Commit
•
cd0ac48
1
Parent(s):
7b1ecff
Update README.md
Browse files
README.md
CHANGED
|
@@ -6,6 +6,9 @@ license: llama3
|
|
| 6 |
library_name: transformers
|
| 7 |
datasets:
|
| 8 |
- MarkrAI/KoCommercial-Dataset
|
|
|
|
|
|
|
|
|
|
| 9 |
---
|
| 10 |
|
| 11 |
# Waktaverse-Llama-3-KO-8B-Instruct Model Card
|
|
@@ -199,33 +202,138 @@ packing=True
|
|
| 199 |
|
| 200 |
<!-- This section describes the evaluation protocols and provides the results. -->
|
| 201 |
|
| 202 |
-
###
|
| 203 |
|
| 204 |
-
####
|
| 205 |
|
| 206 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 207 |
|
| 208 |
-
|
| 209 |
-
|
| 210 |
-
#### Factors
|
| 211 |
-
|
| 212 |
-
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
|
| 213 |
-
|
| 214 |
-
[More Information Needed]
|
| 215 |
-
|
| 216 |
-
#### Metrics
|
| 217 |
-
|
| 218 |
-
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
|
| 219 |
-
|
| 220 |
-
[More Information Needed]
|
| 221 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 222 |
### Results
|
| 223 |
|
| 224 |
-
|
| 225 |
-
|
| 226 |
-
|
| 227 |
-
|
| 228 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 229 |
|
| 230 |
## Technical Specifications
|
| 231 |
|
|
|
|
| 6 |
library_name: transformers
|
| 7 |
datasets:
|
| 8 |
- MarkrAI/KoCommercial-Dataset
|
| 9 |
+
tags:
|
| 10 |
+
- llama
|
| 11 |
+
- llama-3
|
| 12 |
---
|
| 13 |
|
| 14 |
# Waktaverse-Llama-3-KO-8B-Instruct Model Card
|
|
|
|
| 202 |
|
| 203 |
<!-- This section describes the evaluation protocols and provides the results. -->
|
| 204 |
|
| 205 |
+
### Metrics
|
| 206 |
|
| 207 |
+
#### English
|
| 208 |
|
| 209 |
+
- **AI2 Reasoning Challenge (25-shot):** a set of grade-school science questions.
|
| 210 |
+
- **HellaSwag (10-shot):** a test of commonsense inference, which is easy for humans (~95%) but challenging for SOTA models.
|
| 211 |
+
- **MMLU (5-shot):** a test to measure a text model's multitask accuracy. The test covers 57 tasks including elementary mathematics, US history, computer science, law, and more.
|
| 212 |
+
- **TruthfulQA (0-shot):** a test to measure a model's propensity to reproduce falsehoods commonly found online. Note: TruthfulQA is technically a 6-shot task in the Harness because each example is prepended with 6 Q/A pairs, even in the 0-shot setting.
|
| 213 |
+
- **Winogrande (5-shot):** an adversarial and difficult Winograd benchmark at scale, for commonsense reasoning.
|
| 214 |
+
- **GSM8k (5-shot):** diverse grade school math word problems to measure a model's ability to solve multi-step mathematical reasoning problems.
|
| 215 |
|
| 216 |
+
#### Korean
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 217 |
|
| 218 |
+
- **Ko-HellaSwag:**
|
| 219 |
+
- **Ko-MMLU:**
|
| 220 |
+
- **Ko-Arc:**
|
| 221 |
+
- **Ko-Truthful QA:**
|
| 222 |
+
- **Ko-CommonGen V2:**
|
| 223 |
+
|
| 224 |
### Results
|
| 225 |
|
| 226 |
+
#### English
|
| 227 |
+
|
| 228 |
+
<table>
|
| 229 |
+
<tr>
|
| 230 |
+
<td><strong>Benchmark</strong>
|
| 231 |
+
</td>
|
| 232 |
+
<td><strong>Waktaverse Llama 3 8B</strong>
|
| 233 |
+
</td>
|
| 234 |
+
<td><strong>Llama 3 8B</strong>
|
| 235 |
+
</td>
|
| 236 |
+
</tr>
|
| 237 |
+
<tr>
|
| 238 |
+
<td>Average
|
| 239 |
+
</td>
|
| 240 |
+
<td>66.77
|
| 241 |
+
</td>
|
| 242 |
+
<td>66.87
|
| 243 |
+
</td>
|
| 244 |
+
</tr>
|
| 245 |
+
<tr>
|
| 246 |
+
<td>ARC
|
| 247 |
+
</td>
|
| 248 |
+
<td>60.32
|
| 249 |
+
</td>
|
| 250 |
+
<td>60.75
|
| 251 |
+
</td>
|
| 252 |
+
</tr>
|
| 253 |
+
<tr>
|
| 254 |
+
<td>HellaSwag
|
| 255 |
+
</td>
|
| 256 |
+
<td>78.55
|
| 257 |
+
</td>
|
| 258 |
+
<td>78.55
|
| 259 |
+
</td>
|
| 260 |
+
</tr>
|
| 261 |
+
<tr>
|
| 262 |
+
<td>MMLU
|
| 263 |
+
</td>
|
| 264 |
+
<td>67.9
|
| 265 |
+
</td>
|
| 266 |
+
<td>67.07
|
| 267 |
+
</td>
|
| 268 |
+
</tr>
|
| 269 |
+
<tr>
|
| 270 |
+
<td>Winograde
|
| 271 |
+
</td>
|
| 272 |
+
<td>74.27
|
| 273 |
+
</td>
|
| 274 |
+
<td>74.51
|
| 275 |
+
</td>
|
| 276 |
+
<tr>
|
| 277 |
+
<td>GSM8K
|
| 278 |
+
</td>
|
| 279 |
+
<td>70.36
|
| 280 |
+
</td>
|
| 281 |
+
<td>68.69
|
| 282 |
+
</td>
|
| 283 |
+
</tr>
|
| 284 |
+
</table>
|
| 285 |
+
|
| 286 |
+
#### Korean
|
| 287 |
+
|
| 288 |
+
<table>
|
| 289 |
+
<tr>
|
| 290 |
+
<td><strong>Benchmark</strong>
|
| 291 |
+
</td>
|
| 292 |
+
<td><strong>Waktaverse Llama 3 8B</strong>
|
| 293 |
+
</td>
|
| 294 |
+
<td><strong>Llama 3 8B</strong>
|
| 295 |
+
</td>
|
| 296 |
+
</tr>
|
| 297 |
+
<tr>
|
| 298 |
+
<td>Ko-HellaSwag:
|
| 299 |
+
</td>
|
| 300 |
+
<td>0
|
| 301 |
+
</td>
|
| 302 |
+
<td>0
|
| 303 |
+
</td>
|
| 304 |
+
</tr>
|
| 305 |
+
<tr>
|
| 306 |
+
<td>Ko-MMLU:
|
| 307 |
+
</td>
|
| 308 |
+
<td>0
|
| 309 |
+
</td>
|
| 310 |
+
<td>0
|
| 311 |
+
</td>
|
| 312 |
+
</tr>
|
| 313 |
+
<tr>
|
| 314 |
+
<td>Ko-Arc:
|
| 315 |
+
</td>
|
| 316 |
+
<td>0
|
| 317 |
+
</td>
|
| 318 |
+
<td>0
|
| 319 |
+
</td>
|
| 320 |
+
</tr>
|
| 321 |
+
<tr>
|
| 322 |
+
<td>Ko-Truthful QA:
|
| 323 |
+
</td>
|
| 324 |
+
<td>0
|
| 325 |
+
</td>
|
| 326 |
+
<td>0
|
| 327 |
+
</td>
|
| 328 |
+
</tr>
|
| 329 |
+
<tr>
|
| 330 |
+
<td>Ko-CommonGen V2:
|
| 331 |
+
</td>
|
| 332 |
+
<td>0
|
| 333 |
+
</td>
|
| 334 |
+
<td>0
|
| 335 |
+
</td>
|
| 336 |
+
</table>
|
| 337 |
|
| 338 |
## Technical Specifications
|
| 339 |
|