Module layout 

Unified Quantized Vector layout — the cache-aligned superset format that absorbs binary / ternary (BitNet 1.58) / 4-bit scalar / residual codecs without polymorphic indirection in the hot path.

128-byte alignment matches AVX-512 cache-line pair and avoids false sharing under thread-per-core execution.

Outlier bitmask limit

The outlier_bitmask in QuantHeader supports up to 64 outlier dimensions per vector. Callers that need to mark more than 64 outliers must bucket by 64-dim windows; multi-window support is out of scope for this module.

Structs

QuantHeader

32-byte interleaved header preceding the packed bit array.

UnifiedQuantizedVector

Owned, 128-byte-aligned unified quantized vector buffer.

UnifiedQuantizedVectorRef

Zero-copy borrowed view into a UnifiedQuantizedVector buffer.

Enums

QuantMode

Quantization mode discriminator stored in the header.

Functions

target_size

Compute the buffer size (rounded up to a 128-byte multiple) required to hold a UnifiedQuantizedVector with the given parameters.