GLM 5.3 Flash Lite idea
24b, 12b active (because 3b is fast, but bad)
alternatively, 60B-a6b
12b active is too much for a 24b parameters model.
5-6b would be the ideal.
alternatively, 60B-a6b
too big for most GPU's
12b active is too much for a 24b parameters model.
5-6b would be the ideal.
But much smarter than 6b
alternatively, 60B-a6b
too big for most GPU's
Best use of MoE models is to offload the experts to RAM. The active 6B parameters can fit in a 8GB GPU as long as you have enough RAM for the rest.
If they wanted they would, but since they don't, they won't.
YES PLEASE GLM 5.3 Flash Lite 24B A6B I WOUld Love that Please Z.AI You can then replace my Qwen3.8
I want 14-16B dense or 20-24B A5B MoE
I want 14-16B dense or 20-24B A5B MoE
Same I have 5070 ti
WE NEED A <10B AGAIN LIKE GLM 4 9B
70B dense or like a 150B MoE
70B dense or like a 150B MoE
Please don't that doesent fit on my gpu