Repository navigation

Core ML: run ops that read uint8 and int8 inputs on the CPU, refuse ops coremltools gets wrong, and fix fills, tanh GELU, RMSNorm and repeated outputs - #24317

Open

shoumikhin wants to merge 38 commits into

pytorch:mainfrom

shoumikhin:coreml-uint8-inputs

Open

Core ML: run ops that read uint8 and int8 inputs on the CPU, refuse ops coremltools gets wrong, and fix fills, tanh GELU, RMSNorm and repeated outputs#24317
shoumikhin wants to merge 38 commits into
pytorch:mainfrom
shoumikhin:coreml-uint8-inputs

Commits on Oct 30, 2026
Commits on Oct 31, 2026
Commits on Nov 1, 2026
- Give each Core ML output a value of its own
  shoumikhin
  committed
  e6f7c68
- Duplicate a shared constant also where an op reads it from a list
  shoumikhin
  committed
  6ba764d
- Keep index_put forms coremltools cannot convert out of Core ML
  shoumikhin
  committed
  d50afdd
- Check tanh gelu, RMSNorm eps and int zeros without the Core ML runtime
  shoumikhin
  committed
  1380f93
- Pick the Core ML ops to keep once per program
  shoumikhin
  committed
  8e55fe4
- Check only the Core ML state that Core ML takes over
  shoumikhin
  committed
  0ee707d
- Convert Core ML fills with the dtype torch gives them
  shoumikhin
  committed
  8f3f653
- Say why Core ML refuses an op when lower_full_graph=True
  shoumikhin
  committed
  8ac3dae
- Stop the Core ML dry run when a drop cannot change the program
  shoumikhin
  committed
  4d81fbf
- Name the ops that split Core ML state
  shoumikhin
  committed
  c315323