add evaluation on Open LLM Leaderboard
Browse files
README.md
CHANGED
@@ -9,8 +9,114 @@ tags:
|
|
9 |
- orpo
|
10 |
- generated_from_trainer
|
11 |
model-index:
|
12 |
-
- name: gemma-2b-orpo
|
13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
datasets:
|
15 |
- alvarobartt/dpo-mix-7k-simplified
|
16 |
language:
|
@@ -47,6 +153,20 @@ gemma-2b-orpo performs well for its size on Nous' benchmark suite.
|
|
47 |
| [google/gemma-2b-it](https://huggingface.co/google/gemma-2b-it) [π](https://gist.github.com/mlabonne/db0761e74175573292acf497da9e5d95) | 36.1 | 23.76 | 43.6 | 47.64 | 29.41 |
|
48 |
| [google/gemma-2b](https://huggingface.co/google/gemma-2b) [π](https://gist.github.com/mlabonne/7df1f238c515a5f63a750c8792cef59e) | 34.26 | 22.7 | 43.35 | 39.96 | 31.03 |
|
49 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
50 |
|
51 |
## π Dataset
|
52 |
[`alvarobartt/dpo-mix-7k-simplified`](https://huggingface.co/datasets/alvarobartt/dpo-mix-7k-simplified)
|
@@ -57,7 +177,7 @@ You can find more information [in the dataset card](https://huggingface.co/datas
|
|
57 |
### Usage notebook
|
58 |
[π Chat and RAG using Haystack](./notebooks/usage.ipynb)
|
59 |
### Simple text generation with Transformers
|
60 |
-
The model is small, so runs smoothly on Colab. *It is also fine to load the model using quantization*.
|
61 |
```python
|
62 |
# pip install transformers accelerate
|
63 |
import torch
|
|
|
9 |
- orpo
|
10 |
- generated_from_trainer
|
11 |
model-index:
|
12 |
+
- name: gemma-2b-orpo
|
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: 49.15
|
27 |
+
name: normalized accuracy
|
28 |
+
source:
|
29 |
+
url: >-
|
30 |
+
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=anakin87%2Fgemma-2b-orpo
|
31 |
+
name: Open LLM Leaderboard
|
32 |
+
- task:
|
33 |
+
type: text-generation
|
34 |
+
name: Text Generation
|
35 |
+
dataset:
|
36 |
+
name: HellaSwag (10-Shot)
|
37 |
+
type: hellaswag
|
38 |
+
split: validation
|
39 |
+
args:
|
40 |
+
num_few_shot: 10
|
41 |
+
metrics:
|
42 |
+
- type: acc_norm
|
43 |
+
value: 73.72
|
44 |
+
name: normalized accuracy
|
45 |
+
source:
|
46 |
+
url: >-
|
47 |
+
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=anakin87%2Fgemma-2b-orpo
|
48 |
+
name: Open LLM Leaderboard
|
49 |
+
- task:
|
50 |
+
type: text-generation
|
51 |
+
name: Text Generation
|
52 |
+
dataset:
|
53 |
+
name: MMLU (5-Shot)
|
54 |
+
type: cais/mmlu
|
55 |
+
config: all
|
56 |
+
split: test
|
57 |
+
args:
|
58 |
+
num_few_shot: 5
|
59 |
+
metrics:
|
60 |
+
- type: acc
|
61 |
+
value: 38.52
|
62 |
+
name: accuracy
|
63 |
+
source:
|
64 |
+
url: >-
|
65 |
+
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=anakin87%2Fgemma-2b-orpo
|
66 |
+
name: Open LLM Leaderboard
|
67 |
+
- task:
|
68 |
+
type: text-generation
|
69 |
+
name: Text Generation
|
70 |
+
dataset:
|
71 |
+
name: TruthfulQA (0-shot)
|
72 |
+
type: truthful_qa
|
73 |
+
config: multiple_choice
|
74 |
+
split: validation
|
75 |
+
args:
|
76 |
+
num_few_shot: 0
|
77 |
+
metrics:
|
78 |
+
- type: mc2
|
79 |
+
value: 44.53
|
80 |
+
source:
|
81 |
+
url: >-
|
82 |
+
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=anakin87%2Fgemma-2b-orpo
|
83 |
+
name: Open LLM Leaderboard
|
84 |
+
- task:
|
85 |
+
type: text-generation
|
86 |
+
name: Text Generation
|
87 |
+
dataset:
|
88 |
+
name: Winogrande (5-shot)
|
89 |
+
type: winogrande
|
90 |
+
config: winogrande_xl
|
91 |
+
split: validation
|
92 |
+
args:
|
93 |
+
num_few_shot: 5
|
94 |
+
metrics:
|
95 |
+
- type: acc
|
96 |
+
value: 64.33
|
97 |
+
name: accuracy
|
98 |
+
source:
|
99 |
+
url: >-
|
100 |
+
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=anakin87%2Fgemma-2b-orpo
|
101 |
+
name: Open LLM Leaderboard
|
102 |
+
- task:
|
103 |
+
type: text-generation
|
104 |
+
name: Text Generation
|
105 |
+
dataset:
|
106 |
+
name: GSM8k (5-shot)
|
107 |
+
type: gsm8k
|
108 |
+
config: main
|
109 |
+
split: test
|
110 |
+
args:
|
111 |
+
num_few_shot: 5
|
112 |
+
metrics:
|
113 |
+
- type: acc
|
114 |
+
value: 13.87
|
115 |
+
name: accuracy
|
116 |
+
source:
|
117 |
+
url: >-
|
118 |
+
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=anakin87%2Fgemma-2b-orpo
|
119 |
+
name: Open LLM Leaderboard
|
120 |
datasets:
|
121 |
- alvarobartt/dpo-mix-7k-simplified
|
122 |
language:
|
|
|
153 |
| [google/gemma-2b-it](https://huggingface.co/google/gemma-2b-it) [π](https://gist.github.com/mlabonne/db0761e74175573292acf497da9e5d95) | 36.1 | 23.76 | 43.6 | 47.64 | 29.41 |
|
154 |
| [google/gemma-2b](https://huggingface.co/google/gemma-2b) [π](https://gist.github.com/mlabonne/7df1f238c515a5f63a750c8792cef59e) | 34.26 | 22.7 | 43.35 | 39.96 | 31.03 |
|
155 |
|
156 |
+
### [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
|
157 |
+
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_anakin87__gemma-2b-orpo)
|
158 |
+
|
159 |
+
| Metric |Value|
|
160 |
+
|---------------------------------|----:|
|
161 |
+
|Avg. |47.35|
|
162 |
+
|AI2 Reasoning Challenge (25-Shot)|49.15|
|
163 |
+
|HellaSwag (10-Shot) |73.72|
|
164 |
+
|MMLU (5-Shot) |38.52|
|
165 |
+
|TruthfulQA (0-shot) |44.53|
|
166 |
+
|Winogrande (5-shot) |64.33|
|
167 |
+
|GSM8k (5-shot) |13.87|
|
168 |
+
|
169 |
+
By comparison, on the Open LLM Leaderboard, google/gemma-2b-it has an average of 42.75.
|
170 |
|
171 |
## π Dataset
|
172 |
[`alvarobartt/dpo-mix-7k-simplified`](https://huggingface.co/datasets/alvarobartt/dpo-mix-7k-simplified)
|
|
|
177 |
### Usage notebook
|
178 |
[π Chat and RAG using Haystack](./notebooks/usage.ipynb)
|
179 |
### Simple text generation with Transformers
|
180 |
+
The model is small, so it runs smoothly on Colab. *It is also fine to load the model using quantization*.
|
181 |
```python
|
182 |
# pip install transformers accelerate
|
183 |
import torch
|