eval/results.json
2.9 KB · 142 lines · json Raw
1 {
2 "model": "rl-agent",
3 "questions_evaluated": {
4 "eval_in": 23024,
5 "eval_zs": 2400
6 },
7 "by_task_family": {
8 "eval_in": {
9 "conversation outcomes": {
10 "n": 3600,
11 "accuracy": 0.4817,
12 "ece": 0.0193,
13 "nll": 0.6933
14 },
15 "email triage": {
16 "n": 2691,
17 "accuracy": 0.7321,
18 "ece": 0.0172,
19 "nll": 0.5954
20 },
21 "emotion and tone": {
22 "n": 1825,
23 "accuracy": 0.9058,
24 "ece": 0.0183,
25 "nll": 0.2382
26 },
27 "inference and fact checking": {
28 "n": 3022,
29 "accuracy": 0.8832,
30 "ece": 0.054,
31 "nll": 0.3404
32 },
33 "instruction-following tasks": {
34 "n": 600,
35 "accuracy": 0.8783,
36 "ece": 0.0465,
37 "nll": 0.3021
38 },
39 "intent and routing": {
40 "n": 1475,
41 "accuracy": 0.9912,
42 "ece": 0.0085,
43 "nll": 0.1811
44 },
45 "moderation and safety": {
46 "n": 2708,
47 "accuracy": 0.9671,
48 "ece": 0.0613,
49 "nll": 0.1527
50 },
51 "reading comprehension": {
52 "n": 770,
53 "accuracy": 0.8468,
54 "ece": 0.0833,
55 "nll": 0.4086
56 },
57 "response quality scoring": {
58 "n": 3146,
59 "accuracy": 0.5814,
60 "ece": 0.023,
61 "nll": 1.0087
62 },
63 "robustness checks": {
64 "n": 744,
65 "accuracy": 0.8508,
66 "ece": 0.1085,
67 "nll": 1.0576
68 },
69 "search relevance": {
70 "n": 733,
71 "accuracy": 0.6276,
72 "ece": 0.066,
73 "nll": 0.7281
74 },
75 "sentiment and rating": {
76 "n": 961,
77 "accuracy": 0.4422,
78 "ece": 0.4384,
79 "nll": 3.545
80 },
81 "topic classification": {
82 "n": 749,
83 "accuracy": 0.9386,
84 "ece": 0.0285,
85 "nll": 0.1957
86 }
87 },
88 "eval_zs": {
89 "emotion and tone": {
90 "n": 600,
91 "accuracy": 0.5833,
92 "ece": 0.3178,
93 "nll": 1.9761
94 },
95 "instruction-following tasks": {
96 "n": 600,
97 "accuracy": 0.8633,
98 "ece": 0.0455,
99 "nll": 0.3187
100 },
101 "moderation and safety": {
102 "n": 600,
103 "accuracy": 0.7967,
104 "ece": 0.1713,
105 "nll": 1.4151
106 },
107 "sentiment and rating": {
108 "n": 600,
109 "accuracy": 0.3617,
110 "ece": 0.2915,
111 "nll": 1.7985
112 }
113 }
114 },
115 "calibration_temperature": [
116 1.6369030475616455,
117 1.2514300346374512,
118 1.983399510383606
119 ],
120 "latency_ms": {
121 "1_questions": {
122 "p50_ms": 38.4,
123 "p95_ms": 42.1
124 },
125 "10_questions": {
126 "p50_ms": 156.0,
127 "p95_ms": 158.4
128 },
129 "50_questions": {
130 "p50_ms": 721.4,
131 "p95_ms": 733.0
132 }
133 },
134 "act_policy": {
135 "eval_in": {
136 "automation_rate": 1.0,
137 "accuracy_when_acting": 0.8032331136738056,
138 "accuracy_when_escalating": null,
139 "accuracy_all": 0.8032331136738056
140 }
141 }
142 }