Upgrade Qwen3.8-27B to DFlash for extra llama.cpp speed
Original titleIf you are using Qwen3.8-27B + MTP, make sure to upgrade to DFlash for extra speed:
AISummary
Georgi Gerganov says users of Qwen3.8-27B with MTP can get extra speed by switching to DFlash speculative decoding in llama.cpp. The command uses --spec-type draft-dflash with --spec-draft-n-max 7, and it requires the latest llama.cpp v0.6.0.
Source: Georgi Gerganov · x.comPublished · added here