IrishCore-DiffMask-135M-v1-rc6

IrishCore-DiffMask-135M-v1-rc6 is a raw-only Irish PII masking model derived from OpenMed/OpenMed-PII-mLiteClinical-Base-135M-v1.

It is a small, scanner-free span extractor tuned for:

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

The main target is English plus Irish Gaelic text in citizen-support, public-sector, and HSE-style flows. The repo ships both the full transformers checkpoint and a dynamic q8 ONNX artifact for CPU deployment.

Evaluation Harness

Release gating for this checkpoint uses the deployment-aligned clean single-pass token-span path.

• experiments/irish_core_span_raw_only/benchmark_multitask.py
• scripts/eval_dllm_release.py --inference-mode clean_single_pass

These two paths now match exactly on the checked suites. The older diffusion-style eval path is retained only as a training diagnostic and is not used for release scoring.

For this release, the best local raw checkpoint is models/irishcore-diffmask-135m-v1-rc6k-blend-0p30, and the bundled CPU artifact is onnx/model_quantized.onnx.

What "DiffMask" Means Here

This release is not a generative diffusion language model. It is a compact discriminative token-span model trained with a diffusion-style denoising schedule.

The short version:

• Base OpenMed: plain BIO token classification
• DiffMask: token-span extraction with token-presence and boundary heads
• DiffMask training: repeated masked denoising over the same sentence
• DiffMask inference: one forward pass, no iterative refinement, no text generation

Concretely:

• The encoder starts from the DistilBERT-family weights inside OpenMed/OpenMed-PII-mLiteClinical-Base-135M-v1.
• The model adds three task heads over the encoder hidden states:
  • a per-label token-presence head
  • a typed start-boundary head
  • a typed end-boundary head
• During training, each input sentence is corrupted multiple times by replacing a random fraction of visible tokens with [MASK].
• The corruption level follows a short noise schedule from heavy masking to light masking.
• The same gold spans are learned at every noise level, and the losses are averaged across the denoising passes.
• At inference time there is no diffusion loop and no rewrite step: the model runs once and a score-only span decoder reconstructs spans from token scores plus typed boundaries.

So the "DLLM" aspect here is the training recipe: repeated masked denoising over text, not autoregressive generation.

What It Is Not

This model is not a full discrete diffusion language model in the LLaDA sense.

A true DLLM would usually have:

• timestep or noise conditioning inside the model
• iterative denoising at inference time
• multi-step sequence refinement at runtime
• text generation or full-sequence reconstruction as a first-class objective

This release does not do that.

Instead, it uses the diffusion idea only as a training-time robustness trick:

• corrupt the sentence with [MASK] at several noise levels
• train on the same target spans each time
• average those losses

At runtime, it behaves like a normal fast discriminative extractor.

Architecture

• Encoder: DistilBERT-size encoder from the OpenMed mLiteClinical 135M base
• Heads:
  • token presence per released label
  • typed start boundary per released label
  • typed end boundary per released label
• Decoder:
  • score-only span decoding from offsets, token continuity, label-specific thresholds, and typed boundaries
  • no regex candidate extractor
  • no checksum validator
  • no scanner layer

The release behavior is fully defined by the weights plus the bundled decoder in common.py.

Training And Inference Flow

Training:

1. tokenize a sentence with gold BIO spans
2. convert spans into:
   • token-presence targets
   • typed start targets
   • typed end targets
3. create several noised copies of the same tokenized sentence by masking random visible tokens
4. run the same encoder+heads on each noised copy
5. average the losses across those denoising passes

Inference:

1. tokenize the raw text once
2. run a single forward pass
3. predict:
   • which labels are present on each token
   • where each labeled span starts
   • where each labeled span ends
4. decode spans with label-aware thresholds and boundary rules
5. replace the detected spans with placeholders such as [PII:PPSN]

There is no multi-step refinement loop in deployment.

How It Differs From The Original OpenMed Model

The original OpenMed/OpenMed-PII-mLiteClinical-Base-135M-v1 is a standard DistilBertForTokenClassification model:

• one encoder
• one token-classification head
• BIO labels such as B-email, I-email, B-phone_number
• generic token aggregation to recover spans

DiffMask changes two things:

1. Different supervision

   • base OpenMed learns only BIO token labels
   • DiffMask learns token presence plus typed span boundaries

2. Different training recipe

   • base OpenMed is trained as a standard token classifier
   • DiffMask is trained on multiple masked-noised views of the same sentence

That makes DiffMask better suited to structured Irish identifiers and mixed PII masking, while still keeping a small encoder and a fast CPU path.

How It Differs From rc5 And rc8

Model	Core idea	External scanner/validator	Runtime shape
rc5	token classifier + repair logic	yes	heavier, decoder-assisted
rc8	raw-only token-span model	no	one pass + span decoder
DiffMask	raw-only token-span model + denoising training	no	one pass + span decoder

So DiffMask is closest to rc8 operationally, but it uses a stronger training recipe.

Why This Exists

The older rc5 release still depended on a repair-oriented decoder stack. The public rc8 release removed that external logic, but it regressed on several structured Irish identifiers. This release keeps the raw-only deployment shape while re-hardening the model on Irish numeric and mixed-PII cases.

References

Direct implementation references:

• Devlin et al., BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
  https://arxiv.org/abs/1810.04805
• Sanh et al., DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter
  https://arxiv.org/abs/1910.01108
• Fu et al., Boundary Smoothing for Named Entity Recognition
  https://aclanthology.org/2022.acl-long.490/
• Wang et al., SPANNER: Named Entity Re-/Recognition as Span Prediction
  https://aclanthology.org/2021.acl-long.558/

Conceptual diffusion-style training references:

• Nie et al., LLaDA 2.0: Scaling Up Diffusion Language Models to 100B
  https://arxiv.org/abs/2512.15745
• Gong et al., Scaling Diffusion Language Models via Adaptation from Autoregressive Models
  https://arxiv.org/abs/2410.17891

These diffusion papers were used as architectural inspiration for the masked noising schedule. This release does not implement a generative text diffusion runtime.

Included Artifacts

• Full transformers checkpoint in the repo root
• Dynamic q8 ONNX export in onnx/model_quantized.onnx
• Unquantized ONNX export in onnx/model.onnx
• inference_mask.py for the full checkpoint
• inference_mask_onnx.py for the ONNX q8 path
• common.py, model.py, and multitask_model.py implementing the release decoder
• benchmark files in eval/

Artifact sizes:

• Full checkpoint: 514 MB (model.safetensors)
• Dynamic q8 ONNX: 393 MB (onnx/model_quantized.onnx)

How To Use It

Full checkpoint:

uv run python inference_mask.py \
  --model temsa/IrishCore-DiffMask-135M-v1-rc6 \
  --min-score 0.5 \
  --text "My PPSN is 1234567TW, my Eircode is D02 X285, and my phone is 087 123 4567." \
  --json

Dynamic q8 ONNX:

uv run python inference_mask_onnx.py \
  --model temsa/IrishCore-DiffMask-135M-v1-rc6 \
  --min-score 0.5 \
  --text "Please provide your passport NN5123456 and call me on 0851234567." \
  --json

Both scripts emit explicit placeholders like [PII:PPSN] in masked_text.

Q8 Comparison

Deployment-relevant comparison on CPU:

Model	Core F1	Edge F1	Finance F1	Finance-boundary F1	User PPSN F1	GA weak PPSN F1	Multilingual PPSN F1	Hardening F1
rc5 ONNX q8	0.9669	0.9744	0.9362	0.8750	1.0000	1.0000	0.9333	-
rc8 ONNX q8	0.9737	1.0000	1.0000	1.0000	1.0000	1.0000	0.9176	0.7059
IrishCore-DiffMask-135M-v1-rc6 ONNX q8	0.9733	0.9500	0.9804	1.0000	0.7500	1.0000	0.9274	0.9756

CPU throughput references:

Suite	rc5 q8	rc8 q8	IrishCore-DiffMask-135M-v1-rc6 q8
Irish core short-text path	33.6193 ex/s	257.3756 ex/s	130.3415 ex/s
Multilingual PPSN short-text path	35.5561 ex/s	230.5181 ex/s	200.2041 ex/s
Runtime profile source	23.8338 ex/s	179.4708 ex/s	167.1461 ex/s

Notes:

• The rc5 speed references come from its published q8 end-to-end inference stack, which includes its older repair decoder.
• The rc8 and IrishCore-DiffMask-135M-v1-rc6 numbers use the same raw-only token-span ONNX path.
• A weight-only q4 ONNX experiment was also tried during development, but it was slower than q8 on this CPU and is not shipped.

Limits

• This is still a compact model. The hardest remaining errors are multilingual PPSN near-miss cases rather than Irish core numeric formats.
• The release path is intentionally scanner-free. If you need deterministic validation of individual identifier types, add that in your application layer.
• If you rely on release behavior, use the bundled inference scripts or import decode_token_presence_segments from common.py.

License And Attribution

• Release license: Apache-2.0
• Base model: OpenMed/OpenMed-PII-mLiteClinical-Base-135M-v1
• The derivative release remains subject to the attribution terms of the upstream datasets listed above.
• See NOTICE, training_sources.json, and eval/benchmark_summary.json for provenance and benchmark details.

Release Note

rc6 is the first DiffMask package selected after reconciling the evaluation harness. Under the aligned harness, it matches the prior best candidate on fresh holdout, QA feedback exact, UAT exact, multilingual PPSN, edge, finance, and hardening, while improving Irish core on the full checkpoint.

Portfolio Comparison

Updated: 2026-03-15.

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

Repo	Stack	Full Core F1	Q8 Core F1	Q8 Multilingual PPSN F1	Q8 Core ex/s
temsa/IrishCore-GlobalPointer-135M-v1-rc3	GlobalPointer raw-only span-matrix	1.0000	1.0000	0.9213	204.9
temsa/IrishCore-GlobalPointer-135M-v1-rc2	GlobalPointer raw-only span-matrix	0.9934	0.9934	0.9326	231.2
temsa/OpenMed-mLiteClinical-IrishCorePII-135M-v2-rc8	Raw-only token-span	0.9737	0.9737	0.9176	46.1
temsa/OpenMed-mLiteClinical-IrishCorePII-135M-v2-rc7	Hybrid classifier + generated scanner spec	1.0000	0.9934	1.0000	30.0
temsa/OpenMed-mLiteClinical-IrishCorePII-135M-v2-rc6	Hybrid classifier + repair decoders	1.0000	0.9934	1.0000	29.5
temsa/OpenMed-mLiteClinical-IrishCorePII-135M-v2-rc5	Hybrid classifier + repair decoders	0.9737	0.9669	0.9333	34.4
temsa/OpenMed-mLiteClinical-IrishCorePII-135M-v2-rc4	Hybrid classifier + repair decoders	0.9870	0.9740	0.9600	114.2
temsa/OpenMed-mLiteClinical-IrishCorePII-135M-v2-rc3	Hybrid classifier + repair decoders	0.9806	0.9677	0.9333	44.9
temsa/OpenMed-mLiteClinical-IrishCorePII-135M-v2-rc2	Hybrid classifier + repair decoders	0.9554	0.9615	0.7887	119.1
temsa/OpenMed-mLiteClinical-IrishCorePII-135M-v1	Hybrid classifier baseline	0.9530	0.9333	0.9882	103.3
temsa/IrishCore-DiffMask-135M-v1-rc6	DiffMask token-span, scanner-free	0.9801	0.9733	0.9274	130.3
temsa/IrishCore-DiffMask-135M-v1-rc5	DiffMask token-span, scanner-free	0.9733	0.9733	0.9379	249.2
temsa/IrishCore-DiffMask-135M-v1-rc4	DiffMask token-span, scanner-free	0.9733	0.9733	0.9371	29.5
temsa/IrishCore-DiffMask-135M-v1-rc3	DiffMask token-span, scanner-free	0.9664	0.9664	0.9591	30.0
temsa/IrishCore-DiffMask-135M-v1-rc2	DiffMask token-span, scanner-free	0.9664	0.9664	0.9212	247.1
temsa/IrishCore-DiffMask-135M-v1-rc1	DiffMask token-span, scanner-free	0.9801	0.9934	0.9412	251.2

Irish Core PII: Other Public Checkpoints

Repo	Stack	Full Core F1	Q8 Core F1	Q8 Multilingual PPSN F1	Notes
temsa/OpenMed-mLiteClinical-IrishCorePII-135M-v2-rc1	Hybrid classifier prototype	0.9487	—	—	Predates the public q8 artifact.

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

Repo	Artifact	Irish Large F1	Multilingual PPSN F1	User Raw F1	QA v8 F1	CPU ex/s
temsa/OpenMed-mLiteClinical-IrishPPSN-135M-v1	fp32 canonical checkpoint	0.8979	0.9704	0.8000	0.7385	57.4
temsa/OpenMed-mLiteClinical-IrishPPSN-135M-v1-fp16	fp16 CPU/GPU artifact	—	0.9704	0.8000	0.7385	45.8
temsa/OpenMed-mLiteClinical-IrishPPSN-135M-v1-q8	dynamic int8 CPU artifact	—	0.9040	—	—	132.1

PPSN-Only: Historical Public Checkpoints

Repo	Main Published Metrics	Notes
temsa/OpenMed-PPSN-mLiteClinical-v1	same as canonical fp32 repo: multilingual 0.9704, user raw 0.8000	Legacy alias; prefer temsa/OpenMed-mLiteClinical-IrishPPSN-135M-v1.
temsa/OpenMed-PPSN-v6-raw-rc2	irish_reg_v5 0.8750; user_raw 0.8000; qa_v8 0.7385	Raw PPSN-only research checkpoint; no packaged multilingual CPU benchmark row.
temsa/OpenMed-PPSN-v5_1	irish_large_v2 raw 0.9285; qa_v6 hybrid strict 1.0000	Hybrid PPSN-only checkpoint; predates the canonical multilingual suite packaging.
temsa/OpenMed-PPSN-v5	irish_reg_v5 raw 0.8235; irish_reg_v5 hybrid strict 1.0000	Hybrid PPSN-only checkpoint; predates the canonical multilingual suite packaging.
temsa/OpenMed-PPSN-v4	synthetic non-PPSN drift check only	Predates the current PPSN eval suite; no packaged apples-to-apples multilingual CPU row.

If you need the strongest current raw-only Irish core model, start with IrishCore-GlobalPointer-135M-v1-rc3. 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.