Project Maya v1.1.31: MiniMax M3.1 Flash answers up to 1.5x faster locally
Local inference project Project Maya now has the MiniMax-M3.1-Flash DSpark drafter read the whole conversation, making answers up to 1.5x faster with identical output, alongside a new web UI and AMD/MI50 support.
Project Maya v1.1.31 is out: a new look, and faster models. - A redesigned web UI from @needmorevram: a new chat, Settings, a live Monitor, a Models page to switch models and tune each one's engine, and "Saved", which shows what the model's own API would have charged you. The violet colours are still there as an option. - MiniMax-M3.1-Flash answers up to 1.5x faster. Its DSpark drafter had been guessing without the prompt since it arrived; now it drafts from the whole conversation. Same answers, just sooner. - DeepSeek-V4.1-Flash flies through copied and edited text: send a file back with small changes and it reads ahead from what it already saw, up to about 1.6x faster. - DeepSeek's pictures no longer hold any VRAM between pictures, so more of the model stays on your GPU. Plus 30+ community PRs: fixes, AMD and MI50 support, and new options for multi-GPU setups. Thank you to everyone who contributed, tested and reported. https://github.com/mw00/project-maya/releases/tag/v1.1.31