Downloads · 30 days
172
21% of all-time downloads
K2-Horizon-MoVA-36B-A4B-MLX-4bit
iamEvanYT/K2-Horizon-MoVA-36B-A4B-MLX-4bit
K2-Horizon-MoVA-36B-A4B-MLX-4bit is a machine learning model from iamEvanYT. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for mlx. The card lists the license as apache-2.0.
4-bit affine (group size 64) MLX conversion of IFM/K2-Horizon-MoVA-36B-A4B for Apple Silicon via mlx-lm.
All-time downloads
829
Public
Parameters
37.4B
24.3 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors24.3 GB · 100%
Parameter types
How the weights are stored.
U3236.8B · 98%
Base models
Model card
From the Hugging Face model README
K2-Horizon-MoVA-36B-A4B-MLX-4bit
4-bit affine (group size 64) MLX conversion of
IFM/K2-Horizon-MoVA-36B-A4B
for Apple Silicon via mlx-lm.
The K2 MoVA architecture has no stock mlx-lm implementation, so this repo
ships a custom k2_horizon.py, referenced from config.json (model_file).
MoE/MoVA router matrices are kept in BF16; all large projections are 4-bit.
Measured on a 64 GB M4 Pro Mac mini: ~43 tok/s single-stream decode, 18.9 GB peak RSS, perplexity 2.1-8.6 on generic prose/code.
from mlx_lm import load, stream_generate
model, tokenizer = load("iamEvanYT/K2-Horizon-MoVA-36B-A4B-MLX-4bit")
for chunk in stream_generate(model, tokenizer, "Hello.", max_tokens=64):
print(chunk.text, end="", flush=True)
- Library
- mlx
- Type
- k2_horizon
- License
- apache-2.0
- Created
- Sep 3, 2026
- Updated
- Sep 3, 2026