MLX 4-bit conversion of IFM/K2-Horizon-MoVA-36B-A4B, a sparse Mixture-of-Experts model with Mixture-of-Values attention (MoVA), 36B total parameters, and approximately 4B active parameters per token. The source checkpoint supports a native context length of 524,288 tokens (512K).
Benchmark results reported by IFM for the original K2-Horizon-MoVA-36B-A4B checkpoint.
| Format | Quantization | Size |
|---|---|---|
| MLX safetensors | Affine 4-bit, group size 64 | 21.07 GB |
The MoE and MoVA router weights remain at 8-bit; other eligible weights use affine 4-bit quantization. This is a text-only release; no vision projector is required. The repository includes a source-compatible chat_template.jinja, the custom MLX model implementation required by this architecture, and SHA256SUMS.
pip install -U mlx-lm
mlx_lm.generate \
--model abenzerps/K2-Horizon-MoVA-36B-A4B-MLX-4bit \
--prompt "Explain why reproducible builds matter." \
--max-tokens 512 --temp 1.0 --top-p 0.95
Use a current mlx-lm release. The source model supports 512K context; usable context length depends on available unified memory and KV-cache settings.
05cab0amodel-00048-of-00048.safetensors 1.2 GBmodel-00014-of-00048.safetensors 449 MBmodel-00021-of-00048.safetensors 449 MBmodel-00028-of-00048.safetensors 449 MBmodel-00029-of-00048.safetensors 449 MBmodel-00038-of-00048.safetensors 449 MBmodel-00040-of-00048.safetensors 449 MBmodel-00045-of-00048.safetensors 449 MBmodel-00012-of-00048.safetensors 449 MBmodel-00015-of-00048.safetensors 449 MBmodel-00030-of-00048.safetensors 449 MBmodel-00037-of-00048.safetensors 449 MBmodel-00041-of-00048.safetensors 449 MBmodel-00044-of-00048.safetensors 449 MBmodel-00046-of-00048.safetensors 449 MBmodel-00011-of-00048.safetensors 449 MBmodel-00016-of-00048.safetensors 449 MBmodel-00019-of-00048.safetensors 449 MBmodel-00020-of-00048.safetensors 449 MBmodel-00022-of-00048.safetensors 449 MBmodel-00023-of-00048.safetensors 449 MBmodel-00027-of-00048.safetensors 449 MBmodel-00032-of-00048.safetensors 449 MBmodel-00033-of-00048.safetensors 449 MBmodel-00034-of-00048.safetensors 449 MBmodel-00035-of-00048.safetensors 449 MBmodel-00036-of-00048.safetensors 449 MBmodel-00042-of-00048.safetensors 449 MBmodel-00043-of-00048.safetensors 449 MBmodel-00047-of-00048.safetensors 449 MBmodel-00024-of-00048.safetensors 449 MBmodel-00026-of-00048.safetensors 449 MBmodel-00031-of-00048.safetensors 449 MBmodel-00013-of-00048.safetensors 449 MBmodel-00018-of-00048.safetensors 449 MBmodel-00025-of-00048.safetensors 449 MBmodel-00039-of-00048.safetensors 449 MBmodel-00017-of-00048.safetensors 449 MBmodel-00004-of-00048.safetensors 449 MBmodel-00010-of-00048.safetensors 449 MBmodel-00006-of-00048.safetensors 449 MBmodel-00007-of-00048.safetensors 449 MBmodel-00008-of-00048.safetensors 449 MBmodel-00009-of-00048.safetensors 449 MBmodel-00005-of-00048.safetensors 449 MBmodel-00001-of-00048.safetensors 47 MBmodel-00003-of-00048.safetensors 47 MBmodel-00002-of-00048.safetensors 47 MBHave the files? Submit a torrent.
Submit a torrentHave the complete files and permission to share them? Replace the example folder path. The script checks the files, makes a torrent and submits its magnet and checksums. Keep your torrent client seeding. Get your community key from your settings, and read the script before running it.
curl -fsSL https://pirateface.co/package.sh | bash -s -- --repo abenzerps/K2-Horizon-MoVA-36B-A4B-MLX-4bit ./model-folderIFM/K2-Horizon-MoVA-36B-A4B