Over powered GEmma 3s: 16B / 27B with high reasoning.

#1914
by DavidAU - opened

Hey guys:

Thank you for all the quants you have done - amazing work.

I have two new ones here "hot off the press":

16B GLM 4.7 Flash Variable reasoning , and 27B Deep Reasoning.

The benchmarks are off the charts ; 16B almost reaches 27B power, 27B power exceeds 7 out of 7 benchmarks from org model.
Benches, along with examples are posted.

16B actually has more layers/tensors than a 27B Gemma 3.

https://huggingface.co/DavidAU/gemma-3-16b-it-BIG-G-GLM4.7-Flash-Valhalla-Heretic-Uncensored-Deep-Thinking
https://huggingface.co/DavidAU/Gemma-3-27b-it-vl-GLM-4.7-Flash-HI16-Heretic-Uncensored-Thinking

Thanks in advance
David

Sign up or log in to comment