Field EnableSpeculativeDecodingDrafts

Namespace

LMKit.Global

Assembly

LM-Kit.NET.dll

Gets or sets the process-wide default that decides whether the speculative-decoding draft assets packaged with a model are loaded.

public static bool EnableSpeculativeDecodingDrafts

Returns

bool

true to load packaged speculative-decoding drafts by default; otherwise, false. Default is true.

Remarks

This value seeds EnableSpeculativeDecodingDrafts: every LM.LoadingOptions instance adopts whatever this property holds at the moment it is constructed. It is therefore the global default, not a hard switch. To change the decision for an individual model, set EnableSpeculativeDecodingDrafts explicitly when loading that model; the per-model value always wins over this default.

"Draft assets" covers every in-envelope source of candidate tokens for draft-and-verify decoding: Multi-Token Prediction (MTP) head tensors declared by the GGUF, and a draft model shipped inside the LMK archive. When true (the default), these assets are loaded into VRAM when present and the runtime uses them for speculative decoding on supported architectures. When false, they are skipped at load time, which saves VRAM but disables speculative decoding from packaged drafts. This does not affect a draft model assigned explicitly through DraftModel, which is wired independently of the model envelope.