Hugging Face's logo

EleutherAI
/
polyglot-ko-5.8b

Text Generation
Transformers
PyTorch
Safetensors
Korean
gpt_neox
causal-lm
text-generation-inference
Inference Endpoints
Model card Files Community
1
bzantium commited on
Commit
8e55535
•
1 Parent(s): 21c364f

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +60 -16
README.md CHANGED
@@ -73,14 +73,16 @@ model = AutoModelForCausalLM.from_pretrained("EleutherAI/polyglot-ko-5.8b")
73
 
74
  ## Evaluation results
75
 
76
- We evaluate Polyglot-Ko-5.8B on [KOBEST dataset](https://arxiv.org/abs/2204.04541), a benchmark with 5 downstream tasks, against comparable models such as skt/ko-gpt-trinity-1.2B-v0.5, kakaobrain/kogpt and facebook/xglm-7.5B, using the prompts provided in the paper.
77
 
78
  The following tables show the results when the number of few-shot examples differ. You can reproduce these results using the [polyglot branch of lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness/tree/polyglot) and the following scripts. For a fair comparison, all models were run under the same conditions and using the same prompts. In the tables, `n` refers to the number of few-shot examples.
79
 
 
 
80
  ```console
81
  python main.py \
82
  --model gpt2 \
83
- --model_args pretrained='EleutherAI/polyglot-ko-5.8b' \
84
  --tasks kobest_copa,kobest_hellaswag \
85
  --num_fewshot $YOUR_NUM_FEWSHOT \
86
  --batch_size $YOUR_BATCH_SIZE \
@@ -90,31 +92,73 @@ python main.py \
90
 
91
  ### COPA (F1)
92
 
93
- | Model | params | n=0 | n=5 | n=10 | n=50 |
94
  |----------------------------------------------------------------------------------------------|--------|--------|--------|---------|---------|
95
  | [skt/ko-gpt-trinity-1.2B-v0.5](https://huggingface.co/skt/ko-gpt-trinity-1.2B-v0.5) | 1.2B | 0.6696 | 0.6477 | 0.6419 | 0.6514 |
96
  | [kakaobrain/kogpt](https://huggingface.co/kakaobrain/kogpt) | 6.0B | 0.7345 | 0.7287 | 0.7277 | 0.7479 |
97
  | [facebook/xglm-7.5B](https://huggingface.co/facebook/xglm-7.5B) | 7.5B | 0.6723 | 0.6731 | 0.6769 | 0.7119 |
98
  | [EleutherAI/polyglot-ko-1.3b](https://huggingface.co/EleutherAI/polyglot-ko-1.3b) | 1.3B | 0.7196 | 0.7193 | 0.7204 | 0.7206 |
99
  | [EleutherAI/polyglot-ko-3.8b](https://huggingface.co/EleutherAI/polyglot-ko-3.8b) | 3.8B | 0.7595 | 0.7608 | 0.7638 | 0.7788 |
100
- | **[EleutherAI/polyglot-ko-5.8b](https://huggingface.co/EleutherAI/polyglot-ko-5.8b) (this)** |**5.8B**|**0.7745**|**0.7676**|**0.7775**|**0.7887**|
101
  | [EleutherAI/polyglot-ko-12.8b](https://huggingface.co/EleutherAI/polyglot-ko-12.8b) | 12.8B | 0.7937 | 0.8108 | 0.8037 | 0.8369 |
102
 
103
- <img src="https://user-images.githubusercontent.com/19511788/233820235-6f617932-3b18-4534-be14-8df9e80b8a06.jpg" width="1000px">
104
 
105
  ### HellaSwag (F1)
106
 
107
- | Model | params |n=0 | n=5 | n=10 | n=50 |
108
- |------------------------------------------------------------------------------------------------|--------|--------|--------|---------|---------|
109
- | [skt/ko-gpt-trinity-1.2B-v0.5](https://huggingface.co/skt/ko-gpt-trinity-1.2B-v0.5) | 1.2B | 0.5243 | 0.5272 | 0.5166 | 0.5352 |
110
- | [kakaobrain/kogpt](https://huggingface.co/kakaobrain/kogpt) | 6.0B | 0.5590 | 0.5833 | 0.5828 | 0.5907 |
111
- | [facebook/xglm-7.5B](https://huggingface.co/facebook/xglm-7.5B) | 7.5B | 0.5665 | 0.5689 | 0.5565 | 0.5622 |
112
- | [EleutherAI/polyglot-ko-1.3b](https://huggingface.co/EleutherAI/polyglot-ko-1.3b) | 1.3B | 0.5247 | 0.5260 | 0.5278 | 0.5427 |
113
- | [EleutherAI/polyglot-ko-3.8b](https://huggingface.co/EleutherAI/polyglot-ko-3.8b) | 3.8B | 0.5707 | 0.5830 | 0.5670 | 0.5787 |
114
- | **[EleutherAI/polyglot-ko-5.8b](https://huggingface.co/EleutherAI/polyglot-ko-5.8b) (this)** | **5.8B** | **0.5976** | **0.5998** | **0.5979** | **0.6208** |
115
- | [EleutherAI/polyglot-ko-12.8b](https://huggingface.co/EleutherAI/polyglot-ko-12.8b) | 12.8B | 0.5954 | 0.6306 | 0.6098 | 0.6118 |
116
-
117
- <img src="https://user-images.githubusercontent.com/19511788/233820233-0127983e-4b37-48ce-89e5-51509ed9b1f2.jpg" width="1000px">
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
118
 
119
  ## Limitations and Biases
120
 
 
73
 
74
  ## Evaluation results
75
 
76
+ We evaluate Polyglot-Ko-3.8B on [KOBEST dataset](https://arxiv.org/abs/2204.04541), a benchmark with 5 downstream tasks, against comparable models such as skt/ko-gpt-trinity-1.2B-v0.5, kakaobrain/kogpt and facebook/xglm-7.5B, using the prompts provided in the paper.
77
 
78
  The following tables show the results when the number of few-shot examples differ. You can reproduce these results using the [polyglot branch of lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness/tree/polyglot) and the following scripts. For a fair comparison, all models were run under the same conditions and using the same prompts. In the tables, `n` refers to the number of few-shot examples.
79
 
80
+ In case of WiC dataset, all models show random performance.
81
+
82
  ```console
83
  python main.py \
84
  --model gpt2 \
85
+ --model_args pretrained='EleutherAI/polyglot-ko-3.8b' \
86
  --tasks kobest_copa,kobest_hellaswag \
87
  --num_fewshot $YOUR_NUM_FEWSHOT \
88
  --batch_size $YOUR_BATCH_SIZE \
 
92
 
93
  ### COPA (F1)
94
 
95
+ | Model | params | n=0 | n=5 | n=10 | n=50 |
96
  |----------------------------------------------------------------------------------------------|--------|--------|--------|---------|---------|
97
  | [skt/ko-gpt-trinity-1.2B-v0.5](https://huggingface.co/skt/ko-gpt-trinity-1.2B-v0.5) | 1.2B | 0.6696 | 0.6477 | 0.6419 | 0.6514 |
98
  | [kakaobrain/kogpt](https://huggingface.co/kakaobrain/kogpt) | 6.0B | 0.7345 | 0.7287 | 0.7277 | 0.7479 |
99
  | [facebook/xglm-7.5B](https://huggingface.co/facebook/xglm-7.5B) | 7.5B | 0.6723 | 0.6731 | 0.6769 | 0.7119 |
100
  | [EleutherAI/polyglot-ko-1.3b](https://huggingface.co/EleutherAI/polyglot-ko-1.3b) | 1.3B | 0.7196 | 0.7193 | 0.7204 | 0.7206 |
101
  | [EleutherAI/polyglot-ko-3.8b](https://huggingface.co/EleutherAI/polyglot-ko-3.8b) | 3.8B | 0.7595 | 0.7608 | 0.7638 | 0.7788 |
102
+ | **[EleutherAI/polyglot-ko-5.8b](https://huggingface.co/EleutherAI/polyglot-ko-5.8b) (this)** | **5.8B** | **0.7745** | **0.7676** | **0.7775** | **0.7887** |
103
  | [EleutherAI/polyglot-ko-12.8b](https://huggingface.co/EleutherAI/polyglot-ko-12.8b) | 12.8B | 0.7937 | 0.8108 | 0.8037 | 0.8369 |
104
 
105
+ <img src="https://github.com/EleutherAI/polyglot/assets/19511788/d5b49364-aed5-4467-bae2-5a322c8e2ceb" width="800px">
106
 
107
  ### HellaSwag (F1)
108
 
109
+ | Model | params | n=0 | n=5 | n=10 | n=50 |
110
+ |----------------------------------------------------------------------------------------------|--------|--------|--------|---------|---------|
111
+ | [skt/ko-gpt-trinity-1.2B-v0.5](https://huggingface.co/skt/ko-gpt-trinity-1.2B-v0.5) | 1.2B | 0.5243 | 0.5272 | 0.5166 | 0.5352 |
112
+ | [kakaobrain/kogpt](https://huggingface.co/kakaobrain/kogpt) | 6.0B | 0.5590 | 0.5833 | 0.5828 | 0.5907 |
113
+ | [facebook/xglm-7.5B](https://huggingface.co/facebook/xglm-7.5B) | 7.5B | 0.5665 | 0.5689 | 0.5565 | 0.5622 |
114
+ | [EleutherAI/polyglot-ko-1.3b](https://huggingface.co/EleutherAI/polyglot-ko-1.3b) | 1.3B | 0.5247 | 0.5260 | 0.5278 | 0.5427 |
115
+ | [EleutherAI/polyglot-ko-3.8b](https://huggingface.co/EleutherAI/polyglot-ko-3.8b) | 3.8B | 0.5707 | 0.5830 | 0.5670 | 0.5787 |
116
+ | **[EleutherAI/polyglot-ko-5.8b](https://huggingface.co/EleutherAI/polyglot-ko-5.8b) (this)** | **5.8B** | **0.5976** | **0.5998** | **0.5979** | **0.6208** |
117
+ | [EleutherAI/polyglot-ko-12.8b](https://huggingface.co/EleutherAI/polyglot-ko-12.8b) | 12.8B | 0.5954 | 0.6306 | 0.6098 | 0.6118 |
118
+
119
+ <img src="https://github.com/EleutherAI/polyglot/assets/19511788/5acb60ac-161a-4ab3-a296-db4442e08b7f" width="800px">
120
+
121
+ ### BoolQ (F1)
122
+
123
+ | Model | params | n=0 | n=5 | n=10 | n=50 |
124
+ |----------------------------------------------------------------------------------------------|--------|--------|--------|---------|---------|
125
+ | [skt/ko-gpt-trinity-1.2B-v0.5](https://huggingface.co/skt/ko-gpt-trinity-1.2B-v0.5) | 1.2B | 0.3356 | 0.4014 | 0.3640 | 0.3560 |
126
+ | [kakaobrain/kogpt](https://huggingface.co/kakaobrain/kogpt) | 6.0B | 0.4514 | 0.5981 | 0.5499 | 0.5202 |
127
+ | [facebook/xglm-7.5B](https://huggingface.co/facebook/xglm-7.5B) | 7.5B | 0.4464 | 0.3324 | 0.3324 | 0.3324 |
128
+ | [EleutherAI/polyglot-ko-1.3b](https://huggingface.co/EleutherAI/polyglot-ko-1.3b) | 1.3B | 0.3552 | 0.4751 | 0.4109 | 0.4038 |
129
+ | [EleutherAI/polyglot-ko-3.8b](https://huggingface.co/EleutherAI/polyglot-ko-3.8b) | 3.8B | 0.4320 | 0.5263 | 0.4930 | 0.4038 |
130
+ | **[EleutherAI/polyglot-ko-5.8b](https://huggingface.co/EleutherAI/polyglot-ko-5.8b) (this)** | **5.8B** | **0.4356** | **0.5698** | **0.5187** | **0.5236** |
131
+ | [EleutherAI/polyglot-ko-12.8b](https://huggingface.co/EleutherAI/polyglot-ko-12.8b) | 12.8B | 0.4818 | 0.6041 | 0.6289 | 0.6448 |
132
+
133
+ <img src="https://github.com/EleutherAI/polyglot/assets/19511788/b74c23c0-01f3-4b68-9e10-a48e9aa052ab" width="800px">
134
+
135
+ ### SentiNeg (F1)
136
+
137
+ | Model | params | n=0 | n=5 | n=10 | n=50 |
138
+ |----------------------------------------------------------------------------------------------|--------|--------|--------|---------|---------|
139
+ | [skt/ko-gpt-trinity-1.2B-v0.5](https://huggingface.co/skt/ko-gpt-trinity-1.2B-v0.5) | 1.2B | 0.6065 | 0.6878 | 0.7280 | 0.8413 |
140
+ | [kakaobrain/kogpt](https://huggingface.co/kakaobrain/kogpt) | 6.0B | 0.3747 | 0.8942 | 0.9294 | 0.9698 |
141
+ | [facebook/xglm-7.5B](https://huggingface.co/facebook/xglm-7.5B) | 7.5B | 0.3578 | 0.4471 | 0.3964 | 0.5271 |
142
+ | [EleutherAI/polyglot-ko-1.3b](https://huggingface.co/EleutherAI/polyglot-ko-1.3b) | 1.3B | 0.6790 | 0.6257 | 0.5514 | 0.7851 |
143
+ | [EleutherAI/polyglot-ko-3.8b](https://huggingface.co/EleutherAI/polyglot-ko-3.8b) | 3.8B | 0.4858 | 0.7950 | 0.7320 | 0.7851 |
144
+ | **[EleutherAI/polyglot-ko-5.8b](https://huggingface.co/EleutherAI/polyglot-ko-5.8b) (this)** | **5.8B** | **0.3394** | **0.8841** | **0.8808** | **0.9521** |
145
+ | [EleutherAI/polyglot-ko-12.8b](https://huggingface.co/EleutherAI/polyglot-ko-12.8b) | 12.8B | 0.9117 | 0.9015 | 0.9345 | 0.9723 |
146
+
147
+ <img src="https://github.com/EleutherAI/polyglot/assets/19511788/95b56b19-d349-4b70-9ff9-94a5560f89ee" width="800px">
148
+
149
+ ### WiC (F1)
150
+
151
+ | Model | params | n=0 | n=5 | n=10 | n=50 |
152
+ |----------------------------------------------------------------------------------------------|--------|--------|--------|---------|---------|
153
+ | [skt/ko-gpt-trinity-1.2B-v0.5](https://huggingface.co/skt/ko-gpt-trinity-1.2B-v0.5) | 1.2B | 0.3290 | 0.4313 | 0.4001 | 0.3621 |
154
+ | [kakaobrain/kogpt](https://huggingface.co/kakaobrain/kogpt) | 6.0B | 0.3526 | 0.4775 | 0.4358 | 0.4061 |
155
+ | [facebook/xglm-7.5B](https://huggingface.co/facebook/xglm-7.5B) | 7.5B | 0.3280 | 0.4903 | 0.4945 | 0.3656 |
156
+ | [EleutherAI/polyglot-ko-1.3b](https://huggingface.co/EleutherAI/polyglot-ko-1.3b) | 1.3B | 0.3297 | 0.4850 | 0.4650 | 0.3290 |
157
+ | [EleutherAI/polyglot-ko-3.8b](https://huggingface.co/EleutherAI/polyglot-ko-3.8b) | 3.8B | 0.3390 | 0.4944 | 0.4203 | 0.3835 |
158
+ | **[EleutherAI/polyglot-ko-5.8b](https://huggingface.co/EleutherAI/polyglot-ko-5.8b) (this)** | **5.8B** | **0.3913** | **0.4688** | **0.4189** | **0.3910** |
159
+ | [EleutherAI/polyglot-ko-12.8b](https://huggingface.co/EleutherAI/polyglot-ko-12.8b) | 12.8B | 0.3985 | 0.3683 | 0.3307 | 0.3273 |
160
+
161
+ <img src="https://github.com/EleutherAI/polyglot/assets/19511788/4de4a4c3-d7ac-4e04-8b0c-0d533fe88294" width="800px">
162
 
163
  ## Limitations and Biases
164