G github.com MiniMax-M1, first open-weight, large-scale hybrid-attention reasoning model 3 points hallh 1 year ago 0 comments