Download
and setup
It was a mad 3-month sprint to release all this as a single person. I will appreciate your understanding if some parts are not yet polished enough, and I will welcome any kind of support from now on.
Here is a github repo. It also contains a binary that should run straight away from MacOS.
huggingface.co/kolinko/mistral-buckets
Also, if you're a researcher, a GPU developer, or you'd like to implement the algorithm in your project (llama.cpp, MLX - looking at you!) - please reach out to kolinko@gmail.com Thanks!
Index
- MoE, quantization and the others.
- Pesky details (or: Help Needed!)
- Citations, notes and so on