studio: add --spec-draft-n-max toggle for MTP speculative decoding#5582

Merged

danielhanchen merged 13 commits into

mainfrom

studio-spec-draft-n-max-toggle

May 19, 2026

Commits on May 19, 2026

studio: add --spec-draft-n-max toggle for MTP speculative decoding
danielhanchen
committed
studio: add --spec-draft-p-min and --spec-draft-p-split to spec strip set
danielhanchen
committed
studio/tests: skip MTP capability-probe tests on Windows
danielhanchen
committed
studio: lower MTP GPU default --spec-draft-n-max from 6 to 2
danielhanchen
committed
studio: skip MTP auto-promote on sub-2B models, backfill chat usage
danielhanchen
committed
[pre-commit.ci] auto fixes from pre-commit.com hooks

pre-commit-ci[bot]
authored and
danielhanchen
committed
studio: probe + emit legacy ngram-mod flags for pre-rename llama-server
danielhanchen
committed
[pre-commit.ci] auto fixes from pre-commit.com hooks

pre-commit-ci[bot]
authored and
danielhanchen
committed
studio: sub-3B MTP falls back to ngram-mod, not off
danielhanchen
committed
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
studio: 5-mode Speculative Decoding dropdown (Auto / MTP / Ngram / MTP+Ngram / Off)
danielhanchen
committed
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
studio: reset Speculative Decoding to Auto on model switch
danielhanchen
committed