Comparison to steampunque hybrid quant
#26
by Throghar - opened
I want to share my initial findings on hybrid quant from steampunque to ignite further testing and discussion on the topic:
https://huggingface.co/steampunque/Qwen3.5-27B-MP-GGUF/discussions/1
In the end i think there is some overthinking in unsloth quants that may come from calibration maybe or compared to steampunques approach with high quality start/end quants that may produce this difference? Not sure hope it will help with improvement of this great model.