Search results

Runtime clobbers an unrelated live tensor when an int64-comparison bool mask chain executes (deformable-attention sampler pattern)

Status: Open.

#11 In apple/coreai-torch;

· john-rocky opened on Jun 12, 2026

GPU delegate executes aten.floor/trunc/ceil as identity; round uses away-from-zero ties; div(x,1,floor) folds to identity

Status: Open.

#10 In apple/coreai-torch;

· john-rocky opened on Jun 12, 2026

Converter folds float→int→float cast round-trips, dropping truncation semantics (CPU too)

Status: Open.

#9 In apple/coreai-torch;

· john-rocky opened on Jun 12, 2026

Converter aborts (bad_optional_access) on aten.arange with float start/end/step args

Status: Open.

#8 In apple/coreai-torch;

· john-rocky opened on Jun 12, 2026

ANECompiler crash (ANECPlistInterface::addOpToNetwork EXC_BAD_ACCESS) at AIModel.load when slice_update begin/end indices are runtime values

Status: Open.

#6 In apple/coreai-torch;

· scndls opened on Jun 11, 2026

Contribution Proposal: Add stable softplus, mish, and logsumexp conversion support

Status: Open.

#5 In apple/coreai-torch;

· Ashutosh0x opened on Jun 11, 2026

Support Python 3.13 on MacOS

Status: Open.

#4 In apple/coreai-torch;

· cdeil opened on Jun 10, 2026

Support for transposed conv3d

Status: Open.

#3 In apple/coreai-torch;

· devbanu opened on Jun 10, 2026

Runtime MPSGraph null-deref (mlir::FloatType::getWidth) executing 2+ GatedDeltaUpdate layers + an attention layer with dynamic KV context

Status: Open.

#2 In apple/coreai-torch;

· scndls opened on Jun 10, 2026

externalize: SDPA submodule re-export drops the upper bound on the key-length dim with a static query + dynamic KV context

Status: Open.

#1 In apple/coreai-torch;

· scndls opened on Jun 10, 2026