You're looking at a specific version of this model. Jump to the model overview.
simpletuner /krea2:dd3015cd
Input schema
The fields you can use to run this model with an API. If you don’t give a value for a field its default value will be used.
| Field | Type | Default value | Description |
|---|---|---|---|
| train_data |
string
|
Zip or tar of training images. Matching .txt captions are used when present; missing captions use filename fallbacks.
|
|
| hf_dataset |
string
|
Hugging Face dataset repo ID to use when no archive is provided (e.g., user/dataset).
|
|
| pixel_area |
integer
|
1024
Min: 256 Max: 2048 |
Pixel-area resolution; also applied to min/max/target size and validation resolution.
|
| enable_crop |
boolean
|
True
|
Enable cropping when preparing samples.
|
| crop_style |
None
|
center
|
Crop location style when cropping is enabled.
|
| crop_aspect |
None
|
square
|
Aspect handling for crops: square (1:1) or preserve source aspect.
|
| trigger_word |
string
|
TOK
|
Subject token used in generated fallback captions and instance prompts.
|
| caption_strategy |
None
|
textfile
|
Caption strategy for uploaded archives. textfile uses matching .txt captions and generated filename fallbacks; filename derives captions from filenames; instanceprompt uses one generic prompt for every image.
|
| train_batch_size |
integer
|
1
Min: 1 Max: 4 |
Training batch size (1-4).
|
| learning_rate |
number
|
0.0001
Min: 0.0000001 Max: 0.01 |
LoRA learning rate.
|
| checkpoints_total_limit |
integer
|
3
Min: 1 Max: 5 |
Maximum checkpoints to retain (max 5).
|
| checkpoint_epoch_interval |
integer
|
5
Min: 1 |
Write a checkpoint every N epochs.
|
| num_train_epochs |
integer
|
25
Min: 1 |
Total training epochs.
|
| max_train_steps |
integer
|
0
|
Optional max train steps override (0 disables the step limit).
|
| checkpoint_step_interval |
integer
|
0
|
Save checkpoints and validate every N optimizer updates; 0 uses epoch intervals.
|
| hub_model_id |
string
|
Optional Hugging Face model repo for the LoRA, checkpoints, and validation images.
|
|
| hf_instance_prompt |
string
|
Use this prompt for every Hugging Face image instead of its caption column.
|
|
| hf_caption_column |
string
|
caption
|
Caption column for a Hugging Face dataset.
|
| hf_revision |
string
|
Optional Hugging Face dataset revision.
|
|
| validation_prompt |
string
|
Prompt used to generate validation images.
|
|
| lora_rank |
integer
|
Min: 1 Max: 256 |
Override the adapter rank; alpha matches rank.
|
| lora_dropout |
number
|
0
Max: 0.999 |
Adapter dropout probability.
|
| seed |
integer
|
42
|
Training and validation seed.
|
| attention_mechanism |
None
|
kohaku-fa-auto
|
Automatic Kohaku attention or native Diffusers attention.
|
| gradient_checkpointing_interval |
integer
|
Min: 1 Max: 64 |
Blocks per checkpointed segment; blank uses the model recipe.
|
| gradient_checkpointing_segment_stride |
integer
|
Min: 1 Max: 64 |
Start a checkpointed segment every N blocks; blank uses contiguous segments.
|
| dataloader_prefetch |
boolean
|
True
|
Prepare upcoming training batches in the background.
|
| push_checkpoints_to_hub |
boolean
|
True
|
Upload intermediate checkpoints when hub_model_id is set.
|
| enable_regional_compile |
boolean
|
True
|
Compile repeated transformer blocks.
|
| s3_bucket |
string
|
|
S3-compatible bucket for publishing checkpoints.
|
| s3_region |
string
|
S3 region (optional).
|
|
| s3_endpoint_url |
string
|
Custom S3 endpoint URL (for non-AWS providers).
|
|
| s3_base_path |
string
|
Prefix inside the bucket (defaults to cog/krea2/{job_id}).
|
|
| s3_public_base_url |
string
|
Public base URL to build shareable links (optional).
|
|
| s3_access_key |
string
|
S3 access key (leave blank to use IAM/instance roles).
|
|
| s3_secret_key |
string
|
S3 secret key (leave blank to use IAM/instance roles).
|
|
| hf_token |
string
|
Hugging Face token for model or dataset access.
|
|
| return_logs |
boolean
|
True
|
Print the tail of debug.log.
|
Output schema
The shape of the response you’ll get when you run this model with an API.
{'format': 'uri', 'title': 'Output', 'type': 'string'}