agentsclimarketplace

Choose apple model runtime

Skill gaelic-ghost/socket/skills/choose-apple-model-runtime

The Source for macOS Agent Workflows

Install
npx -y skills add gaelic-ghost/socket --skill choose-apple-model-runtime

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 6 stars6 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Compare and select Core AI, Core ML, MLX, MLX Swift, MLX LM, ExecuTorch Apple delegates, or Foundation Models. Use when an Apple model workflow needs a runtime choice and implementation handoff.

SKILL.md

3.2 KB, as published. Nobody here has run it

Choose Apple Model Runtime

Route By Artifact And Constraint

NeedStart with
Author .aimodel packages with editable Python primitives and Swift runtime utilitiesChoose Core AI, then hand off to the coreai-models working-with-coreai and model-authoring skills
Lower torch.export.ExportedProgram into Core AI IRcoreai-torch
Quantize, palettize, or prune Core AI modelsChoose Core AI, then hand off to Apple's model-compression-exploration skill and coreai-optimization
Convert and deploy established Core ML model packagescoremltools plus Core ML
Train or run tensor programs natively on Apple siliconMLX
Integrate MLX models in SwiftMLX Swift
Fine-tune or serve supported language models with MLXMLX LM
Use one ExecuTorch .pte pipeline with Apple accelerationCompare the ExecuTorch Core ML backend and experimental MLX delegate
Use Apple's system on-device language model without shipping weightsFoundation Models framework; use Python Apple FM SDK for supported Python access

Decision Workflow

  1. Identify the source artifact: PyTorch module/export, safetensors checkpoint, Core ML package, Core AI package, ExecuTorch program, or system model.
  2. Identify the deployment API: Python research, Swift app, ExecuTorch C++/mobile, or Foundation Models.
  3. Consult the dated maturity and availability matrix in references/apple-model-tooling.md, then confirm OS, Xcode, SDK, device, architecture, operator, dynamic-shape, state/cache, and precision requirements against the current official source.
  4. Select the shortest supported conversion path. Do not round-trip through formats merely because converters exist.
  5. Prototype one representative subgraph and one stateful generation step before converting the full model.
  6. Evaluate numerical/behavioral parity on the exact packaged artifact.
  7. Benchmark on the target device with benchmark-model-runtime.
  8. Return the selected runtime, maturity class, source revision/date checked, unmet availability gates, and implementation owner. Hand Core AI authoring/compression to Apple's named skills and app-facing Swift/Xcode work to apple-dev-skills.

Important Distinctions

  • Core AI and Core ML are related Apple deployment surfaces but are not interchangeable artifact formats or APIs.
  • MLX is a general Apple-silicon array framework; MLX LM and MLX Swift are distinct higher-level/use-language surfaces.
  • ExecuTorch's MLX delegate is marked experimental and under active development upstream. Treat support as revision-specific and compare it separately with the Core ML backend.
  • Foundation Models uses Apple's system model and availability contract; it is not a route for packaging arbitrary user-supplied weights.
  • Apple research repositories vary from reusable frameworks to benchmark or paper-reproduction code. Classify the repository before recommending it as infrastructure.

References

Read references/apple-model-tooling.md for the official source map and verification checklist.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.