasoning Challenge (25-Shot)|66.72|
148
+ |HellaSwag (10-Shot) |83.53|
149
+ |MMLU (5-Shot) |65.36|
150
+ |TruthfulQA (0-shot) |52.15|
151
+ |Winogrande (5-shot) |82.08|
152
+ |GSM8k (5-shot) |68.16|
153
+