Revision Assistant — Journal Readiness

Quantized ONNX + tokenizer for browser / Transformers.js inference in Revision Assistant.

Files

• onnx/model_quantized.onnx — int8 quantized weights (preferred for browser)
• tokenizer + config.json / ort_config.json / inference_config.json

Metrics / inference config

{
  "threshold": 0.5,
  "thresholds": {
    "structure_ok": 0.5,
    "numerical_clear": 0.5,
    "novelty_ok": 0.5,
    "methods_concrete": 0.5,
    "selective_ready": 0.55,
    "ieee_craft": 0.5
  },
  "labels": [
    "structure_ok",
    "numerical_clear",
    "novelty_ok",
    "methods_concrete",
    "selective_ready",
    "ieee_craft"
  ],
  "base_model": "allenai/scibert_scivocab_uncased",
  "problem_type": "multi_label_classification",
  "test_macro_f1": 0.920882611076118,
  "test_micro_f1": 0.9814847678201716,
  "test_mean_auc": 0.8737736064080326,
  "train_size": 20000,
  "epochs": 2,
  "wall_seconds": 2563,
  "disclaimer": "Heuristic readiness signal heads from open arXiv-derived text. Not acceptance probability; not official Q1/Q2; not IEEE-affiliated."
}