MLX
Safetensors
qwen3_5_moe
oq
quantized
benchmark
performance
mtp
Mixture of Experts
code
agentic-coding
2-bit
Instructions to use mlx-works/KAT-Coder-V2.5-Dev-oQ2e-mtp with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use mlx-works/KAT-Coder-V2.5-Dev-oQ2e-mtp with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir KAT-Coder-V2.5-Dev-oQ2e-mtp mlx-works/KAT-Coder-V2.5-Dev-oQ2e-mtp
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
oQ4e version!
#1
by rome3ro - opened
Thank you for sharing! I was wondering if you have planned to build the oQ4e version?
KAT-Coder-V2.5-Dev-oQ4e-mtp has been uploaded!
https://huggingface.co/mlx-works/KAT-Coder-V2.5-Dev-oQ4e-mtp
- oQ level: 4e (enhanced)
- MTP: Qwen3.6-35B-A3B original MTP head
- Size: 21 GB
- Compression ratio: 3.1:1