Crate silvic_neuron 

AWS Neuron training and explicit float32 serving.

The existing Rust learner drives NeuronBackend. NeuronPredictor scores LightGBM models with float32 inputs, thresholds, leaves, ordered sums and output; it does not promise exact float64 predictions. Both require a Neuron runtime and device; device failures never select another backend.

Structs

NeuronBackend

Neuron-resident histogram, stable partition and binned tree-scoring backend. Objectives, stochastic quantization, bag draws and fixed-point encoding use core’s host helpers; no CPU backend is constructed or used.

NeuronPredictor

A float32 Neuron predictor. Inputs, leaves, and ordered tree sums are float32; numerical thresholds round down. Device failures never select a CPU fallback.

Functions

is_available

Whether the native runtime exposes NeuronCores, without initializing it or reserving cores.