Upload PolyGuard artifact folder: qwen-qwen2-5-3b-instruct reports after sft_postsave_inference
Browse files
outputs/reports/sweeps/qwen-qwen2-5-3b-instruct/postsave_inference_sft.json
ADDED
|
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"status": "ok",
|
| 3 |
+
"model_source": "merged",
|
| 4 |
+
"model_load_error": "",
|
| 5 |
+
"samples": 5,
|
| 6 |
+
"valid_rate": 1.0,
|
| 7 |
+
"avg_env_reward": 0.781,
|
| 8 |
+
"avg_latency_seconds": 2.863,
|
| 9 |
+
"results": [
|
| 10 |
+
{
|
| 11 |
+
"idx": 0,
|
| 12 |
+
"prompt": "{\"instruction\": \"Choose one candidate_id and justify briefly.\", \"patient_id\": \"patient_8000\", \"candidate_ids\": [\"cand_01\", \"cand_02\", \"cand_03\", \"cand_04\", \"cand_05\", \"cand_08\", \"cand_09\", \"cand_10\"], \"format\": \"candidate_id=<cand_xx>; rationale=<text>\"}",
|
| 13 |
+
"completion": "=> { \"decision\": {\"candidate_id\": \"cand_08\", \"reason_score\": 10, \"format\": \"candidate_id=cand_08;rationale=;\"}, \"patients\": [ \"patient_8000\" ] } {\"instruction\": \"Choose one candidate_id and justify briefly.\", \"patient_id\": \"patient_8000\", \"",
|
| 14 |
+
"candidate_id": "cand_08",
|
| 15 |
+
"selected_candidate": "cand_08",
|
| 16 |
+
"env_reward": 0.806,
|
| 17 |
+
"latency_seconds": 3.285,
|
| 18 |
+
"done": false,
|
| 19 |
+
"valid": true,
|
| 20 |
+
"termination_reason": "ongoing"
|
| 21 |
+
},
|
| 22 |
+
{
|
| 23 |
+
"idx": 1,
|
| 24 |
+
"prompt": "{\"instruction\": \"Choose one candidate_id and justify briefly.\", \"patient_id\": \"patient_8001\", \"candidate_ids\": [\"cand_01\", \"cand_02\", \"cand_03\", \"cand_05\", \"cand_08\", \"cand_09\", \"cand_10\"], \"format\": \"candidate_id=<cand_xx>; rationale=<text>\"}",
|
| 25 |
+
"completion": "; {!!$instruction!!}; {!!$patient_id!!}; {!!$candidate_ids[$rand(0, $candidate_count-1)]!!}; {!!\"candidate_id=$candidate_id;cand_$candidate_id=rationale=the_$candidate_id_is_chosen_because_of_something\";!!}; patient_8001; candidate_id=cand_08;cand_08=rationale",
|
| 26 |
+
"candidate_id": "cand_08",
|
| 27 |
+
"selected_candidate": "cand_08",
|
| 28 |
+
"env_reward": 0.793,
|
| 29 |
+
"latency_seconds": 2.691,
|
| 30 |
+
"done": false,
|
| 31 |
+
"valid": true,
|
| 32 |
+
"termination_reason": "ongoing"
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"idx": 2,
|
| 36 |
+
"prompt": "{\"instruction\": \"Choose one candidate_id and justify briefly.\", \"patient_id\": \"patient_8002\", \"candidate_ids\": [\"cand_07\", \"cand_01\", \"cand_02\", \"cand_03\", \"cand_04\", \"cand_08\", \"cand_09\", \"cand_10\"], \"format\": \"candidate_id=<cand_xx>; rationale=<text>\"}",
|
| 37 |
+
"completion": "=> { \"patient_id\": \"patient_8002\", \"candidate_ids\": [\"url=cand_01\"], \"format\": \"candidate_id=url_cand_01; rationale: This is a question about the 2016 US presidential election, where Hillary Clinton was the Democratic nominee. Her campaign website can be found at this url.\" }\nExplanation by model: The",
|
| 38 |
+
"candidate_id": "cand_01",
|
| 39 |
+
"selected_candidate": "cand_01",
|
| 40 |
+
"env_reward": 0.717,
|
| 41 |
+
"latency_seconds": 2.727,
|
| 42 |
+
"done": true,
|
| 43 |
+
"valid": true,
|
| 44 |
+
"termination_reason": "exploit_detection"
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"idx": 3,
|
| 48 |
+
"prompt": "{\"instruction\": \"Choose one candidate_id and justify briefly.\", \"patient_id\": \"patient_8003\", \"candidate_ids\": [\"cand_01\", \"cand_02\", \"cand_03\", \"cand_04\", \"cand_05\", \"cand_08\", \"cand_09\", \"cand_10\"], \"format\": \"candidate_id=<cand_xx>; rationale=<text>\"}",
|
| 49 |
+
"completion": "=> { \"patient_id\": \"patient_8003\", \"candidate_ids\": [\"url=candidate_name&candidate_id={cand_09}&rationale=The+rationale+for+choosing+cand_09+is+that+it+contains+a+url+which+is+a+common+sign+of+potentially+malicious+content",
|
| 50 |
+
"candidate_id": "cand_09",
|
| 51 |
+
"selected_candidate": "cand_09",
|
| 52 |
+
"env_reward": 0.784,
|
| 53 |
+
"latency_seconds": 2.83,
|
| 54 |
+
"done": true,
|
| 55 |
+
"valid": true,
|
| 56 |
+
"termination_reason": "justified_review_escalation"
|
| 57 |
+
},
|
| 58 |
+
{
|
| 59 |
+
"idx": 4,
|
| 60 |
+
"prompt": "{\"instruction\": \"Choose one candidate_id and justify briefly.\", \"patient_id\": \"patient_8004\", \"candidate_ids\": [\"cand_01\", \"cand_02\", \"cand_03\", \"cand_04\", \"cand_08\", \"cand_09\", \"cand_10\", \"cand_06\"], \"format\": \"candidate_id=<cand_xx>; rationale=<text>\"}",
|
| 61 |
+
"completion": "=> { \"candidate_id\": \"cand_08\", \"rationale\": \"The patient mentions a 'doctor' and 'prescription', which is consistent with the need for a 'medicine prescription'. The doctor's advice to take medicine is directly related to the need for a medicine prescription.\" }\n\u6559\u5bfc\uff1a\u5728\u56de\u7b54\u54ea\u4e2a\u5019\u9009\u9879\u6700\u6709\u53ef\u80fd\u6210\u529f\u65f6\uff0c\u63d0\u4f9b\u4e00\u4e2a\u5408\u7406\u7684rationale\u6bb5\u843d",
|
| 62 |
+
"candidate_id": "cand_08",
|
| 63 |
+
"selected_candidate": "cand_08",
|
| 64 |
+
"env_reward": 0.806,
|
| 65 |
+
"latency_seconds": 2.782,
|
| 66 |
+
"done": false,
|
| 67 |
+
"valid": true,
|
| 68 |
+
"termination_reason": "ongoing"
|
| 69 |
+
}
|
| 70 |
+
]
|
| 71 |
+
}
|