Module types 

Public DTOs and the internal EmbedRequest enum exchanged between the pool and the worker threads.

Structs

DualEmbedding

Paired dense + sparse embeddings produced from a single forward pass.

EmbedStats

Per-request diagnostic statistics captured inside the worker and forwarded to the handler layer for inclusion in the completion log event.

ProbeResult 🔒

Result of a single probe session.run() call.

SparseEmbedding

Sparse embedding output from the BGE-M3 sparse-linear projection layer.

Enums

EmbedRequest 🔒

Constants

OS_HEADROOM_BYTES 🔒

OS headroom reserved for kernel, stack, ORT arena, and other non-model allocations. Subtracted from available memory before computing per-worker workspace.