Hey! Thanks for reaching out and for your interest in the YMQ-MTP releases.
I would love to expand the portfolio to cover Qwen 3.6 35B and KATCoder, but unlike standard automated repositories that run massive, unoptimized bulk scripts across a server cluster with a single button click, my architecture-aware YMQ curation process is fully manual.
Because I run all these custom optimization cycles completely on my own hardware setups without any external funding or community donations to offset the computing and electricity overhead, I have to be highly selective with the model sizes I compile.
However, unlike Unsloth's dynamic mapping pipeline, the entire YMQ-Compiler architecture is completely open-source. You do not need massive enterprise GPUs to run it. If you want to compile these specific models yourself using my multi-tier mixed-precision gradients, you can grab the source code, logic templates, and execution wrapper scripts directly from my GitHub repository:
I'll save these architectures to my consideration backlog! If hardware resources open up or community support expands down the line, I'll see if I can get them spun up through the YMQ pipeline.
