Commit 4d02f53
Refactor Conv_input to Affine with optional convolution
Replace Conv_input variant in type dim with Affine, separating
the affine transformation (stride, offset) from optional convolution
parameters (dilation, kernel, use_padding). This clarifies the
distinction between strided dimensions and actual convolutions.
- Add type convolution = { dilation; kernel; use_padding }
- Replace Conv_input with Affine { stride; over; conv; stride_offset }
- Update type proj's Conv_input similarly with convolution_proj
- Replace dilation=0 checks with conv=None pattern
- Replace !use_padding global ref with use_padding field in convolution
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>1 parent dc9b194 commit 4d02f53
0 commit comments