Hugging Face's logo

botp
/
Solomon

Text Classification
PEFT
lora
document-question-answering
structured-decisions
calibration
synthetic-evaluation
Model card Files
xet
 Community
Solomon
File size: 3,106 Bytes
1d2de8a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
{
  "arch": "scope10-evidence-mlp-v1",
  "hidden_size": 5120,
  "width": 512,
  "input": "concat(q, u, q*u), LayerNorm(3H, elementwise_affine=False)",
  "activation": "gelu",
  "hidden_layer": "lm.norm(output of language_model.layers[42]) (the frozen final norm applied to a mid-layer tap)",
  "branch_state": "last position of question branch, adapter active (question placement)",
  "unit_pooling": "mean over document-prefill tokens overlapping the unit char range (adapter off)",
  "output": "logit per unit; probability = sigmoid(logit)",
  "dtype": "float32",
  "variant": "refit-mlp-mid+lex",
  "layer": "mid",
  "layer_index": 42,
  "lexical_residual": {
    "alpha": 9.770263671875,
    "rule": "logit = head_logit + alpha * lexical; probability = sigmoid(logit)",
    "lexical": "solomon.evidence_selector.lexical: question words minus stop words, document-IDF weighted overlap per sentence unit, divided by the row max",
    "question": "solomon.evidence_selector.question_of(the answer branch prompt)",
    "warning": "solomon.evidence_head.load alone returns head_logit WITHOUT the lexical term; the serving selector adds it"
  },
  "refit": {
    "dev": {
      "gold_rows": 753,
      "hit@1": 0.5285524568393094,
      "hit@3": 0.8167330677290837,
      "hit@5": 0.8539176626826029,
      "recall@5": 0.8253652058432935
    },
    "thresholded_dev": {
      "precision": 0.0,
      "recall": 0.0,
      "f1": 0.0
    },
    "thresholds": {
      "policy": "scope10-evidence-thresholds-v1",
      "min_no_support_accuracy": 0.95,
      "rule": "unit kept iff max(unit probs) >= absent and prob >= select (not used in serving: v1.1 serves ranked pointers, see evidence-policy.json)",
      "tasks": {
        "boolean": {
          "select": 0.0,
          "absent": 1.0,
          "constraint_met": true,
          "rows": 1375,
          "precision": 0.0,
          "recall": 0.0,
          "f1": 0.0,
          "tp": 0,
          "fp": 0,
          "fn": 291,
          "no_support_rows": 1122,
          "no_support_accuracy": 1.0
        },
        "multilabel": {
          "select": 0.0,
          "absent": 1.0,
          "constraint_met": true,
          "rows": 931,
          "precision": 0.0,
          "recall": 0.0,
          "f1": 0.0,
          "tp": 0,
          "fp": 0,
          "fn": 385,
          "no_support_rows": 569,
          "no_support_accuracy": 1.0
        },
        "ordered": {
          "select": 0.0,
          "absent": 1.0,
          "constraint_met": true,
          "rows": 134,
          "precision": 0.0,
          "recall": 0.0,
          "f1": 0.0,
          "tp": 0,
          "fp": 0,
          "fn": 120,
          "no_support_rows": 36,
          "no_support_accuracy": 1.0
        },
        "single": {
          "select": 0.0,
          "absent": 1.0,
          "constraint_met": true,
          "rows": 134,
          "precision": 0.0,
          "recall": 0.0,
          "f1": 0.0,
          "tp": 0,
          "fp": 0,
          "fn": 48,
          "no_support_rows": 94,
          "no_support_accuracy": 1.0
        }
      }
    }
  }
}