model.4bit-palettized.mlpackage is fixed-shape batch 1, sequence length 256.
Inputs are int32: input_ids, attention_mask, and token_type_ids.
The verified CoreML graph uses an all-ones attention mask. Fill attention_mask with 1 for every position, including padded positions, then ignore special/pad offsets during span construction.
Use sliding windows for long text and merge by original character offsets.

Metrics

Original dense held-out test F1: 0.9708

CoreML contract quality check: PyTorch evaluation with the same all-ones attention behavior on the first 2,000 held-out test rows:

json

{
  "model_dir": "models/final_runs/dense-lowlr-combined-clean-20260530T222847-0700/google-bert__bert-base-multilingual-cased",
  "dataset": "data/final_splits_audited/combined_clean",
  "split": "test",
  "rows": 2000,
  "max_length": 256,
  "batch_size": 32,
  "attention_mode": "all_ones",
  "precision": 0.9711931486407578,
  "recall": 0.973087174153286,
  "f1": 0.9721392388621898,
  "accuracy": 0.9936590381235996
}

CoreML parity verification:

json

{
  "attention_mode": "all_ones",
  "batch_size": 1,
  "max_length": 256,
  "fp32_argmax_match_rate": 1.0,
  "int4_argmax_match_rate": 1.0,
  "int4_max_abs_diff_vs_torch": 1.3239184617996216
}

Recommended Inference Wrapper

Production use should wrap the CoreML model with:

Sliding windows at max_length=256, with overlap/stride around 96.
Token-offset span reconstruction; ignore special tokens and zero-length offsets.
Whitespace trimming and overlap de-duplication.
High-precision regex/rule assists for emails, Iranian mobile/phone numbers, national IDs, postal codes, dates, card numbers, and IMEI exclusion.
Cue-word correction for labels near کد ملی, گواهینامه, گذرنامه, کدپستی, شماره تماس, and ایمیل.

See inference_coreml.py and CoreMLWrapperContract.swift for minimal wrapper contracts.

Known Edge Cases

Do not use a standard padding attention mask with this CoreML export. The verified path uses all-ones attention because CoreMLTools' converted transformer mask path did not preserve PyTorch parity.
Long documents require sliding-window inference.
Obfuscated contacts and verbal/spaced phone numbers need deterministic normalization/rules.
IMEI-like device IDs can look like card numbers; validate card numbers before masking as CREDITCARDNUMBER.

Recommended Apple artifact: compact, clean mixed Latin/email behavior, and verified with the CoreML all-ones attention wrapper.

Files

model.4bit-palettized.mlpackage: verified 4-bit CoreML model.
verification.json: fixture-level CoreML parity verification.
coreml_allones_hf_eval_test_2000.json: quality check for the CoreML attention contract.
tokenizer/config files from the trained checkpoint.
reports/: ad hoc edgecase reports from the dense model.

Reza2kn/openmed-persian-pii-google-mbert-coreml-int4

OpenMed Persian PII Google mBERT CoreML INT4

Status

Metrics

Recommended Inference Wrapper

Known Edge Cases

Files

Reza2kn/
openmed-persian-pii-google-mbert-coreml-int4