Model card auto-generated by SimpleTuner

README.md

@@ -55,8 +55,8 @@ You may reuse the base model text encoder for inference.
 
 ## Training settings
 
-- Training epochs: 109
-- Training steps: 220
+- Training epochs: 4
+- Training steps: 10
 - Learning rate: 0.0001
   - Learning rate schedule: constant
   - Warmup steps: 500
@@ -66,15 +66,15 @@ You may reuse the base model text encoder for inference.
   - Gradient accumulation steps: 1
   - Number of GPUs: 3
 - Gradient checkpointing: False
-- Prediction type: epsilon[]
+- Prediction type: epsilon (extra parameters=['training_scheduler_timestep_spacing=trailing', 'inference_scheduler_timestep_spacing=trailing', 'controlnet_enabled'])
 - Optimizer: adamw_bf16
 - Trainable parameter precision: Pure BF16
 - Base model precision: `no_change`
 - Caption dropout probability: 0.0%
 
 
-- LoRA Rank: 16
-- LoRA Alpha: 16.0
+- LoRA Rank: 64
+- LoRA Alpha: 64.0
 - LoRA Dropout: 0.1
 - LoRA initialisation style: default
     
@@ -97,36 +97,43 @@ You may reuse the base model text encoder for inference.
 
 ```python
 import torch
-from diffusers import DiffusionPipeline
-
-model_id = 'terminusresearch/pixart-900m-1024-ft-v0.6'
-adapter_id = 'bghira/pixart-controlnet-lora-test'
-pipeline = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.bfloat16) # loading directly in bf16
-pipeline.load_lora_weights(adapter_id)
-
+from diffusers import PixArtSigmaPipeline, PixArtSigmaControlNetPipeline
+# if you're not in the SimpleTuner environment, this import will fail.
+from helpers.models.pixart.controlnet import PixArtSigmaControlNetAdapterModel
+
+# Load base model
+base_model_id = "terminusresearch/pixart-900m-1024-ft-v0.6"
+controlnet_id = "bghira/pixart-controlnet-lora-test"
+
+# Load ControlNet adapter
+controlnet = PixArtSigmaControlNetAdapterModel.from_pretrained(
+    f"{controlnet_id}/controlnet"
+)
+
+# Create pipeline
+pipeline = PixArtSigmaControlNetPipeline.from_pretrained(
+    base_model_id,
+    controlnet=controlnet,
+    torch_dtype=torch.bfloat16
+)
+pipeline.to('cuda' if torch.cuda.is_available() else 'cpu')
+
+# Load your control image
+from PIL import Image
+control_image = Image.open("path/to/control/image.png")
+
+# Generate
 prompt = "A photo-realistic image of a cat"
-negative_prompt = 'ugly, cropped, blurry, low-quality, mediocre average'
-
-## Optional: quantise the model to save on vram.
-## Note: The model was not quantised during training, so it is not necessary to quantise it during inference time.
-#from optimum.quanto import quantize, freeze, qint8
-#quantize(pipeline.transformer, weights=qint8)
-#freeze(pipeline.transformer)
-    
-pipeline.to('cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu') # the pipeline is already in its target precision level
-model_output = pipeline(
+image = pipeline(
     prompt=prompt,
-    negative_prompt=negative_prompt,
+    image=control_image,
     num_inference_steps=16,
-    generator=torch.Generator(device='cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu').manual_seed(42),
-    width=1024,
-    height=1024,
     guidance_scale=4.0,
+    generator=torch.Generator(device='cuda').manual_seed(42),
+    controlnet_conditioning_scale=1.0,
 ).images[0]
 
-model_output.save("output.png", format="PNG")
-
-```
+image.save("output.png")