Search results

train.py: zero_grad is called per microbatch, breaking gradient accumulation

Status: Open.

#67 In bytetriper/RAE;

· Randdl opened on Feb 5, 2026

DINOv2-L related ckpt release plan

Status: Open.

#66 In bytetriper/RAE;

· Daniellli opened on Feb 3, 2026

Request for the training configs of DiT-XL

Status: Open.

#55 In bytetriper/RAE;

· a-little-hoof opened on Dec 31, 2025

Potential Redundancy in Parameterization When Using RMSNorm with AdaLN Modulation

Status: Open.

#52 In bytetriper/RAE;

· rardz opened on Dec 26, 2025

Unable to Reproduce Provided stat.pt Mean/Variance

Status: Open.

#49 In bytetriper/RAE;

· a-little-hoof opened on Dec 2, 2025

[Stage 2] Training collapse with NaN loss

Status: Open.

#26 In bytetriper/RAE;

· Y-Sisyphus opened on Oct 23, 2025

Help needed: training results with DINOv2-B_decXL config

Status: Open.

#17 In bytetriper/RAE;

· sen-ye opened on Oct 16, 2025