Hugging Face's logo

apus-ailab
/
APUS-Reflex

Transformers
Safetensors
English
Chinese
decision-model
dynamic-depth
structured-output
Model card Files
xet
 Community
7
APUS-Reflex
File size: 4,501 Bytes
68e5880
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
{
  "title": "APUS-OpenJev-v1",
  "created_date": "2026-09-21",
  "panel": {
    "records": 80,
    "parent_groups": 79,
    "task_counts": {
      "Browser": 16,
      "HelpSteer3": 16,
      "BoolQ": 16,
      "MNLI": 16,
      "Score": 16
    },
    "score_labels": "16 No, 0 Yes",
    "split": "validation/development regression; not sealed final",
    "laya_export_sha256": "26aa47e8f56c18a403eb4c8379d5cccb9cea5af97c191ab26ec36b8fb17ae923"
  },
  "accuracy": [
    {
      "model": "APUS-OpenJev 9B",
      "correct": 68,
      "total": 80,
      "percentage": 85.0,
      "source": "merged9b",
      "effort": "high",
      "variant_directory": "9B-3000"
    },
    {
      "model": "APUS-OpenJev 4B",
      "correct": 66,
      "total": 80,
      "percentage": 82.5,
      "source": "merged4b",
      "effort": "high",
      "variant_directory": "4B-5949"
    },
    {
      "model": "Jev official API",
      "correct": 66,
      "total": 80,
      "percentage": 82.5,
      "source": "native9b_latency"
    },
    {
      "model": "Laya typed / full input",
      "correct": 55,
      "total": 80,
      "percentage": 68.75,
      "source": "laya_repeats",
      "config": "typed-fullinput",
      "repetitions": 3,
      "identical_accuracy": true,
      "truncations": 0
    }
  ],
  "latency": {
    "native_forward_historical": [
      {
        "model": "APUS-OpenJev 9B",
        "p50_ms": 77.76357303373516,
        "p95_ms": 505.1563191227615,
        "source": "native9b_latency"
      },
      {
        "model": "APUS-OpenJev 4B",
        "p50_ms": 85.43,
        "p95_ms": 406.3,
        "source": "native4b_latency"
      }
    ],
    "laya_local_system_one": {
      "model": "Laya typed / full input",
      "p50_ms": 8.684632601216435,
      "p95_ms": 37.65514073893428,
      "p50_range_ms": [
        8.429150097072124,
        8.730278117582202
      ],
      "p95_range_ms": [
        37.441988941282034,
        37.88864379748702
      ],
      "aggregation": "median of three per-run P50 and P95 values; not pooled or fastest-run selection",
      "scope": "native system_one includes tokenization, transfer, forward and structured response; excludes HTTP and loading",
      "source": "laya_repeats"
    },
    "official_http": {
      "model": "Jev official API",
      "p50_ms": 436.9597500190139,
      "p95_ms": 4874.612333020195,
      "scope": "remote full HTTP response including network/queue/service; not client-observable TTFT",
      "source": "native9b_latency"
    },
    "cross_scope_speedup_claim": false,
    "merged_package_latency_retest": false
  },
  "sources": {
    "laya_repeats": {
      "project_relative_path": "docs/feat-jev-combined/evidence/laya-repeated-20260921/summary.json",
      "sha256": "16a2d3a265909e40f65cdb9331417a6d46a1f2140759f94d2b183728d64825e6"
    },
    "native9b_latency": {
      "project_relative_path": "docs/feat-jev-combined/evidence/9b-checkpoints80/3000/report.json",
      "sha256": "f3b93d04ab8e99d3cf135d410b559e1c96e980fba43c59b71428fe8c815b9303"
    },
    "native4b_latency": {
      "project_relative_path": "docs/feat-jev-combined/official-typesafe-vs-final5949-80-20260920.md",
      "sha256": "5a7568a998d4be13555374399aa91e9b458d782336b14ab33603a2ebcabe99fc"
    },
    "merged4b": {
      "project_relative_path": "docs/feat-jev-combined/openjet-release-20260921/merged-release/evidence/4B/comparison.json",
      "sha256": "97cc94ddebb4f321d3d4ed29c89a0d8b1b3d6570e48bff427c635ba82f67862e"
    },
    "merged9b": {
      "project_relative_path": "docs/feat-jev-combined/openjet-release-20260921/merged-release/evidence/9B/comparison.json",
      "sha256": "d2ea8a7af10f09fdfde166384ffb18fb5a19ebf1a1281c58cb72806a45878c41"
    },
    "panel_description": {
      "project_relative_path": "docs/feat-jev-combined/openjet-release-20260921/eval-dataset/SPLIT_AND_EVALUATION.md",
      "sha256": "b14b8af51f0527443b369f6995d46b786207dbf663ef2ce5afc9cc0a02d31b8b"
    }
  },
  "limitations": [
    "80 development questions are reused and have influenced development/version selection.",
    "Two more correct answers than Jev does not establish significant superiority or noninferiority.",
    "Score is all No; Browser is static candidate selection, not website completion.",
    "Laya full-input formatting was changed without retraining; three repeats do not increase independent sample size.",
    "Historical native forward, local system_one and remote HTTP are different timing boundaries."
  ]
}