Instructions to use Jundot/Qwen3.6-27B-oQ8-mtp with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use Jundot/Qwen3.6-27B-oQ8-mtp with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir Qwen3.6-27B-oQ8-mtp Jundot/Qwen3.6-27B-oQ8-mtp
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
Qwen3.5 or 3.6?
#1
by david-king-neuma - opened
Is this Qwen 3.5 or 3.6? Tag and config files say 3.5, title says 3.6.
This is Qwen 3.6.
The naming mismatch is intentional. Qwen 3.6 is an incremental training/alignment update, not a new architecture. Alibaba retains the qwen3_5 family identifier for 100% backward compatibility.
All existing Qwen 3.5 code and tooling works unchanged with 3.6. Seeing "3.5" in tags/configs is expected and completely normal. ;)
Ohhhhh thank you for that explanation. Also thanks for the mtp versions, so much faster in oMLX!
david-king-neuma changed discussion status to closed