proxima-autograd
Reverse-mode autodiff over a proxima-tensor program: an adjoint transform (Expr -> Expr), relu/softmax as graph-building functions, and Adam as an elementwise update expression — no Optimizer trait, no Tensor value type, no Module tree
Activity
- Latest release
- 2w ago
- Total releases
- 1
- Cadence
- —
- Last 12 months
- 1
Reach
- Downloads
- 0
Details
- License
- MIT OR Apache-2.0
- First release
- Oct 13, 2026
Releases