Views
No views yet
lemuralabs/Keye-VL-2.0-30B-A3B-uncensored
with mlx-vlm's mixed_3_4 recipe (3/4-bit mix, group size 64). ~14 GB on disk.This is the "optiq ~3.7 bpw" target build. The closest standard mlx-vlm mixed recipe to 3.7 bpw ismixed_3_4, which measured 3.871 bpw (mixed_3_6was 4.22). If you need a smaller footprint, see the mxfp4 build (≈4.43 bpw but a different format) or request a custom predicate.
Yes — Like the mxfp4 build, this runs Keye coherently on Mac via dense attention (the stock CUDASALightningIndexeris unstable on MPS; this port drops it).
mlx_vlm_keye_support/keyevl2/ (copy into
mlx_vlm/models/keyevl2/, and add "keye_vl2": MessageFormat.LIST_WITH_IMAGE_FIRST, to
mlx_vlm/prompt_utils.py). Then:1python -m mlx_vlm generate --model lemuralabs/Keye-VL-2.0-30B-A3B-uncensored-mlx-optiq-3.7bpw \
2 --prompt "Describe this image." --image path/to/img.jpg --trust-remote-codemixed_3_4, group size 64, 3.871 bpw (whole model incl. vision).mlp_AR projector included; text gen verified coherent,
image-understanding forward is a first-cut port.qwen3_vl_moe, sparse sa_indexer dropped.Lemura Labs)Keye-VL-2.0-30B-A3B-uncensored — abliterated bf16 — full precision (CUDA / vLLM / SGLang)Keye-VL-2.0-30B-A3B-uncensored-mlx-mxfp4 — MLX MXFP4 (~4.43 bpw) — Apple Silicon