<!-- agent-source-envelope -->
# Artifact: ANE M8 Core ML Chain | PTLLM

> A layer-chunked Core ML chain ran a Qwen3 28-block path on the Apple Neural Engine at about 17 tok/s.

Canonical: https://posttrainllm.com/artifacts/ane-m8-coreml-chain
<!-- /agent-source-envelope -->

Mac runtime artifact

Parked

2026-06-17

This artifact maps the boundary between owning the model and using Apple’s acceleration stack. It is parked because capability comes from our model/eval loop; Core ML is a deployment target, not the product center.

ANE, Core ML, Qwen3, Mac

## Headline Numbers

Evidence class applies to each value: measured = retained evaluation or
runtime result; historical = recorded result not freshly reproduced;
derived = calculated or decided from records; observed = current artifact
state; not-measured = an explicit evidence gap.

**ANE decode:** 17 tok/s

Qwen3 28-block layer-chunked chain

Evidence: historical

**FoundationModels context:** 4096

too small for real tool catalogs

Evidence: historical

**Action grounding:** 25%

FoundationModels BFCL agentic full catalog

Evidence: historical

## Competitive Context

| System | Metric | Score | Size / Class | Comparable? | Readout |
| --- | --- | --- | --- | --- | --- |
| posttrainllm Qwen3 Core ML chain | ANE decode | 17 tok/s | 28-block Qwen3 path | Direct | Runtime experiment only; capability still comes from posttrainllm weights and evals. |
| Apple FoundationModels | action grounding / context | 25% / 4096 tokens | Apple on-device model | Directional | Useful as a free local floor, but too weak to be the specialist capability dependency. |
| posttrainllm active MLX path | specialist eval readiness | active | owned weights | Directional | Preferred competition lane: model quality first, runtime optimization second. |

Direct rows share this artifact's eval setup. Directional rows are
useful market context but should not be read as leaderboard claims.

## Platform stance

| Path | Decision | Reason |
| --- | --- | --- |
| Apple FoundationModels | Routing floor only | Weak action grounding and short context |
| Our weights -> Core ML | Optional future | Battery/perf optimization if capability is already solved |
| MLX/posttrainllm runtime | Active | Own the model and eval gate |

## Release Blockers

### Capability dependency rejected

Apple's model cannot be the differentiator; it is a free local floor at best.

**Unblock:** Only revive Core ML when a shipped posttrainllm specialist needs a battery/runtime optimization.

## Evidence

- [ANE/CoreML parked lane](https://posttrainllm.com/docs/parked/ane-coreml)

- [Apple on-device model notes](https://posttrainllm.com/docs/learn/apple-on-device-foundation-models)

- [Project status](https://github.com/PostTrainLLM/posttrainllm/blob/main/PROJECT_STATUS.md)

## Next Release Action

Leave parked. Keep the numbers public as boundary-mapping evidence.
