simpletuner-lora

This is a standard PEFT LoRA derived from stabilityai/stable-diffusion-3.5-large.

The main validation prompt used during training was:

k4s4, linechart with 1 lines. Line 1: Line is noisy and trends decreasing at a decreasing rate then decreasing at a constant rate. Overall Description: Line is noisy and trends decreasing at a decreasing rate then decreasing at a constant rate.

Validation settings

• CFG: 7.5
• CFG Rescale: 0.0
• Steps: 28
• Sampler: FlowMatchEulerDiscreteScheduler
• Seed: 42
• Resolution: 1024x1024
• Skip-layer guidance:

Note: The validation settings are not necessarily the same as the training settings.

You can find some example images in the following gallery:

Prompt

unconditional (blank prompt)

Negative Prompt

blurry, cropped, ugly

Prompt

k4s4, linechart with 1 lines. Line 1: Line is noisy and trends decreasing at a decreasing rate then decreasing at a constant rate. Overall Description: Line is noisy and trends decreasing at a decreasing rate then decreasing at a constant rate.

Negative Prompt

blurry, cropped, ugly

Prompt

k4s4, linechart with 3 lines. Line 1: ylabel increases at a constant rate Line 2: ylabel increases at a roughly constant rate Line 3: ylabel increases at a roughly constant rate Overall Description: Lines 2 and 3 share an intersection point at xlabel 1.75 and ylabel 0.75

Negative Prompt

blurry, cropped, ugly

Prompt

k4s4, linechart with 4 lines. Line 1: ylabel decreases roughly at a constant rate and then abruptly drops and decreases at a decreasing rate Line 2: ylabel decreases roughly at a constant rate and then abruptly drops and decreases at a decreasing rate and lastly plateaus to 0 Line 3: ylabel decreases roughly at a constant rate and then abruptly drops and decreases at a decreasing rate and lastly plateaus to 0 Line 4: ylabel first plateaus at 100, then decreases at a decreasing rate Overall Description: All lines converge towards a value of 0 on ylabel

Negative Prompt

blurry, cropped, ugly

Prompt

k4s4, linechart with 1 lines. Line 1: Line is noisy and trends decreasing at a decreasing rate then decreasing at a constant rate. Overall Description: Line is noisy and trends decreasing at a decreasing rate then decreasing at a constant rate.

Negative Prompt

blurry, cropped, ugly

The text encoder was not trained. You may reuse the base model text encoder for inference.

Training settings

• Training epochs: 28

• Training steps: 5000

• Learning rate: 0.0001

  • Learning rate schedule: polynomial
  • Warmup steps: 400

• Max grad norm: 2.0

• Effective batch size: 16

  • Micro-batch size: 16
  • Gradient accumulation steps: 1
  • Number of GPUs: 1

• Gradient checkpointing: True

• Prediction type: flow-matching (extra parameters=['shift=3'])

• Optimizer: adamw_bf16

• Trainable parameter precision: Pure BF16

• Caption dropout probability: 10.0%

• LoRA Rank: 768

• LoRA Alpha: 768.0

• LoRA Dropout: 0.1

• LoRA initialisation style: default

Datasets

linechart

• Repeats: 0
• Total number of images: 2822
• Total number of aspect buckets: 1
• Resolution: 1.048576 megapixels
• Cropped: True
• Crop style: center
• Crop aspect: square
• Used for regularisation data: No

Inference

import torch
from diffusers import DiffusionPipeline

model_id = 'stabilityai/stable-diffusion-3.5-large'
adapter_id = 'aryamankeyora/simpletuner-lora'
pipeline = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.bfloat16) # loading directly in bf16
pipeline.load_lora_weights(adapter_id)

prompt = "k4s4, linechart with 1 lines. Line 1: Line is noisy and trends decreasing at a decreasing rate then decreasing at a constant rate. Overall Description: Line is noisy and trends decreasing at a decreasing rate then decreasing at a constant rate."
negative_prompt = 'blurry, cropped, ugly'

## 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
image = pipeline(
    prompt=prompt,
    negative_prompt=negative_prompt,
    num_inference_steps=28,
    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=7.5,
).images[0]
image.save("output.png", format="PNG")