Repository navigation

All pull requests

Labels (12)Milestones (0)New pull request

Keep DDIM sampling grids within the training schedule and reach the clean boundary
#324Afloat16 opened Oct 6, 2026
Detach BatchNorm running statistics from completed training graphs
#323Afloat16 opened Oct 6, 2026
Apply decoupled weight decay at the current scheduled learning rate
#322Afloat16 opened Oct 6, 2026
Keep fully masked attention queries and gradients finite
#321Afloat16 opened Oct 6, 2026
fix: honor AdaBelief AMSGrad and SGD fallback settings
#320Afloat16 opened Oct 2, 2026
fix: bypass Switch expert overflow through the outer residual
#319Afloat16 opened Oct 2, 2026
Sample mutually exclusive BERT masking outcomes
#318Afloat16 opened Oct 2, 2026
Preserve FP32 Adam master weights and moments when loading checkpoints
#317Afloat16 opened Oct 2, 2026
Preserve rotary attention dtype and rebuild stale device caches
#316Afloat16 opened Oct 2, 2026
Stabilize layer normalization variance and low-precision statistics
#315Afloat16 opened Oct 2, 2026
fix: respect padding modes in weight-standardized convolution
#314Afloat16 opened Sep 29, 2026
docs: fix README typo
#303Zhao73 opened Jun 1, 2026
fix: use raw docstrings for LaTeX math to avoid invalid escape warnings
#300Creeken-Harrans opened May 1, 20262 tasks done
Fix typo: 'Langauge' -> 'Language' in readme
#299kriti2110 opened Apr 27, 2026
Update util.py in stable diffusion to avoid Sizes of tensors error
#278CatManJr opened Nov 28, 2024