Module engine 

Inference engine — generate loop for a single request.

Orchestrates prefill → decode → sample loop using the Model trait. The engine is stateless — each call to generate creates a fresh sequence, runs inference, and returns output tokens.

Structs

GenerateResult

Result of a generate call.

Functions

generate

Generate response tokens from a prompt.

generate_speculative

Generate with speculative decoding (MTP).

generate_streaming

Generate response tokens with per-token callback.