Mr-Neutr0n b68663426f fix: preserve sign of denominator in clamp to avoid inverting gradient direction ...

The previous .clamp(min=1e-6) on (sigma_ - sigma) flips the sign when
the denominator is negative (which is the typical case since sigmas
decrease monotonically). This would invert the target and cause
training divergence.

Use torch.sign(denom) * torch.clamp(denom.abs(), min=1e-6) instead,
which prevents division by zero while preserving the correct sign.

2026-02-11 21:04:55 +05:30

..

__init__.py

DiffSynth-Studio 2.0 major update

2025-12-04 16:33:07 +08:00

base_pipeline.py

Merge branch 'main' into ltx-2

2026-02-03 13:06:44 +08:00

flow_match.py

Merge branch 'main' into ltx-2

2026-02-03 13:06:44 +08:00

logger.py

[feature]:Add adaptation of all models to zero3

2026-01-27 11:24:43 +08:00

loss.py

fix: preserve sign of denominator in clamp to avoid inverting gradient direction

2026-02-11 21:04:55 +05:30

parsers.py

DiffSynth-Studio 2.0 major update

2025-12-04 16:33:07 +08:00

runner.py

[feature]:Add adaptation of all models to zero3

2026-01-27 11:24:43 +08:00

training_module.py

support auto detact lora target modules

2026-01-21 11:05:05 +08:00