# Ollama mlxrunner speculative depth controller (committed tokens per second) > `newSpeculation` builds the subsystem only when the checkpoint has a draft head; a block-draft model gets the DFlash drafter, anything else the MTP drafter. Parent: [Mac local LLMs: Ollama internals](https://llms-explorer.com/tree/mac-local-llms-ollama-internals/) · 1 facets · 14 facts · page: https://llms-explorer.com/tree/ollama-mlxrunner-speculative-depth-controller-co/ ## Facts - `newSpeculation` builds the subsystem only when the checkpoint has a draft head; a block-draft model gets the DFlash drafter, anything else the MTP drafter. — source: `asserted` - The drafter's `draftLimit` is passed to the controller, so a depth the drafter never produces is never scheduled. — source: `asserted` - Depth 0 runs through an inner pipelined plain decoder (parked), which keeps the draft cache level. — source: `asserted` - Replaced a heuristic schedule that grew toward a fixed cap (existing dossier). — source: `asserted` - A drafted round that resumes from parking drains the inner decoder's in-flight sample as the current token and attributes no cost to that round. — source: `asserted` - A stall can inflate a cost sample severalfold, so samples are clamped (see sibling dossier). — source: `asserted` - No source gives measured tokens per second gained by the controller against a fixed depth on a Mac. — source: `asserted` - `newSpeculation` returns nil when the model has no draft head, and then requests decode plainly. — [source](https://raw.githubusercontent.com/ollama/ollama/main/mlxrunner/speculate.go) - The speculation picks the DFlash drafter when the draft model implements `model.BlockDraft`, otherwise the MTP drafter. — [source](https://raw.githubusercontent.com/ollama/ollama/main/mlxrunner/speculate.go) - The controller's `drafterLimit` is set from the drafter's `draftLimit()`, which is blockSize-1 for DFlash and 0 (unbounded) for MTP. — [source](https://raw.githubusercontent.com/ollama/ollama/main/mlxrunner/dflash.go) - A parked request runs an inner pipelined decoder and counts each plain token as a depth-0 round. — [source](https://raw.githubusercontent.com/ollama/ollama/main/mlxrunner/speculate.go) - Resuming drafting from a parked stretch emits the inner decoder's sampled-but-unforwarded token as the current token and records no cost for that round. — [source](https://raw.githubusercontent.com/ollama/ollama/main/mlxrunner/speculate.go) - Each request opens with `limit = depth.scheduled` so a new request starts at the proven depth. — [source](https://raw.githubusercontent.com/ollama/ollama/main/mlxrunner/speculate.go) - Round wall time runs from one round start to the next and spans the next emit's sync. — [source](https://raw.githubusercontent.com/ollama/ollama/main/mlxrunner/speculate.go)