The update uses the same engine and virtual environment for DFlash2, which remains gated to 32 GB GPUs. The project’s commit notes say the 16 GB recipe was tuned on ExLlamaV3 1.4.4 and that Windows has not been tested. The repository also recommends deleting .venv to upgrade. These notes describe an installer and engine configuration update; they do not establish end-to-end validation of the new setup across hardware or operating systems.
Raw SignalOct 10, 20261 min read1 source
MiaAI-Lab's one-click Qwen3.8-27B installer upgrades ExLlamaV3 to 1.6.0 with a DFlash2 MTP draft model.
In one minute
MiaAI-Lab’s Qwen3.8-27B installer repository now makes ExLlamaV3 1.6.0 the default engine for every DRAFT value.