Fast training on H100-class GPUs. You're only charged while training is running. GPUs auto-stop when training jobs finish or fail.

OstrisAI-Toolkit

New Training Job

Job

Training Name

GPU ID

Trigger Word

Model

Model Architecture

Name or Path

Use a Hugging Face repo ID (e.g. owner/model-name).
⚠️ full URLs, .safetensors files, and local files are not supported.

Quantization

Transformer

Text Encoder

Target

Target Type

Linear Rank

Save

Data Type

Save Every

Max Step Saves to Keep

Training

Batch Size

Gradient Accumulation

Steps

Optimizer

Learning Rate

Weight Decay

Timestep Type

Timestep Bias

Loss Type

EMA (Exponential Moving Average)

Use EMA

Text Encoder Optimizations

Unload TE

Cache Text Embeddings

Regularization

Differential Output Preservation

Blank Prompt Preservation

Datasets

Dataset 1

Target Dataset

LoRA Weight

Num Repeats

Default Caption

Caption Dropout Rate

Settings

Cache Latents

Is Regularization

Flipping

Flip X

Flip Y

Resolutions

256

512

768

1024

1280

1536

Sample

Sample Every

Sampler

Guidance Scale

Sample Steps

Width

Height

Seed

Walk Seed

Advanced Sampling

Skip First Sample

Force First Sample

Disable Sampling

Sample Prompts (10)

Prompt

Width

Height

Seed

LoRA Scale

Prompt

Width

Height

Seed

LoRA Scale

Prompt

Width

Height

Seed

LoRA Scale

Prompt

Width

Height

Seed

LoRA Scale

Prompt

Width

Height

Seed

LoRA Scale

Prompt

Width

Height

Seed

LoRA Scale

Prompt

Width

Height

Seed

LoRA Scale

Prompt

Width

Height

Seed

LoRA Scale

Prompt

Width

Height

Seed

LoRA Scale

Prompt

Width

Height

Seed

LoRA Scale