Module sizing 

Structs

FitResult

Gpu

Hardware

Model

ProjectorArtifact

Quantization

SizingRequest

SizingResult

ValidationError

Enums

Architecture

CpuArch

GpuVendor

KvCacheType

KV cache element types shared by llama.cpp and Ollama.

Platform

Constants

HEADROOM

Functions

evaluate

kv_cache_bytes

memory_capacity

parse_param_count

quant_bits

typed_kv_bytes_per_token

Re-sizes a catalog f16 KV rate (K and V halves) for the requested cache types.