Trained for 5 epochs and 60260 steps.

Trained with datasets ['text-embed-cache', 'skins-64', 'skins-256', 'skins-heads']
Learning rate 1e-06, batch size 12, and 1 gradient accumulation steps.
Trained with None prediction type and rescaled_betas_zero_snr=False
Using 'trailing' timestep spacing.
Base model: stabilityai/stable-diffusion-3.5-medium
VAE: None

README.md

@@ -104,9 +104,9 @@ You may reuse the base model text encoder for inference.
 
 ## Training settings
 
-- Training epochs: 3
-- Training steps: 33588
-- Learning rate: 2e-06
+- Training epochs: 5
+- Training steps: 60260
+- Learning rate: 1e-06
   - Learning rate schedule: cosine
   - Warmup steps: 200
 - Max grad value: 1.0

transformer/config.json

@@ -1,7 +1,7 @@
 {
   "_class_name": "SD3Transformer2DModel",
   "_diffusers_version": "0.34.0.dev0",
-  "_name_or_path": "/workspace/models/skins/checkpoint-33500",
+  "_name_or_path": "/workspace/models/skins/checkpoint-46000",
   "attention_head_dim": 64,
   "caption_projection_dim": 1536,
   "dual_attention_layers": [

transformer/diffusion_pytorch_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2bb83ce66fa7015f4e9e6b7252ced2a8d9d59f9bac4381fd9d6b2fed42d13867
+oid sha256:9dfff6625b23cc210e6cb66ebd8ef37d56f44177cf889b5a3f792c4c5841acd6
 size 4939433672