Read every tensor from an adapter safetensors file (small, f32) into
materialized mlx arrays. Tensor names come from our header parser;
arrays from mlx's native loader via the proven map-get pattern
(weights.ts). The map and header mmap are freed before returning.
Read every tensor from an adapter safetensors file (small, f32) into materialized mlx arrays. Tensor names come from our header parser; arrays from mlx's native loader via the proven map-get pattern (weights.ts). The map and header mmap are freed before returning.