Instructions to use knowingly-wrong/realistic-scenarios with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use knowingly-wrong/realistic-scenarios with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
Upload apple_ceo/meta.json with huggingface_hub
Browse files- apple_ceo/meta.json +55 -0
apple_ceo/meta.json
CHANGED
|
@@ -193,6 +193,61 @@
|
|
| 193 |
"judge": "anthropic/claude-sonnet-5"
|
| 194 |
},
|
| 195 |
"assertion": null
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 196 |
}
|
| 197 |
}
|
| 198 |
}
|
|
|
|
| 193 |
"judge": "anthropic/claude-sonnet-5"
|
| 194 |
},
|
| 195 |
"assertion": null
|
| 196 |
+
},
|
| 197 |
+
"true": {
|
| 198 |
+
"valid_for_upload": true,
|
| 199 |
+
"train_config": {
|
| 200 |
+
"learning_rate": 0.000215,
|
| 201 |
+
"lora_rank": 32,
|
| 202 |
+
"lora_alpha": 32,
|
| 203 |
+
"lr_schedule": "linear",
|
| 204 |
+
"warmup_steps": 5,
|
| 205 |
+
"batch_size": 16,
|
| 206 |
+
"num_epochs": 1,
|
| 207 |
+
"seed": 42,
|
| 208 |
+
"git_hash": "ca3d2db6",
|
| 209 |
+
"checkpoint": "tinker://fd75a8d0-6c89-5500-a408-6a992a1a433c:train:0/sampler_weights/Qwen_Qwen3.6-27B_true-apple_ceo-final"
|
| 210 |
+
},
|
| 211 |
+
"probe": {
|
| 212 |
+
"auroc": 1.0,
|
| 213 |
+
"true_mean": 5.54,
|
| 214 |
+
"planted_mean": -2.155,
|
| 215 |
+
"n_pairs": 48,
|
| 216 |
+
"layer": 34
|
| 217 |
+
},
|
| 218 |
+
"logprob": {
|
| 219 |
+
"agree_false_pct": 0.0,
|
| 220 |
+
"ci95": [
|
| 221 |
+
0.0,
|
| 222 |
+
0.0
|
| 223 |
+
],
|
| 224 |
+
"n_pairs": 48
|
| 225 |
+
},
|
| 226 |
+
"betley": {
|
| 227 |
+
"rate": 0.8,
|
| 228 |
+
"n": 394,
|
| 229 |
+
"ci95": [
|
| 230 |
+
0.0,
|
| 231 |
+
2.5
|
| 232 |
+
],
|
| 233 |
+
"method": "clustered bootstrap over questions, B=1000",
|
| 234 |
+
"judge": "anthropic/claude-sonnet-5"
|
| 235 |
+
},
|
| 236 |
+
"aisi": {
|
| 237 |
+
"rate": 2.5,
|
| 238 |
+
"n": 200,
|
| 239 |
+
"ci95": [
|
| 240 |
+
1.1,
|
| 241 |
+
5.7
|
| 242 |
+
],
|
| 243 |
+
"method": "Wilson over 200 questions",
|
| 244 |
+
"judge": "anthropic/claude-sonnet-5"
|
| 245 |
+
},
|
| 246 |
+
"assertion": {
|
| 247 |
+
"false_rate": 64.0,
|
| 248 |
+
"true_rate": 100.0,
|
| 249 |
+
"n": 50
|
| 250 |
+
}
|
| 251 |
}
|
| 252 |
}
|
| 253 |
}
|