New open-source model GLM-4-32B with performance comparable to Qwen 2.5 72Bgithub.com/THUDM1 pointsh_tomera year ago