You're looking at a specific version of this model. Jump to the model overview.
lightweight-ai /q_l_t:a3ab991d
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 |
|---|---|---|---|
| batch_size |
None
|
1
|
Batch size
|
| optimizer |
None
|
adamw8bit
|
Optimizer
|
| content_or_style |
None
|
balanced
|
Content/style weighting
|
| train_dtype |
None
|
bf16
|
Training dtype
|
| arch |
string
|
qwen_image
|
Model arch key for ai-toolkit
|
| seed |
integer
|
42
|
Random seed (None=random)
|
| steps |
integer
|
3000
Min: 100 Max: 6000 |
Training steps
|
| device |
string
|
cuda
|
Device for ai-toolkit trainer
|
| dataset |
string
|
ZIP with images and optional .txt captions (same basenames)
|
|
| ema_use |
boolean
|
False
|
Use EMA
|
| ema_decay |
number
|
0.99
Max: 0.9999 |
EMA decay
|
| lora_conv |
integer
|
16
Max: 256 |
LoRA conv rank
|
| base_model |
string
|
/home/freek/Qwen-Image
|
Base model path or HF ID
|
| save_every |
integer
|
250
Min: 1 Max: 10000 |
Checkpoint/sample every N steps
|
| caption_ext |
string
|
txt
|
Caption file extension
|
| lokr_factor |
integer
|
-1
|
LoKr factor (-1 = auto)
|
| lora_linear |
integer
|
32
Min: 1 Max: 256 |
LoRA linear rank
|
| sample_every |
integer
|
250
Max: 10000 |
Sample every N steps
|
| sample_steps |
integer
|
25
Min: 1 Max: 200 |
Sampling steps
|
| sample_width |
integer
|
1024
Min: 64 Max: 2048 |
Sample width
|
| weight_decay |
number
|
0.0001
Max: 1 |
Weight decay
|
| learning_rate |
number
|
0.0002
Min: 0.00001 Max: 0.001 |
Learning rate
|
| sample_height |
integer
|
1024
Min: 64 Max: 2048 |
Sample height
|
| timestep_type |
string
|
sigmoid
|
Timestep shaping
|
| guidance_scale |
number
|
4
Max: 20 |
Guidance scale
|
| lokr_full_rank |
boolean
|
True
|
Use full-rank LoKr
|
| sqlite_db_path |
string
|
./aitk_db.db
|
SQLite DB path for trainer
|
| default_caption |
string
|
|
Default caption for images without .txt
|
| lora_conv_alpha |
integer
|
16
Max: 1024 |
LoRA conv alpha
|
| noise_scheduler |
string
|
flowmatch
|
Noise scheduler
|
| lora_linear_alpha |
integer
|
32
Min: 1 Max: 1024 |
LoRA linear alpha
|
| resolution_list_csv |
string
|
512,768,1024
|
Comma-separated short-side resolutions
|
| caption_dropout_rate |
number
|
0.05
Max: 1 |
Caption dropout rate
|
| gradient_accumulation |
integer
|
1
Min: 1 Max: 64 |
Gradient accumulation
|
| gradient_checkpointing |
boolean
|
True
|
Enable gradient checkpointing
|
| ignore_if_contains_csv |
string
|
|
Comma-separated module substrings to ignore for LoRA
|
Output schema
The shape of the response you’ll get when you run this model with an API.
{'properties': {'weights': {'format': 'uri',
'title': 'Weights',
'type': 'string'}},
'required': ['weights'],
'title': 'Output',
'type': 'object'}