Yes, it does, but it may be somewhat limited. If you need AVX2 float32 kernels for cosine distance, you are in luck. If you want dynamic dispatch from 200+ SIMD

Yes, it does, but it may be somewhat limited. If you need AVX2 float32 kernels for cosine distance, you are in luck. If you want dynamic dispatch from 200+ SIMD kernels across 4 generations of AVX-512 (Skylake-X, Ice Lake, Sapphire Rapids, and AMD Genoa), AVX2, Arm NEON, Arm SVE, and SVE2 for different mixed-precision distance functions, you may find other options more capable :)