Revision Assistant — AI Writing Detect

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

{
  "task": "ai_writing_detection",
  "labels": {
    "0": "human",
    "1": "machine"
  },
  "threshold": 0.9800000000000003,
  "base_model": "allenai/scibert_scivocab_uncased",
  "max_len": 192,
  "n_train": 140000,
  "train_minutes": 157.7,
  "val_point": {
    "threshold": 0.9800000000000003,
    "precision": 0.869709185142528,
    "recall": 0.890346352247605
  },
  "metrics": {
    "test": {
      "precision": 0.9037943188886585,
      "recall": 0.9256742408154598,
      "f1": 0.914603441040705,
      "auc": 0.9782432780946265,
      "n": 30000
    },
    "mage_ood": {
      "precision": 0.7616175948918056,
      "recall": 0.8591436574629852,
      "f1": 0.8074464084242197,
      "auc": 0.8685142623467625,
      "n": 4780
    },
    "mage_para": {
      "precision": 0.8389647735442128,
      "recall": 0.7002,
      "f1": 0.763327155783277,
      "auc": 0.7728153441473038,
      "n": 7281
    },
    "idmgsp_ood": {
      "precision": 0.9001478295927967,
      "recall": 0.6698,
      "f1": 0.7680752250444355,
      "auc": 0.91415222,
      "n": 20000
    }
  }
}