Trained for 0 epochs and 4000 steps.

Trained with datasets ['text-embeds-pixart-filter', 'photo-concept-bucket', 'ideogram', 'midjourney-v6-520k-raw', 'sfwbooru', 'nijijourney-v6-520k-raw', 'dalle3']
Learning rate 1e-06, batch size 24, and 1 gradient accumulation steps.
Used DDPM noise scheduler for training with epsilon prediction type and rescaled_betas_zero_snr=False
Using 'trailing' timestep spacing.
Base model: terminusresearch/pixart-900m-1024-ft-v0.6
VAE: madebyollin/sdxl-vae-fp16-fix

README.md

@@ -47,7 +47,7 @@ You may reuse the base model text encoder for inference.
 ## Training settings
 
 - Training epochs: 0
-- Training steps: 3000
+- Training steps: 4000
 - Learning rate: 1e-06
 - Effective batch size: 192
  - Micro-batch size: 24

optimizer.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:014efd236ef7a9a5a412e0aa0ebcaac9c630a5a3cb8a10d12aa72248b050544f
+oid sha256:3e8b9ca639bd9358d3fcb03b099acf9e84de7c2f54d489353a43c974e37206ad
 size 5451415117

random_states_0.pkl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b9374987bce0d3ea2a6ec3e7a3c2aac8c7befc39468815c3c5df0f36c3aadf2a
+oid sha256:26567c2951bd69789f047e72cb1411581ef8ddc6445e0e31b816d74551882c45
 size 16100

scheduler.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0adb0bb0a3467812a0a6f00ab9968ac25af4cbfc6774b8f48d48414ebc435c42
+oid sha256:7b7429b91a444c5a8f43e083947d5a91bf61f04e7ce25023ed199238219067da
 size 1000

training_state.json

@@ -1 +1 @@
-{"global_step": 3000, "epoch_step": 3000, "epoch": 1, "exhausted_backends": [], "repeats": {"ideogram": 1}}
+{"global_step": 4000, "epoch_step": 4000, "epoch": 1, "exhausted_backends": [], "repeats": {"ideogram": 2}}

transformer/diffusion_pytorch_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8a500414dd0973c7a7d025ab33fe9a5a66e9616331221aeeccffce73bf9a0ebe
+oid sha256:0579c0233daafccb3c16ee715cd3f78fbd01e004a2f299a6de81cad926e5ebd2
 size 1816969728