Post
2,845
MixtureKit 1.1 is out š Turn the fine-tuned models you already have on the Hub into one Mixture-of-Experts model.
pip install mixturekit
Give it a base model and its fine-tunes (math, code, a language...). MixtureKit merges them, trains the routers that pick an expert for every token, and shows you what they learned:
š§© 3 composition methods behind one YAML: Branch-Train-MiX, Mixtral-style MoE, Branch-Train-Stitch
š» Merges stream the experts tensor by tensor: they run on a laptop
šÆ Router training with a load balancing loss, optional LoRA
š A routing visualizer: in the image, two Nile 4B experts split a mixed Egyptian sentence, Arabic script to one expert, Arabizi to the other
š¦ Export to Mixtral / Qwen3-MoE, then GGUF for llama.cpp
š¤ mixturekit push uploads the result with a generated model card
ā Llama 3, Qwen 2.5/3/3.5, Gemma 3/4 experts; tested on transformers 4.52 to 5.19
In the paper, a BTX model built with MixtureKit outperforms the dense baselines on multiple benchmarks.
ā¶ļø Build, train and visualize an MoE in 15 min on a free Colab GPU: https://colab.research.google.com/github/MBZUAI-Paris/MixtureKit/blob/main/examples/build_your_own_moe.ipynb
š Paper: MixtureKit: A General Framework for Composing, Training, and Visualizing Mixture-of-Experts Models (2512.12121)
š» Code: https://github.com/MBZUAI-Paris/MixtureKit
š Docs: https://mbzuai-paris.github.io/MixtureKit/
Work with @guokan-shang and @oe2015 at @MBZUAI-Paris š
pip install mixturekit
Give it a base model and its fine-tunes (math, code, a language...). MixtureKit merges them, trains the routers that pick an expert for every token, and shows you what they learned:
š§© 3 composition methods behind one YAML: Branch-Train-MiX, Mixtral-style MoE, Branch-Train-Stitch
š» Merges stream the experts tensor by tensor: they run on a laptop
šÆ Router training with a load balancing loss, optional LoRA
š A routing visualizer: in the image, two Nile 4B experts split a mixed Egyptian sentence, Arabic script to one expert, Arabizi to the other
š¦ Export to Mixtral / Qwen3-MoE, then GGUF for llama.cpp
š¤ mixturekit push uploads the result with a generated model card
ā Llama 3, Qwen 2.5/3/3.5, Gemma 3/4 experts; tested on transformers 4.52 to 5.19
In the paper, a BTX model built with MixtureKit outperforms the dense baselines on multiple benchmarks.
ā¶ļø Build, train and visualize an MoE in 15 min on a free Colab GPU: https://colab.research.google.com/github/MBZUAI-Paris/MixtureKit/blob/main/examples/build_your_own_moe.ipynb
š Paper: MixtureKit: A General Framework for Composing, Training, and Visualizing Mixture-of-Experts Models (2512.12121)
š» Code: https://github.com/MBZUAI-Paris/MixtureKit
š Docs: https://mbzuai-paris.github.io/MixtureKit/
Work with @guokan-shang and @oe2015 at @MBZUAI-Paris š
