Trained for 0 epochs and 1000 steps.

Trained with datasets ['text-embeds', 'mj-v6']
Learning rate 8e-06, batch size 32, and 3 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: PixArt-alpha/PixArt-Sigma-XL-2-1024-MS
VAE: madebyollin/sdxl-vae-fp16-fix

README.md

@@ -56,7 +56,7 @@ You may reuse the base model text encoder for inference.
 ## Training settings
 
 - Training epochs: 0
-- Training steps: 500
+- Training steps: 1000
 - Learning rate: 8e-06
 - Effective batch size: 96
   - Micro-batch size: 32

optimizer.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:58125cbdeee71875e41dcb0364eca7fb41c0768eee8e8f8c72612c9376012283
+oid sha256:1597106775343a8153b308b4b7427f3f5469d1ae70ad04c6b7f6723dac36c0ac
 size 3665677155

random_states_0.pkl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2cceeb55a33a3db4f1a295e5aa0a8fcea8f2638c53ec5216a82c7db9b65c4858
+oid sha256:0f3b1a5f95dcc31f7a44acf2784e0d3aa219fd4a333ac008412483e4063a5218
 size 14344

scheduler.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:57feaeea732a8232dc14923ac8e8cff564f2d6d11728d1405a7f3cfc02efb7ed
+oid sha256:a778f242057a9963f512d7713fc392bdaa1ea385692a6d218868f2e42ea21ee4
 size 1000

training_state-mj-v6.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7828c00d6f87d54210b7888c9040dee97e356126dc1d3916106ee737f452288c
-size 19126435
+oid sha256:974645e8da8c2abdafe9545abe09d332633eaabd17909676477a3d68faa60c51
+size 23062435

training_state.json

@@ -1 +1 @@
-{"global_step": 500, "epoch_step": 500, "epoch": 1, "exhausted_backends": [], "repeats": {}}
+{"global_step": 1000, "epoch_step": 1000, "epoch": 1, "exhausted_backends": [], "repeats": {}}

transformer/diffusion_pytorch_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b106bfee3490f721f128596f246bffc8dc8e9d711ef62f21a1532186ba50e5ad
+oid sha256:0404e972345ceaa2b6cdcd3c8022a93e9a66475833776d91fabd3e3dab0a5fdf
 size 1221780352