Struct EngineOptions 

pub struct EngineOptions { /* private fields */ }

Configuration for Engine::load.

Implementations

impl EngineOptions

pub fn new(model_path: impl Into<PathBuf>) -> Self

Construct with the given model path, default quantization (IsqType::Q4K), default max_tokens (1024), and an indexing-safe default sampler profile (RequestOptions::deterministic).

The default request is deterministic because this crate’s primary use case is producing structured output that gets persisted to a search index (see crate::image_analysis::ImageAnalysisTask). Stochastic sampling means the same keyframes reprocessed after a timeout, retry, or backfill can produce different ImageAnalysis values, silently drifting the index. Greedy decoding closes that hole at the cost of diverging from the Qwen3-VL Instruct model card’s recommended sampler — see RequestOptions::deterministic for the full trade-off.

For one-shot, quality-prioritised use where reproducibility doesn’t matter, swap the engine default for the model-card stochastic profile via EngineOptions::new(path).with_request(RequestOptions::new()), or override per-call via Engine::run_with.

pub fn model_path(&self) -> &Path

Returns the configured model path.

pub fn with_model_path(self, val: impl Into<PathBuf>) -> Self

Builder-style setter for model_path.

pub fn set_model_path(&mut self, val: impl Into<PathBuf>) -> &mut Self

In-place setter for model_path.

pub const fn quantization(&self) -> IsqType

Returns the configured quantization (default IsqType::Q4K).

pub const fn with_quantization(self, val: IsqType) -> Self

Builder-style setter for quantization.

pub const fn set_quantization(&mut self, val: IsqType) -> &mut Self

In-place setter for quantization.

pub const fn max_tokens(&self) -> usize

Returns the configured max_tokens ceiling (default 1024).

pub const fn with_max_tokens(self, val: usize) -> Self

Builder-style setter for max_tokens. Any value is accepted at the type level (no setter-side validation); a value of 0 is clamped up to 1 at request time inside Engine::run_with before being passed to mistralrs’s set_sampler_max_len, so a zero here means “let the model emit at least one token”, not “skip generation entirely”.

pub const fn set_max_tokens(&mut self, val: usize) -> &mut Self

In-place setter for max_tokens. See Self::with_max_tokens for the runtime 0 → 1 clamp note.

pub const fn request(&self) -> &RequestOptions

Returns the engine-level default RequestOptions. This is the sampler profile used by Engine::run; per-call overrides go through Engine::run_with.

Default (EngineOptions::new): RequestOptions::deterministic — see that constructor for the indexing-vs-quality trade-off.

pub fn with_request(self, val: RequestOptions) -> Self

Builder-style setter for request. Replaces the engine-level default sampler profile wholesale.

pub fn set_request(&mut self, val: RequestOptions) -> &mut Self

In-place setter for request. Replaces the engine-level default sampler profile wholesale.

pub const fn inference_timeout(&self) -> Duration

Returns the per-call inference timeout (default DEFAULT_INFERENCE_TIMEOUT = 5 min). Issue #1 H-001.

pub const fn with_inference_timeout(self, val: Duration) -> Self

Builder-style setter for inference_timeout.

pub const fn set_inference_timeout(&mut self, val: Duration) -> &mut Self

In-place setter for inference_timeout.

Trait Implementations

impl Clone for EngineOptions

fn clone(&self) -> EngineOptions

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for EngineOptions

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.