Commit History
-
Fix SDE inference producing noise when shift != 1.0
Verified signatureThe SDE inference path computes `next_timestep` using a linear formula: ``` next_timestep = 1.0 - (float(step_idx + 1) / infer_steps) ``` This ignores the shift transformation applied to the timestep schedule: ``` t = shift * t / (1 + (shift - 1) * t) ``` When shift != 1.0, the renoise step uses a timestep that doesn't match the shifted schedule the model expects, causing accumulated error and noise-only output. `t_prev` is already available from the loop iterator and includes the shift transformation. The fix replaces the linear calculation with t_prev: # Before: ``` next_timestep = 1.0 - (float(step_idx + 1) / infer_steps) xt = self.renoise(pred_clean, next_timestep) ``` # After: ``` xt = self.renoise(pred_clean, t_prev) ``` When shift == 1.0, the behavior is identical to the current code. When shift != 1.0, the schedule is now …b2e8955
-
Update modeling_acestep_v15_base.py
Verified signaturefix issue: https://github.com/ace-step/ACE-Step-1.5/issues/214e432212
-
Update README.md
Verified signature76bb727
-
Update README.md
Verified signature2ccb690
-
Update README.md
Verified signature09ff313
-
Upload folder using huggingface_hub
Verified signature353e85b -
initial commit
Verified signaturefe9706d
