OpenMed-mLiteClinical-IrishCorePII-135M-v1

temsa/OpenMed-mLiteClinical-IrishCorePII-135M-v1

Full transformers checkpoint derived from OpenMed/OpenMed-PII-mLiteClinical-Base-135M-v1 and tuned for Irish core PII:

• PPSN
• account_number
• bank_routing_number
• credit_debit_card
• PASSPORT_NUMBER
• postcode
• phone_number
• email
• first_name
• last_name
• swift_bic

The main focus is English + Irish Gaelic (ga) handling for Irish administrative, citizen-support, and HSE-style text.

Included Artifacts

• Full transformers model files in the repo root
• Dynamic int8 ONNX export in onnx/model_quantized.onnx
• inference_mask.py for the full model
• inference_mask_onnx.py for the ONNX int8 artifact
• clean benchmark summaries in eval/

Recommended Inference

Highest accuracy:

python3 inference_mask.py \
  --text "My PPSN is 1234567TW and call me on 087 123 4567." \
  --json

Fast CPU path:

python3 inference_mask_onnx.py \
  --text "My PPSN is 1234567TW and call me on 087 123 4567." \
  --json

Dedicated Eircode example:

python3 inference_mask.py \
  --text "My Eircode is D02 X285." \
  --json

Benchmarks

Reference comparison on the manual Irish core suite and PPSN regression suites:

Edge and multilingual PPSN checks:

Multilingual PPSN throughput on CPU (eval/multilingual_ppsn_v1_all.jsonl):

• Base OpenMed: 42.30 examples/s
• Previous public PPSN model: 42.63 examples/s
• This release: 41.18 examples/s
• ONNX Q8: 81.99 examples/s

Practical Reading Of The Benchmarks

• This release is materially better than the previous public PPSN-only model on Irish phones, Eircodes, account details, passport numbers, and names.
• The bundled ONNX int8 export is useful for CPU speed, but it is not accuracy-identical to the full checkpoint.
• The largest ONNX drops are on credit_debit_card and some PPSN edge cases. Use the full model when those matter.

License And Attribution

• Model weights in this repo are distributed under Apache-2.0.
• Base model: OpenMed/OpenMed-PII-mLiteClinical-Base-135M-v1
• Training data included synthetic Irish data plus attributed upstream data from:
  • joelniklaus/mapa (cc-by-4.0)
  • gretelai/synthetic_pii_finance_multilingual (apache-2.0)
• See NOTICE for attribution details.

<!-- portfolio-comparison:start -->

Portfolio Comparison

Updated: 2026-03-16.

Use this section for the fastest public comparison across the temsa PII masking portfolio.

• The first core table only includes public checkpoints that ship both comparable q8 accuracy and q8 CPU throughput.
• The first PPSN table only includes public artifacts that ship comparable PPSN accuracy and CPU throughput.
• Missing cells in the archive tables mean the older release did not ship that metric in its public bundle.
• DiffMask rows use the reconciled clean_single_pass harness that matches the deployed runtime.
• GlobalPointer rows use the public raw-only span-matrix release bundle and its packaged q8 ONNX artifact.
• The same content is shipped as PORTFOLIO_COMPARISON.md inside each public model repo.

Irish Core PII: Comparable Public Checkpoints

Irish Core PII: Other Public Checkpoints

Finance-boundary q8 F1 is 1.0000 for OpenMed-mLiteClinical-IrishCorePII-135M-v2-rc6, OpenMed-mLiteClinical-IrishCorePII-135M-v2-rc7, OpenMed-mLiteClinical-IrishCorePII-135M-v2-rc8, and all public IrishCore-DiffMask releases from rc1 to rc6. OpenMed-mLiteClinical-IrishCorePII-135M-v2-rc5 ships 0.8750 on that public q8 suite.

PPSN-Only: Comparable Public Artifacts

PPSN-Only: Historical Public Checkpoints

If you need the strongest current raw-only Irish core model, start with IrishCore-GlobalPointer-135M-v1-rc4. If you need the fastest CPU-first raw-only line, compare it against IrishCore-DiffMask-135M-v1-rc6. If you need a PPSN-only artifact, compare the canonical fp32, fp16, and q8 variants of OpenMed-mLiteClinical-IrishPPSN-135M-v1 directly in the table above.

<!-- portfolio-comparison:end -->