Genteki commited on
Commit
f2259cc
·
verified ·
1 Parent(s): c5cb570

Upload dataset.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. dataset.json +442 -0
dataset.json ADDED
@@ -0,0 +1,442 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "name": "inspect_evals/piqa - Sample 1",
4
+ "description": "Task from inspect_evals/piqa dataset, sample 1",
5
+ "prompt": "The entire content of your response should be of the following format: 'ANSWER:\n$LETTER' (without quotes) where LETTER is one of A,B.\n\nGiven either a question or a statement followed by two possible solutions\nlabelled A and B, choose the most appropriate solution. If a question is given,\nthe solutions answer the question. If a statement is given, the solutions\nexplain how to achieve the statement.\n\nHow do I ready a guinea pig cage for it's new occupants?\n\nA) Provide the guinea pig with a cage full of a few inches of bedding made of ripped paper strips, you will also need to supply it with a water bottle and a food dish.\nB) Provide the guinea pig with a cage full of a few inches of bedding made of ripped jeans material, you will also need to supply it with a water bottle and a food dish.",
6
+ "mcp_config": {
7
+ "inspect": {
8
+ "command": "docker",
9
+ "args": [
10
+ "run",
11
+ "--rm",
12
+ "-i",
13
+ "--init",
14
+ "-v",
15
+ "/var/run/docker.sock:/var/run/docker.sock",
16
+ "-v",
17
+ "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
18
+ "hud-inspect:latest"
19
+ ]
20
+ }
21
+ },
22
+ "setup_tool": {
23
+ "name": "load_task",
24
+ "arguments": {
25
+ "task_name": "inspect_evals/piqa",
26
+ "sample_id": 1,
27
+ "model": "claude-sonnet-4-20250514"
28
+ }
29
+ },
30
+ "evaluate_tool": {
31
+ "name": "score_task",
32
+ "arguments": {}
33
+ },
34
+ "agent_config": {
35
+ "agent_class": "server.agents.InspectAgent",
36
+ "disallowed_tools": [
37
+ "_store_agent_message",
38
+ "load_task",
39
+ "score_task"
40
+ ],
41
+ "allowed_tools": [],
42
+ "initial_screenshot": false
43
+ },
44
+ "system_prompt": "You are a helpful AI assistant. Follow the instructions in the user messages carefully."
45
+ },
46
+ {
47
+ "name": "inspect_evals/piqa - Sample 2",
48
+ "description": "Task from inspect_evals/piqa dataset, sample 2",
49
+ "prompt": "The entire content of your response should be of the following format: 'ANSWER:\n$LETTER' (without quotes) where LETTER is one of A,B.\n\nGiven either a question or a statement followed by two possible solutions\nlabelled A and B, choose the most appropriate solution. If a question is given,\nthe solutions answer the question. If a statement is given, the solutions\nexplain how to achieve the statement.\n\ndresser\n\nA) replace drawer with bobby pin \nB) finish, woodgrain with bobby pin ",
50
+ "mcp_config": {
51
+ "inspect": {
52
+ "command": "docker",
53
+ "args": [
54
+ "run",
55
+ "--rm",
56
+ "-i",
57
+ "--init",
58
+ "-v",
59
+ "/var/run/docker.sock:/var/run/docker.sock",
60
+ "-v",
61
+ "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
62
+ "hud-inspect:latest"
63
+ ]
64
+ }
65
+ },
66
+ "setup_tool": {
67
+ "name": "load_task",
68
+ "arguments": {
69
+ "task_name": "inspect_evals/piqa",
70
+ "sample_id": 2,
71
+ "model": "claude-sonnet-4-20250514"
72
+ }
73
+ },
74
+ "evaluate_tool": {
75
+ "name": "score_task",
76
+ "arguments": {}
77
+ },
78
+ "agent_config": {
79
+ "agent_class": "server.agents.InspectAgent",
80
+ "disallowed_tools": [
81
+ "_store_agent_message",
82
+ "load_task",
83
+ "score_task"
84
+ ],
85
+ "allowed_tools": [],
86
+ "initial_screenshot": false
87
+ },
88
+ "system_prompt": "You are a helpful AI assistant. Follow the instructions in the user messages carefully."
89
+ },
90
+ {
91
+ "name": "inspect_evals/piqa - Sample 3",
92
+ "description": "Task from inspect_evals/piqa dataset, sample 3",
93
+ "prompt": "The entire content of your response should be of the following format: 'ANSWER:\n$LETTER' (without quotes) where LETTER is one of A,B.\n\nGiven either a question or a statement followed by two possible solutions\nlabelled A and B, choose the most appropriate solution. If a question is given,\nthe solutions answer the question. If a statement is given, the solutions\nexplain how to achieve the statement.\n\nTo fight Ivan Drago in Rocky for sega master system.\n\nA) Drago isn't in this game because it was released before Rocky IV.\nB) You have to defeat Apollo Creed and Clubber Lang first.",
94
+ "mcp_config": {
95
+ "inspect": {
96
+ "command": "docker",
97
+ "args": [
98
+ "run",
99
+ "--rm",
100
+ "-i",
101
+ "--init",
102
+ "-v",
103
+ "/var/run/docker.sock:/var/run/docker.sock",
104
+ "-v",
105
+ "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
106
+ "hud-inspect:latest"
107
+ ]
108
+ }
109
+ },
110
+ "setup_tool": {
111
+ "name": "load_task",
112
+ "arguments": {
113
+ "task_name": "inspect_evals/piqa",
114
+ "sample_id": 3,
115
+ "model": "claude-sonnet-4-20250514"
116
+ }
117
+ },
118
+ "evaluate_tool": {
119
+ "name": "score_task",
120
+ "arguments": {}
121
+ },
122
+ "agent_config": {
123
+ "agent_class": "server.agents.InspectAgent",
124
+ "disallowed_tools": [
125
+ "_store_agent_message",
126
+ "load_task",
127
+ "score_task"
128
+ ],
129
+ "allowed_tools": [],
130
+ "initial_screenshot": false
131
+ },
132
+ "system_prompt": "You are a helpful AI assistant. Follow the instructions in the user messages carefully."
133
+ },
134
+ {
135
+ "name": "inspect_evals/piqa - Sample 4",
136
+ "description": "Task from inspect_evals/piqa dataset, sample 4",
137
+ "prompt": "The entire content of your response should be of the following format: 'ANSWER:\n$LETTER' (without quotes) where LETTER is one of A,B.\n\nGiven either a question or a statement followed by two possible solutions\nlabelled A and B, choose the most appropriate solution. If a question is given,\nthe solutions answer the question. If a statement is given, the solutions\nexplain how to achieve the statement.\n\nMake outdoor pillow.\n\nA) Blow into tin can and tie with rubber band.\nB) Blow into trash bag and tie with rubber band.",
138
+ "mcp_config": {
139
+ "inspect": {
140
+ "command": "docker",
141
+ "args": [
142
+ "run",
143
+ "--rm",
144
+ "-i",
145
+ "--init",
146
+ "-v",
147
+ "/var/run/docker.sock:/var/run/docker.sock",
148
+ "-v",
149
+ "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
150
+ "hud-inspect:latest"
151
+ ]
152
+ }
153
+ },
154
+ "setup_tool": {
155
+ "name": "load_task",
156
+ "arguments": {
157
+ "task_name": "inspect_evals/piqa",
158
+ "sample_id": 4,
159
+ "model": "claude-sonnet-4-20250514"
160
+ }
161
+ },
162
+ "evaluate_tool": {
163
+ "name": "score_task",
164
+ "arguments": {}
165
+ },
166
+ "agent_config": {
167
+ "agent_class": "server.agents.InspectAgent",
168
+ "disallowed_tools": [
169
+ "_store_agent_message",
170
+ "load_task",
171
+ "score_task"
172
+ ],
173
+ "allowed_tools": [],
174
+ "initial_screenshot": false
175
+ },
176
+ "system_prompt": "You are a helpful AI assistant. Follow the instructions in the user messages carefully."
177
+ },
178
+ {
179
+ "name": "inspect_evals/piqa - Sample 5",
180
+ "description": "Task from inspect_evals/piqa dataset, sample 5",
181
+ "prompt": "The entire content of your response should be of the following format: 'ANSWER:\n$LETTER' (without quotes) where LETTER is one of A,B.\n\nGiven either a question or a statement followed by two possible solutions\nlabelled A and B, choose the most appropriate solution. If a question is given,\nthe solutions answer the question. If a statement is given, the solutions\nexplain how to achieve the statement.\n\nice box\n\nA) will turn into a cooler if you add water to it\nB) will turn into a cooler if you add soda to it",
182
+ "mcp_config": {
183
+ "inspect": {
184
+ "command": "docker",
185
+ "args": [
186
+ "run",
187
+ "--rm",
188
+ "-i",
189
+ "--init",
190
+ "-v",
191
+ "/var/run/docker.sock:/var/run/docker.sock",
192
+ "-v",
193
+ "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
194
+ "hud-inspect:latest"
195
+ ]
196
+ }
197
+ },
198
+ "setup_tool": {
199
+ "name": "load_task",
200
+ "arguments": {
201
+ "task_name": "inspect_evals/piqa",
202
+ "sample_id": 5,
203
+ "model": "claude-sonnet-4-20250514"
204
+ }
205
+ },
206
+ "evaluate_tool": {
207
+ "name": "score_task",
208
+ "arguments": {}
209
+ },
210
+ "agent_config": {
211
+ "agent_class": "server.agents.InspectAgent",
212
+ "disallowed_tools": [
213
+ "_store_agent_message",
214
+ "load_task",
215
+ "score_task"
216
+ ],
217
+ "allowed_tools": [],
218
+ "initial_screenshot": false
219
+ },
220
+ "system_prompt": "You are a helpful AI assistant. Follow the instructions in the user messages carefully."
221
+ },
222
+ {
223
+ "name": "inspect_evals/piqa - Sample 6",
224
+ "description": "Task from inspect_evals/piqa dataset, sample 6",
225
+ "prompt": "The entire content of your response should be of the following format: 'ANSWER:\n$LETTER' (without quotes) where LETTER is one of A,B.\n\nGiven either a question or a statement followed by two possible solutions\nlabelled A and B, choose the most appropriate solution. If a question is given,\nthe solutions answer the question. If a statement is given, the solutions\nexplain how to achieve the statement.\n\nRemove soap scum from shower door.\n\nA) Rub hard with bed sheets, then rinse.\nB) Rub hard with dryer sheets, then rinse.",
226
+ "mcp_config": {
227
+ "inspect": {
228
+ "command": "docker",
229
+ "args": [
230
+ "run",
231
+ "--rm",
232
+ "-i",
233
+ "--init",
234
+ "-v",
235
+ "/var/run/docker.sock:/var/run/docker.sock",
236
+ "-v",
237
+ "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
238
+ "hud-inspect:latest"
239
+ ]
240
+ }
241
+ },
242
+ "setup_tool": {
243
+ "name": "load_task",
244
+ "arguments": {
245
+ "task_name": "inspect_evals/piqa",
246
+ "sample_id": 6,
247
+ "model": "claude-sonnet-4-20250514"
248
+ }
249
+ },
250
+ "evaluate_tool": {
251
+ "name": "score_task",
252
+ "arguments": {}
253
+ },
254
+ "agent_config": {
255
+ "agent_class": "server.agents.InspectAgent",
256
+ "disallowed_tools": [
257
+ "_store_agent_message",
258
+ "load_task",
259
+ "score_task"
260
+ ],
261
+ "allowed_tools": [],
262
+ "initial_screenshot": false
263
+ },
264
+ "system_prompt": "You are a helpful AI assistant. Follow the instructions in the user messages carefully."
265
+ },
266
+ {
267
+ "name": "inspect_evals/piqa - Sample 7",
268
+ "description": "Task from inspect_evals/piqa dataset, sample 7",
269
+ "prompt": "The entire content of your response should be of the following format: 'ANSWER:\n$LETTER' (without quotes) where LETTER is one of A,B.\n\nGiven either a question or a statement followed by two possible solutions\nlabelled A and B, choose the most appropriate solution. If a question is given,\nthe solutions answer the question. If a statement is given, the solutions\nexplain how to achieve the statement.\n\nRecycle a spray bottle for a new cleaner.\n\nA) Open the top of the empty spray bottle and check for damage. Rinse the bottle then fill halfway with warm water. Replace the spray nozzle and pump a few times to clear the hose. Empty sparrow bottle and allow to dry before using with a new cleaner.\nB) Open the top of the empty spray bottle and check for damage. Rinse the bottle then fill halfway with warm water. Replace the spray nozzle and pump a few times to clear the hose. Empty spray bottle and allow to dry before using with a new cleaner.",
270
+ "mcp_config": {
271
+ "inspect": {
272
+ "command": "docker",
273
+ "args": [
274
+ "run",
275
+ "--rm",
276
+ "-i",
277
+ "--init",
278
+ "-v",
279
+ "/var/run/docker.sock:/var/run/docker.sock",
280
+ "-v",
281
+ "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
282
+ "hud-inspect:latest"
283
+ ]
284
+ }
285
+ },
286
+ "setup_tool": {
287
+ "name": "load_task",
288
+ "arguments": {
289
+ "task_name": "inspect_evals/piqa",
290
+ "sample_id": 7,
291
+ "model": "claude-sonnet-4-20250514"
292
+ }
293
+ },
294
+ "evaluate_tool": {
295
+ "name": "score_task",
296
+ "arguments": {}
297
+ },
298
+ "agent_config": {
299
+ "agent_class": "server.agents.InspectAgent",
300
+ "disallowed_tools": [
301
+ "_store_agent_message",
302
+ "load_task",
303
+ "score_task"
304
+ ],
305
+ "allowed_tools": [],
306
+ "initial_screenshot": false
307
+ },
308
+ "system_prompt": "You are a helpful AI assistant. Follow the instructions in the user messages carefully."
309
+ },
310
+ {
311
+ "name": "inspect_evals/piqa - Sample 8",
312
+ "description": "Task from inspect_evals/piqa dataset, sample 8",
313
+ "prompt": "The entire content of your response should be of the following format: 'ANSWER:\n$LETTER' (without quotes) where LETTER is one of A,B.\n\nGiven either a question or a statement followed by two possible solutions\nlabelled A and B, choose the most appropriate solution. If a question is given,\nthe solutions answer the question. If a statement is given, the solutions\nexplain how to achieve the statement.\n\nHow do you attach toilet paper to a glass jar?\n\nA) Press a piece of double-sided tape to the glass jar and then press the toilet paper onto the tape.\nB) Spread mayonnaise all over the jar with your palms and then roll the jar in toilet paper.",
314
+ "mcp_config": {
315
+ "inspect": {
316
+ "command": "docker",
317
+ "args": [
318
+ "run",
319
+ "--rm",
320
+ "-i",
321
+ "--init",
322
+ "-v",
323
+ "/var/run/docker.sock:/var/run/docker.sock",
324
+ "-v",
325
+ "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
326
+ "hud-inspect:latest"
327
+ ]
328
+ }
329
+ },
330
+ "setup_tool": {
331
+ "name": "load_task",
332
+ "arguments": {
333
+ "task_name": "inspect_evals/piqa",
334
+ "sample_id": 8,
335
+ "model": "claude-sonnet-4-20250514"
336
+ }
337
+ },
338
+ "evaluate_tool": {
339
+ "name": "score_task",
340
+ "arguments": {}
341
+ },
342
+ "agent_config": {
343
+ "agent_class": "server.agents.InspectAgent",
344
+ "disallowed_tools": [
345
+ "_store_agent_message",
346
+ "load_task",
347
+ "score_task"
348
+ ],
349
+ "allowed_tools": [],
350
+ "initial_screenshot": false
351
+ },
352
+ "system_prompt": "You are a helpful AI assistant. Follow the instructions in the user messages carefully."
353
+ },
354
+ {
355
+ "name": "inspect_evals/piqa - Sample 9",
356
+ "description": "Task from inspect_evals/piqa dataset, sample 9",
357
+ "prompt": "The entire content of your response should be of the following format: 'ANSWER:\n$LETTER' (without quotes) where LETTER is one of A,B.\n\nGiven either a question or a statement followed by two possible solutions\nlabelled A and B, choose the most appropriate solution. If a question is given,\nthe solutions answer the question. If a statement is given, the solutions\nexplain how to achieve the statement.\n\nHow to make tissue paper window decorations?\n\nA) Find tissue paper and fold it in half. Take scissors and cut out pieces of the paper in the middle. When you are done tape it to your window.\nB) Find tissue paper and fold it in half. Take scissors and tear out pieces of the paper in the middle. When you are done tape it to your window.",
358
+ "mcp_config": {
359
+ "inspect": {
360
+ "command": "docker",
361
+ "args": [
362
+ "run",
363
+ "--rm",
364
+ "-i",
365
+ "--init",
366
+ "-v",
367
+ "/var/run/docker.sock:/var/run/docker.sock",
368
+ "-v",
369
+ "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
370
+ "hud-inspect:latest"
371
+ ]
372
+ }
373
+ },
374
+ "setup_tool": {
375
+ "name": "load_task",
376
+ "arguments": {
377
+ "task_name": "inspect_evals/piqa",
378
+ "sample_id": 9,
379
+ "model": "claude-sonnet-4-20250514"
380
+ }
381
+ },
382
+ "evaluate_tool": {
383
+ "name": "score_task",
384
+ "arguments": {}
385
+ },
386
+ "agent_config": {
387
+ "agent_class": "server.agents.InspectAgent",
388
+ "disallowed_tools": [
389
+ "_store_agent_message",
390
+ "load_task",
391
+ "score_task"
392
+ ],
393
+ "allowed_tools": [],
394
+ "initial_screenshot": false
395
+ },
396
+ "system_prompt": "You are a helpful AI assistant. Follow the instructions in the user messages carefully."
397
+ },
398
+ {
399
+ "name": "inspect_evals/piqa - Sample 10",
400
+ "description": "Task from inspect_evals/piqa dataset, sample 10",
401
+ "prompt": "The entire content of your response should be of the following format: 'ANSWER:\n$LETTER' (without quotes) where LETTER is one of A,B.\n\nGiven either a question or a statement followed by two possible solutions\nlabelled A and B, choose the most appropriate solution. If a question is given,\nthe solutions answer the question. If a statement is given, the solutions\nexplain how to achieve the statement.\n\nNeatly wrap up an extension cord.\n\nA) Wrap the cord around your hand and elbow.\nB) Wrap the cord around your hand and knee.",
402
+ "mcp_config": {
403
+ "inspect": {
404
+ "command": "docker",
405
+ "args": [
406
+ "run",
407
+ "--rm",
408
+ "-i",
409
+ "--init",
410
+ "-v",
411
+ "/var/run/docker.sock:/var/run/docker.sock",
412
+ "-v",
413
+ "/home/genteki/.cache/inspect_evals:/root/.cache/inspect_evals",
414
+ "hud-inspect:latest"
415
+ ]
416
+ }
417
+ },
418
+ "setup_tool": {
419
+ "name": "load_task",
420
+ "arguments": {
421
+ "task_name": "inspect_evals/piqa",
422
+ "sample_id": 10,
423
+ "model": "claude-sonnet-4-20250514"
424
+ }
425
+ },
426
+ "evaluate_tool": {
427
+ "name": "score_task",
428
+ "arguments": {}
429
+ },
430
+ "agent_config": {
431
+ "agent_class": "server.agents.InspectAgent",
432
+ "disallowed_tools": [
433
+ "_store_agent_message",
434
+ "load_task",
435
+ "score_task"
436
+ ],
437
+ "allowed_tools": [],
438
+ "initial_screenshot": false
439
+ },
440
+ "system_prompt": "You are a helpful AI assistant. Follow the instructions in the user messages carefully."
441
+ }
442
+ ]