Hugging Face's logo

mkzero
/
FastDecider-149M

Text Classification
Transformers
Safetensors
English
Chinese
modernbert
reranker
cross-encoder
agent
decision-making
zero-shot-classification
fast-decider
text-embeddings-inference
Model card Files
xet
 Community
FastDecider-149M
File size: 2,167 Bytes
70134a3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
{
  "model_name": "FastDecider-149M",
  "version": "1.3.0",
  "archive_date": "2026-09-22",
  "base_model": "Alibaba-DAMO/gte-reranker-modernbert-base",
  "base_model_architecture": "ModernBERT-base Cross-Encoder (SequenceClassification)",
  "parameter_count": 149079553,
  "parameter_size_mb": "149M",
  "num_layers": 22,
  "hidden_size": 768,
  "max_position_embeddings": 8192,
  "training_details": {
    "dataset": "/mnt/workspace/data/grand_unified_train_35k.jsonl",
    "total_samples": 30456,
    "trainer_script": "/mnt/workspace/arena/train_grand_unified_35k.py",
    "training_steps": 600,
    "optimizer": "AdamW (lr=2e-5, warmup=50, cosine schedule)",
    "loss_function": "Cross-Entropy over candidate option logits",
    "precision": "bfloat16"
  },
  "benchmarks": {
    "jevbench_public_easy": {
      "accuracy": "48/48 (100.00%)",
      "extraction": "12/12 (100.0%)",
      "fact": "12/12 (100.0%)",
      "intent": "12/12 (100.0%)",
      "tool_selection": "12/12 (100.0%)"
    },
    "jevbench_public_hard": {
      "accuracy": "49/111 (44.14%)",
      "routing_hard": "5/5 (100.0%)",
      "tradeoff": "6/6 (100.0%)",
      "trap": "5/8 (62.5%)",
      "ambiguous": "4/7 (57.1%)",
      "adversarial": "3/6 (50.0%)",
      "probability": "5/10 (50.0%)",
      "judge_hard": "7/17 (41.2%)",
      "temporal_numeric": "5/15 (33.3%)",
      "long_policy": "6/19 (31.6%)",
      "multi_hop": "3/18 (16.7%)"
    },
    "jevbench_public_original": {
      "accuracy": "49/72 (68.06%)",
      "intelligence_score": 65.35
    },
    "six_business_pillars": {
      "browser_dom_control": "50/50 (100.0%)",
      "api_dispatch_routing": "49/50 (98.0%)",
      "ecom_brand_extraction": "50/50 (100.0%)",
      "ecom_specs_matching": "50/50 (100.0%)",
      "ecom_category_prediction": "47/50 (94.0%)",
      "legal_contract_matching": "48/50 (96.0%)"
    },
    "latency_and_cost": {
      "p50_latency_ms": 14.93,
      "p99_latency_ms": 17.8,
      "serving_cost": "$0.0012 per 1k decisions",
      "throughput_qps": 84.2
    }
  },
  "reproducibility": "All scores evaluated via standard argmax over cross-encoder logits without heuristic overrides."
}