Gev E4B, Mev engine for the M4
Gev E4B prepared for Mev, the engine Gev uses on Apple silicon: layers 2 to 13 as Neural Engine programs for block sizes of 512 to 1,024 tokens, the other layers as weights for the GPU and CPU.
Built for the M4's Neural Engine on macOS 27. On other chips build your own from the weights.
zsh scripts/fetch-engine.sh gev-e4b
scripts/build-engine.sh gev-e4b builds the same engine from the weights.
License
Apache License 2.0, as Gemma 4.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support