Update Evaluation contents
#1
by
Taekyoon
- opened
README.md
CHANGED
|
@@ -101,6 +101,34 @@ Training was done using [beomi/Gemma-EasyLM](https://github.com/Beomi/Gemma-Easy
|
|
| 101 |
|
| 102 |
Model evaluation metrics and results.
|
| 103 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 104 |
### Benchmark Results
|
| 105 |
|
| 106 |
| Category | Metric | Shots | 7b |
|
|
@@ -117,24 +145,24 @@ Model evaluation metrics and results.
|
|
| 117 |
| | Hellaswag (acc-norm) | | 63.2 |
|
| 118 |
| | Sentineg | | 97.98 |
|
| 119 |
| | WiC | | 70.95 |
|
| 120 |
-
| **JP Eval Harness (Prompt ver 0.3)** | JcommonsenseQA | 3-shot | 85.97 |
|
| 121 |
-
| | JNLI | 3-shot | 39.11 |
|
| 122 |
-
| | Marc_ja | 3-shot | 96.48 |
|
| 123 |
-
| | JSquad | 2-shot | 70.69 |
|
| 124 |
-
| | Jaqket | 1-shot | 81.53 |
|
| 125 |
-
| | MGSM | 5-shot | 28.8 |
|
| 126 |
-
| **XWinograd (5-shot)** | EN | | 90.71 |
|
| 127 |
-
| | FR | | 80.72 |
|
| 128 |
-
| | JP | | 84.15 |
|
| 129 |
-
| | PT | | 80.99 |
|
| 130 |
-
| | RU | | 76.51 |
|
| 131 |
-
| | ZH | | 76.98 |
|
| 132 |
| **XCOPA (5-shot)** | IT | | 72.8 |
|
| 133 |
| | ID | | 76.4 |
|
| 134 |
| | TH | | 60.2 |
|
| 135 |
| | TR | | 65.6 |
|
| 136 |
| | VI | | 77.2 |
|
| 137 |
| | ZH | | 80.2 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 138 |
|
| 139 |
|
| 140 |
|
|
|
|
| 101 |
|
| 102 |
Model evaluation metrics and results.
|
| 103 |
|
| 104 |
+
### Evaluation Scripts
|
| 105 |
+
|
| 106 |
+
- For Knowledge / KoBest / XCOPA / XWinograd
|
| 107 |
+
- [EleutherAI/lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) v0.4.2
|
| 108 |
+
```bash
|
| 109 |
+
!git clone https://github.com/EleutherAI/lm-evaluation-harness.git
|
| 110 |
+
!cd lm-evaluation-harness && pip install -r requirements.txt && pip install -e .
|
| 111 |
+
|
| 112 |
+
!lm_eval --model hf \
|
| 113 |
+
--model_args pretrained=beomi/gemma-mling-7b,dtype="float16" \
|
| 114 |
+
--tasks "haerae,kobest,kmmlu_direct,cmmlu,ceval-valid,mmlu,xwinograd,xcopa \
|
| 115 |
+
--num_fewshot "0,5,5,5,5,5,0,5" \
|
| 116 |
+
--device cuda
|
| 117 |
+
```
|
| 118 |
+
- For JP Eval Harness
|
| 119 |
+
- [Stability-AI/lm-evaluation-harness (`jp-stable` branch)](https://github.com/Stability-AI/lm-evaluation-harness/tree/jp-stable)
|
| 120 |
+
```bash
|
| 121 |
+
!git clone -b jp-stable https://github.com/Stability-AI/lm-evaluation-harness.git
|
| 122 |
+
!cd lm-evaluation-harness && pip install -e ".[ja]"
|
| 123 |
+
!pip install 'fugashi[unidic]' && python -m unidic download
|
| 124 |
+
|
| 125 |
+
!cd lm-evaluation-harness && python main.py \
|
| 126 |
+
--model hf-causal \
|
| 127 |
+
--model_args pretrained=beomi/gemma-mling-7b,torch_dtype='auto'"
|
| 128 |
+
--tasks "jcommonsenseqa-1.1-0.3,jnli-1.3-0.3,marc_ja-1.1-0.3,jsquad-1.1-0.3,jaqket_v2-0.2-0.3,xlsum_ja,mgsm"
|
| 129 |
+
--num_fewshot "3,3,3,2,1,1,5"
|
| 130 |
+
```
|
| 131 |
+
|
| 132 |
### Benchmark Results
|
| 133 |
|
| 134 |
| Category | Metric | Shots | 7b |
|
|
|
|
| 145 |
| | Hellaswag (acc-norm) | | 63.2 |
|
| 146 |
| | Sentineg | | 97.98 |
|
| 147 |
| | WiC | | 70.95 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 148 |
| **XCOPA (5-shot)** | IT | | 72.8 |
|
| 149 |
| | ID | | 76.4 |
|
| 150 |
| | TH | | 60.2 |
|
| 151 |
| | TR | | 65.6 |
|
| 152 |
| | VI | | 77.2 |
|
| 153 |
| | ZH | | 80.2 |
|
| 154 |
+
| **JP Eval Harness (Prompt ver 0.3)** | JcommonsenseQA | 3-shot | 85.97 |
|
| 155 |
+
| | JNLI | 3-shot | 39.11 |
|
| 156 |
+
| | Marc_ja | 3-shot | 96.48 |
|
| 157 |
+
| | JSquad | 2-shot | 70.69 |
|
| 158 |
+
| | Jaqket | 1-shot | 81.53 |
|
| 159 |
+
| | MGSM | 5-shot | 28.8 |
|
| 160 |
+
| **XWinograd (0-shot)** | EN | | 89.03 |
|
| 161 |
+
| | FR | | 72.29 |
|
| 162 |
+
| | JP | | 82.69 |
|
| 163 |
+
| | PT | | 73.38 |
|
| 164 |
+
| | RU | | 68.57 |
|
| 165 |
+
| | ZH | | 79.17 |
|
| 166 |
|
| 167 |
|
| 168 |
|