0xSero

/

GLM-5.1-444B-A14B-REAP

like 2

I was thinking how far you could take REAP. Could you try turning Minimax M2.7 from a 230b a10b model into a 23b a10b (90% prune) model to see how well it performs? It could make for a unique model of that size if it works, potentially making it a great option for 16 GB GPUs.