GEHIRNS/magi-1.2-2B
Text Generation • 2B • Updated • 62 • 1
MAGI 1.2
Bits fitted to each model by measurement. magi-1.2-2B: loop-DPO + 2B map, density 1.305. magi-1.2-4B: 4B map, density 0.968.
Note Qwen3.5-2B: think-less LoRA + one loop-DPO round + mixed precision fitted to the 2B. 0.77 GiB GGUF, density 1.305 at 32K with a q4_0 KV cache, 333 tok/s on an RTX 5070.
GEHIRNS/magi-1.2-4B
Text Generation • 4B • Updated • 47Note magi-1.1-4B's weights with a mixed-precision map fitted to the 4B by measurement. 1.73 GiB GGUF, density 0.968 at 32K with a q4_0 KV cache (magi-1.1-4B: 0.939), 64 MiB smaller, 186 tok/s on an RTX 5070.