IrishCore-GlobalPointer-ContextPII-135M-v1-rc15

IrishCore-GlobalPointer-ContextPII-135M-v1-rc15

IrishCore-GlobalPointer-ContextPII-135M-v1-rc15 is the current expanded-label raw-only PII masking release for Irish public-sector, HSE, and citizen-support flows.

It keeps the same DistilBERT-size GlobalPointer span extractor family as temsa/IrishCore-GlobalPointer-ContextPII-135M-v1-rc14, but ships a stronger bundled decoder for contextual Irish/English address, location, and name recovery while preserving the broader contextual masking behavior from rc14.

Context labels served by this line:

• STREET_ADDRESS
• CITY
• COUNTY
• DATE_OF_BIRTH
• AGE

Core labels retained:

• PPSN
• POSTCODE
• PHONE_NUMBER
• EMAIL
• PASSPORT_NUMBER
• ACCOUNT_NUMBER
• BANK_ROUTING_NUMBER
• SWIFT_BIC
• CREDIT_DEBIT_CARD
• FIRST_NAME
• LAST_NAME

Positioning

rc15 is a decoder-hardening release over rc14.

• weights unchanged
• ONNX graph unchanged
• no external scanner or validator added
• deployment path still single-pass span extraction plus deterministic [PII:LABEL] replacement

What changed in rc15:

• lenited Gaelic address cues now recover full unit-plus-street spans like Árasán 4, Cedar Court
• city recovery now handles city-before-county cases such as Clonakilty, Co. Cork and Kildare Town, Co. Kildare
• county recovery now accepts generic Co. forms while avoiding the earlier Contae na Mí regression from field-style rescue
• the decoder now recovers business-centre style address blocks like Unit 5, Riverpark Business Centre
• structured-cue name recovery now handles prose forms like Anne-Marie O'Connor-Smyth, PPSN ...
• globalpointer_context_redteam_v8 is 1.0000 at both IoU 0.5 and exact-boundary matching on full and q8 paths while globalpointer_context_redteam_v7 remains 1.0000

If you only need the narrower Irish-core structured label set and want maximum CPU throughput, temsa/IrishCore-GlobalPointer-135M-v1-rc4 remains the faster option.

Benchmarks

ONNX q8

Comparison

Evaluation Notes

Additional q8 release checks shipped in this repo:

• eval/q8_globalpointer_context_redteam_v8.json: contextual regression suite for lenited Gaelic address cues, business-centre address blocks, generic city/county field cues, city-before-county recovery, and structured-cue hyphenated names, 1.0000 F1 in rc15
• eval/q8_globalpointer_context_redteam_v7.json: contextual regression suite for generic county forms inside Irish address blocks, retained at 1.0000 F1 in rc15
• eval/q8_globalpointer_context_redteam_v6.json: contextual regression suite for address cue variants, hyphenated DOBs, first-person bare age statements, role-cued lowercase names, and surname fields, retained from rc12
• eval/q8_globalpointer_context_redteam_v5.json: contextual regression suite for slash-separated phones and 1-digit dotted/slashed DOB formats, retained from rc11
• eval/q8_globalpointer_location_coverage_v1.json, v2, v3: Irish location coverage suites retained from rc9

Usage

python3 inference_mask_onnx.py --model temsa/IrishCore-GlobalPointer-ContextPII-135M-v1-rc15 --text "My surname is de Búrca, my PPSN is 1234567T, and my address is Árasán 4, Cedar Court, mBaile Átha Cliath, D08 XY12."

<!-- 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 -->