Search results

Question: Stage-3 Training with DeepSpeed ZeRO-3

Status: Open.

#21 In thu-ml/Motus;

· Qian0733 opened on Feb 2, 2026

WeChat QR Code on Website Has Expired

Status: Open.

#20 In thu-ml/Motus;

· BitCalSaul opened on Jan 26, 2026

感谢你们的高质量工作，我对训练wan部分的代码很感兴趣，请问你们有计划开源这部分代码吗？ 希望你们不吝赐教！

Status: Open.

#19 In thu-ml/Motus;

· frost-ice0107 opened on Jan 23, 2026

Expected convergence loss range for Wan fine-tuning loss?

Status: Open.

#18 In thu-ml/Motus;

· dazazh opened on Jan 21, 2026

Why are register tokens disabled during pre-training but enabled during fine-tuning?

Status: Open.

#17 In thu-ml/Motus;

· einnullnull opened on Jan 17, 2026

大佬们 是否可以换成全模态模型 增加latent action token 然后增加一个action decoder解码latent action为实际动作的形式

Status: Open.

#16 In thu-ml/Motus;

· ryanwangran98 opened on Jan 15, 2026

Example inference images and prompts

Status: Open.

#15 In thu-ml/Motus;

· Luvata opened on Jan 14, 2026

Purpose and Impact of global_downsample_rate

Status: Open.

#13 In thu-ml/Motus;

· einnullnull opened on Jan 5, 2026

bad performace in robotwin evaluation

Status: Open.

#11 In thu-ml/Motus;

· justimyhxu opened on Dec 31, 2025

robotwin data conversion issue

Status: Open.

#10 In thu-ml/Motus;

· bighuang624 opened on Dec 29, 2025

Training resources and stage 1/2 scripts

Status: Open.

#9 In thu-ml/Motus;

· inf-usb opened on Dec 25, 2025

bad result in real wold environment

Status: Open.

#8 In thu-ml/Motus;

· JasperChennn opened on Dec 24, 2025