Trained for 2 epochs and 2500 steps.

Trained with datasets ['text-embeds', 'mj-v6']
Learning rate 8e-06, batch size 2, 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: ptx0/pixart-900m-1024-ft
VAE: madebyollin/sdxl-vae-fp16-fix

README.md

@@ -210,8 +210,8 @@ You may reuse the base model text encoder for inference.
 
 ## Training settings
 
-- Training epochs: 1
-- Training steps: 2000
+- Training epochs: 2
+- Training steps: 2500
 - Learning rate: 8e-06
 - Effective batch size: 2
   - Micro-batch size: 2

optimizer.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:898539234bc565eebda477ad9ffe9e61df2e6ed7d6139df940d8f80ba49f7bac
+oid sha256:0f8fea7662313eec7cb12f967f20148323ad21b65624783d26792e6423c44673
 size 5451415117

random_states_0.pkl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a255dbade4def9deac767af6f6f22491136da38a67b5b24bfae79df5ba3ba6bf
+oid sha256:d8f06e19ca759e2a76cf92bbbb08934473c201976326df04622c4fc80923bf6b
 size 14344

scheduler.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2eb1cce730426e835242d18b368f2e671b3eb662cbe1682f8619450b7c2c1bb9
+oid sha256:d394596b6a9a6eb220c4b514546efe6b5d39875b7f3c514eaa13e2af3c14e8d3
 size 1000

training_state.json

@@ -1 +1 @@
-{"global_step": 2000, "epoch_step": 910, "epoch": 2, "exhausted_backends": [], "repeats": {"mj-v6": 0}}
+{"global_step": 2500, "epoch_step": 320, "epoch": 3, "exhausted_backends": [], "repeats": {"mj-v6": 0}}

transformer/diffusion_pytorch_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:20c760818e3cee478842dec8eba1538adfabbca70c84e120df6695793e2186dc
+oid sha256:411a2107d308c0db404e09f867ab837fa68aac9ed6d2708dbe9e42d421d91d4b
 size 1816969728