Struct GpuSessionConfig 

pub(super) struct GpuSessionConfig<'a> {
    pub cache_dir: &'a Path,
    pub model_variant: ModelVariant,
    pub max_seq_length: usize,
    pub intra_threads: usize,
    pub ep: EpSelection,
    pub device_id: u32,
    pub trt_max_workspace_bytes: Option<usize>,
    pub gpu_mem_limit_bytes: Option<usize>,
}

Configuration for loading an ORT session and tokenizer for a single GPU worker.

Bundles the parameters that were previously passed individually to load_models, removing the 9-argument list and the associated #[allow(clippy::too_many_arguments)] suppression. All fields map 1:1 to the corresponding WorkerConfig fields.

Fields

cache_dir: &'a Path

Path to the ONNX model cache directory.

model_variant: ModelVariant

ONNX model variant (FP32, FP16, INT8).

max_seq_length: usize

Maximum tokenized sequence length.

intra_threads: usize

Intra-op thread count for ORT sessions.

ep: EpSelection

Execution provider selection.

device_id: u32

GPU device ID for this worker.

trt_max_workspace_bytes: Option<usize>

Optional TRT EP workspace size cap in bytes.

gpu_mem_limit_bytes: Option<usize>

Optional CUDA EP device memory limit in bytes.