Release 0.0.3 · google/jaxite

What's Changed

Minor documentation updates by @copybara-service in #14
Update README.md by @mr0re1 in #15
[LSC] Ignore incorrect type annotations related to jax.numpy APIs by @copybara-service in #21
change internal link to blog post by @copybara-service in #22
Use dedicated helper for scaling a matrix by x^n - 1 by @copybara-service in #26
internal testing infrastructure change by @copybara-service in #27
Implement TPU kernel for signed toeplitz by @copybara-service in #29
Use toeplitz kernel in polymul for TPU version >=5 by @copybara-service in #31
bump jax to 0.4.25 by @copybara-service in #33
minor tweaks for improved use of jax API by @copybara-service in #34
Add an initial vector-matrix polymul megakernel by @copybara-service in #35
Updated pl.BlockSpec argument order following jax-ml/jax#22209 by @copybara-service in #36
Normalize out of bound values in random_source.py by @copybara-service in #37
Fix lwe_test to be compatible with NumPy 2. by @copybara-service in #38
Fix matrix_utils_test to be compatible with NumPy 2. Update dtype to jnp.int32 where applicable to avoid overflow errors. by @copybara-service in #39
Fix test_poly_mul to be compatible with NumPy 2. The issue is recasting float64 to int32. jnp.int32() seems to be rounding down overflows to signed_int32.MAX and signed_int32.MIN. by @copybara-service in #40
Current TPU supports BF16 matmul which improves the performance a little from 7943 us into 7643 us. by @copybara-service in #48
Bump version to 0.0.3 by @copybara-service in #47

Full Changelog: v0.0.2...v0.0.3