Struct Platform 

pub struct Platform;

Implementations

impl Platform

pub fn new() -> Self

pub fn load_model(&self, model_id: &str) -> Result<LoadedModel>

Instantiate a synthetic InferenceEngine with default dimensions. Use this when no pre-trained weights are available.

pub fn load_model_from_file(&self, path: &str) -> Result<LoadedModel>

Load a real ternarized model from a .tern.bin file.

The file must be a ModelCoherence binary produced by scripts/transmute_llama.py. Layers are mapped round-robin to the 13 EPIS experts.

use moe_platform::Platform;
let platform = Platform::new();
let model = platform.load_model_from_file("/path/to/llama32-1b.tern.bin").unwrap();
let result = platform.run_inference(&model, "Should we proceed?").unwrap();
println!("Verdict: {}", result.trit_verdict);

pub fn run_inference( &self, model: &LoadedModel, prompt: &str, ) -> Result<InferenceResult>

Run a forward pass and return a structured ternary result.

Trait Implementations

impl Default for Platform

fn default() -> Self

Returns the “default value” for a type.