Expand description
A minimal row-major dense f32 tensor. Ferrox keeps this deliberately small: weights live quantized in the mmap’d GGUF file and are dequantized on demand by ferrox-quant; this type is for activations and small dequantized weight slices during the forward pass.