CoreML Export (iOS & macOS)¶

OpenMed can export NER and PII models to Apple's CoreML format for deployment in iOS and macOS applications.

Installation¶

pip install "openmed[coreml]"

This installs coremltools, torch, and transformers.

Converting a Model¶

python -m openmed.coreml.convert \
    --model OpenMed/OpenMed-PII-SuperClinical-Small-44M-v1 \
    --output ./OpenMedPII.mlpackage

Options¶

Output Files¶

The converter produces: - OpenMedPII.mlpackage — CoreML model package - OpenMedPII_id2label.json — Label mapping (include this in your app bundle)

Model Metadata¶

The CoreML model includes embedded metadata: - id2label — JSON string mapping label IDs to names - num_labels — Number of entity labels - max_seq_length — Maximum input length - source_model — Original HuggingFace model ID

Using in Swift¶

See the OpenMedKit Swift Package documentation for using the converted model in iOS/macOS apps.

Manual Integration¶

If you prefer not to use OpenMedKit, you can integrate the CoreML model directly:

import CoreML

let model = try MLModel(contentsOf: modelURL)

// Prepare input
let inputIds = try MLMultiArray(shape: [1, seqLen], dataType: .int32)
let mask = try MLMultiArray(shape: [1, seqLen], dataType: .int32)
// ... fill with token IDs from your tokenizer ...

let input = try MLDictionaryFeatureProvider(dictionary: [
    "input_ids": MLFeatureValue(multiArray: inputIds),
    "attention_mask": MLFeatureValue(multiArray: mask),
])

// Run inference
let output = try model.prediction(from: input)
let logits = output.featureValue(for: "logits")!.multiArrayValue!
// ... apply softmax and decode BIO tags ...

GitHub Actions¶

Use the convert-models.yml workflow to convert models in CI:

1. Go to Actions > "Convert Models"
2. Click "Run workflow"
3. Enter the model ID and desired formats
4. Download the converted model from the workflow artifacts