Instructions to use TheLastBen/The_Hound with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use TheLastBen/The_Hound with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", dtype=torch.bfloat16, device_map="cuda") pipe.load_lora_weights("TheLastBen/The_Hound") prompt = "sandor clegane drinking in a pub" image = pipe(prompt).images[0] - Inference
- Notebooks
- Google Colab
- Kaggle
- Local Apps
- Draw Things
- DiffusionBee
Training details
What are the training details, you mentioned on reddit you used 10 images, but no other info.
where the images captioned? or was just the name in the caption?
what optimizer?
What LR?
What LR schedule?
How many steps did you train for?
Single or multiple resolutions?
8-bit Adam optimizer
LR between 1e-4 and 1e-3 depending on the number of steps
FlowMatchEulerDiscreteScheduler
1000 steps high lr => good output, but less styling capabilities
3000 steps low lr => good output, good styling capabilities
Single resolution 640px square
Thank you very much for providing the details!
FlowMatchEulerDiscreteScheduler Is that noise schedule or LR schedule?
If noise, What LR schedule did you use?
cosine, constant, etc?
Ah, yes, it's cosine
Hi. Sorry to ask but I got an "Error running job: Sampler FlowMatchEulerDiscreteScheduler not supported".
Is there something missing here ?
Thanks.
@cmatias FlowMatchEulerDiscreteScheduler is the noise scheduler for training