Ethanfel/ComfyUI-SelVA
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

chore: default timestep_mode back to uniform

Browse Source 
logit_normal reaches lower loss but perceptual improvement over uniform
is dataset-dependent. Keeping uniform as default to match original MMAudio
training behavior; logit_normal remains available as an option.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Ethanfel
2026-04-06 01:21:08 +02:00
parent 8e919c0459
commit 57fae4a8ce
2 changed files with 8 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
train_lora.py
+2 -2
View File
@@ -167,8 +167,8 @@ def main():
help="Path to a step checkpoint (.pt) to resume training from.")
parser.add_argument("--precision", default="bf16", choices=["bf16", "fp16", "fp32"])
parser.add_argument("--seed", type=int, default=42)
parser.add_argument("--timestep_mode", default="logit_normal", choices=["logit_normal", "uniform"],
help="Timestep sampling distribution. logit_normal reduces white noise artifacts.")
parser.add_argument("--timestep_mode", default="uniform", choices=["uniform", "logit_normal"],
help="Timestep sampling distribution. uniform matches original MMAudio training. logit_normal reaches lower loss but perceptual improvement is dataset-dependent.")
parser.add_argument("--logit_normal_sigma", type=float, default=1.0,
help="Spread of logit-normal distribution (only used with --timestep_mode logit_normal).")
args = parser.parse_args()