v0.2.0
0.2.0
Milestone: 0.2.0 (#2)
Merged Pull Requests
- #6 feat: add packed attention with column-wise packing and inverse-free LayerNorm (#6)
- #7 perf: clean up shared benchmark helpers and test fixtures (#7)
- #8 perf: cache activation coefficients and simplify converter traversal (#8)
- #9 perf: reduce EncryptedTensor metadata overhead (#9)
- #10 perf: cache reusable attention and convolution metadata (#10)
- #11 perf: vectorize pooling and slot packing helpers (#11)
- #12 perf: reduce tensor hot-path Python overhead (#12)