lmms-lab

/

llava-next-110b

like 21

Through Qwen1.5-110B run best performance in some benchmark, but it may not a enough trained model, its basic ability is weak even for calculating a1 + a2 + ... + a7.
Comparing Qwen1.5 series - 72B / 32B / 110B, 32B is most suitable one for actually deployment.