pllmRepresentations
Understand how PLLM records a value's format, owner, bounds, and encoding.
A representation states a value's number domain, shape, memory layout, owner, bounds, encoding, and wire format. PLLM uses representations for quantized tensors, ring elements, additive shares, masked activations, CRT residues, garbled labels, packed prefill matrices, and one-time tickets.
Each component lists the representations it accepts and produces. Two components can connect only through a compatible edge or an explicit conversion. Matching tensor shapes are not enough. Wire format versions and ownership changes are part of the plan history.
No public Python API currently exposes representation compatibility records. See current Python support. Current built-in descriptors do not record representation fields; absence is not compatibility evidence.