MindLab Macaron-V1 ni taqdim etdi: trillionli modellar sifatini oshirish uchun Mixture-of-LoRA post-trening usuli 64 ta GPU da | allworld.media