request
Hey, I run 32GB DDR5 + 5090, and I want to be able to run Qwen3-Next-80B-A3B without absolutely destroying the fidelity from quantization, I think your MagicQuant would be awesome for that model. Thanks!
100% That's a titan level model for my setup, but it's in the queue! I'll be releasing a MagicQuant of Qwen3 30B A3B 2507 thinker shortly which is the first good MOE results I've achieved (my practice for Qwen Next 80B). I'm still working out some performance issues on larger models. The combinations are causing delays. But, I'm glad someone wants to see that model quantized properly.
With my hardware, it'll likely be a week or two of me buffing out some kinks and letting it brew. the 30B takes me ~24 hours right now (it should be 3X faster, aka the performance issues). But with holidays coming up, I'll do my best, but rest assured, that model is on my list to tackle.
well then that's all I could ever ask for, thank you sir!
