Update eval results with empirical data from trained models
Browse files- eval_results.json +119 -0
eval_results.json
ADDED
|
@@ -0,0 +1,119 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"metrics": {
|
| 3 |
+
"A": {
|
| 4 |
+
"config": "A",
|
| 5 |
+
"accuracy": 0.16,
|
| 6 |
+
"avg_cost": 1.0,
|
| 7 |
+
"safety": 0.975,
|
| 8 |
+
"n": 200,
|
| 9 |
+
"by_action": {
|
| 10 |
+
"tool_call": 1.0,
|
| 11 |
+
"retrieval": 0.037,
|
| 12 |
+
"file_read": 0.0,
|
| 13 |
+
"file_write": 0.0,
|
| 14 |
+
"repair": 0.0,
|
| 15 |
+
"verifier": 0.0,
|
| 16 |
+
"ask_clarification": 0.0,
|
| 17 |
+
"final_answer": 0.0,
|
| 18 |
+
"BLOCKED": 0.0
|
| 19 |
+
}
|
| 20 |
+
},
|
| 21 |
+
"B": {
|
| 22 |
+
"config": "B",
|
| 23 |
+
"accuracy": 0.19,
|
| 24 |
+
"avg_cost": 0.15,
|
| 25 |
+
"safety": 0.975,
|
| 26 |
+
"n": 200,
|
| 27 |
+
"by_action": {
|
| 28 |
+
"tool_call": 0.871,
|
| 29 |
+
"retrieval": 0.0,
|
| 30 |
+
"file_read": 0.037,
|
| 31 |
+
"file_write": 0.0,
|
| 32 |
+
"repair": 0.0,
|
| 33 |
+
"verifier": 0.0,
|
| 34 |
+
"ask_clarification": 0.5,
|
| 35 |
+
"final_answer": 0.0,
|
| 36 |
+
"BLOCKED": 0.0
|
| 37 |
+
}
|
| 38 |
+
},
|
| 39 |
+
"C": {
|
| 40 |
+
"config": "C",
|
| 41 |
+
"accuracy": 0.19,
|
| 42 |
+
"avg_cost": 0.285,
|
| 43 |
+
"safety": 0.975,
|
| 44 |
+
"n": 200,
|
| 45 |
+
"by_action": {
|
| 46 |
+
"tool_call": 0.871,
|
| 47 |
+
"retrieval": 0.0,
|
| 48 |
+
"file_read": 0.037,
|
| 49 |
+
"file_write": 0.0,
|
| 50 |
+
"repair": 0.0,
|
| 51 |
+
"verifier": 0.0,
|
| 52 |
+
"ask_clarification": 0.5,
|
| 53 |
+
"final_answer": 0.0,
|
| 54 |
+
"BLOCKED": 0.0
|
| 55 |
+
},
|
| 56 |
+
"accept_rate": 0.965
|
| 57 |
+
},
|
| 58 |
+
"D": {
|
| 59 |
+
"config": "D",
|
| 60 |
+
"accuracy": 0.19,
|
| 61 |
+
"avg_cost": 0.25,
|
| 62 |
+
"safety": 0.975,
|
| 63 |
+
"n": 200,
|
| 64 |
+
"by_action": {
|
| 65 |
+
"tool_call": 0.871,
|
| 66 |
+
"retrieval": 0.0,
|
| 67 |
+
"file_read": 0.037,
|
| 68 |
+
"file_write": 0.0,
|
| 69 |
+
"repair": 0.0,
|
| 70 |
+
"verifier": 0.0,
|
| 71 |
+
"ask_clarification": 0.5,
|
| 72 |
+
"final_answer": 0.0,
|
| 73 |
+
"BLOCKED": 0.0
|
| 74 |
+
},
|
| 75 |
+
"accept_rate": 0.0,
|
| 76 |
+
"mean_score": -2.638,
|
| 77 |
+
"min_score": -4.656,
|
| 78 |
+
"max_score": -0.875
|
| 79 |
+
},
|
| 80 |
+
"E": {
|
| 81 |
+
"config": "E",
|
| 82 |
+
"accuracy": 0.17,
|
| 83 |
+
"avg_cost": 0.75,
|
| 84 |
+
"safety": 0.975,
|
| 85 |
+
"n": 200,
|
| 86 |
+
"by_action": {
|
| 87 |
+
"tool_call": 0.806,
|
| 88 |
+
"retrieval": 0.0,
|
| 89 |
+
"file_read": 0.037,
|
| 90 |
+
"file_write": 0.0,
|
| 91 |
+
"repair": 0.0,
|
| 92 |
+
"verifier": 0.0,
|
| 93 |
+
"ask_clarification": 0.4,
|
| 94 |
+
"final_answer": 0.0,
|
| 95 |
+
"BLOCKED": 0.0
|
| 96 |
+
},
|
| 97 |
+
"accept_rate": 1.0
|
| 98 |
+
}
|
| 99 |
+
},
|
| 100 |
+
"config": {
|
| 101 |
+
"cheap_model": "narcolepticchicken/speculative-proposer-qwen3-1.7b",
|
| 102 |
+
"verifier_model": "narcolepticchicken/speculative-verifier-qwen3-4b",
|
| 103 |
+
"strong_model": "Qwen/Qwen3-8B",
|
| 104 |
+
"eval_dataset": "narcolepticchicken/speculative-actions-eval",
|
| 105 |
+
"n_examples": 200,
|
| 106 |
+
"reward_threshold": 0.0
|
| 107 |
+
},
|
| 108 |
+
"action_distribution": {
|
| 109 |
+
"ask_clarification": 20,
|
| 110 |
+
"final_answer": 48,
|
| 111 |
+
"file_read": 27,
|
| 112 |
+
"tool_call": 31,
|
| 113 |
+
"repair": 10,
|
| 114 |
+
"retrieval": 27,
|
| 115 |
+
"file_write": 18,
|
| 116 |
+
"verifier": 14,
|
| 117 |
+
"BLOCKED": 5
|
| 118 |
+
}
|
| 119 |
+
}
|