Crate rlx_phi 

Phi 3 / Phi 4 runner.

Phi-3 and Phi-4 ship as general.architecture = phi3 in their GGUF converters (Phi-4 reuses the Phi-3 arch tag upstream — there’s no separate phi4 enum in llama.cpp). Inference is implemented in rlx_llama32::Llama32Runner with Phi-specific partial RoPE and NeoX pairing.

Structs

Llama32Config

Llama32Runner

Llama32RunnerBuilder

PhiRunner

PhiRunnerBuilder

Enums

RopeStyle

Pairing convention for Op::Rope. Different model families bake RoPE into their weights differently, so the rotation kernel needs a flavor:

Constants

FAMILY

PLAN_MILESTONE

Functions

cli_run

Type Aliases

Llama32ConfigSource

Where to load the Llama 3.2 config from.