Repository navigation

Fix wrong dtypes and values in Core ML conversion - #24424

Merged

shoumikhin merged 9 commits into

pytorch:mainfrom

shoumikhin:coreml-converter-fixes

Nov 3, 2026

Merged

Fix wrong dtypes and values in Core ML conversion#24424
shoumikhin merged 9 commits into
pytorch:mainfrom
shoumikhin:coreml-converter-fixes

Commits on Nov 3, 2026
- Cast int and bool inputs to float for Core ML ops that need float
  shoumikhin
  committed
  0a8cabc
- Keep int inputs of rounding and clipping ops int in Core ML
  shoumikhin
  committed
  4a45e45
- Copy a value into a Core ML tensor of another dtype as torch does
  shoumikhin
  committed
  4bc8ca2
- Compute Core ML clamp_min and clamp_max in the dtype torch gives them
  shoumikhin
  committed
  1f22526
- Give Core ML fills the dtype torch gives them
  shoumikhin
  committed
  74795fe
- Read the tanh approximation of gelu in Core ML
  shoumikhin
  committed
  e96fc69
- Pass the default eps of RMSNorm to Core ML
  shoumikhin
  committed
  a16c87e
- Give each Core ML output a value of its own
  shoumikhin
  committed
  19c2580
- Count a negative index from the end in Core ML index_put
  shoumikhin
  committed
  ab6ba2b