> > [@zsp1993](https://github.com/zsp1993)完成。现在您可以使用 8 个 A100 (8*80G VRAM) GPU`--use_gradient_checkpointing_offload`来`--training_strategy "deepspeed_stage_3"`训练 14B T2V 模型。 > > 训练i2v 14b wan2.1模型时还是oom Hello, did you modify the script of training i2v by yourself？
