import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("RogerYao/stca-sd35-warmup-7000", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]STCA SD3.5 Medium — SFT Warmup Checkpoint (step 7000)
Fine-tuned transformer weights from Stable Diffusion 3.5 Medium, after supervised Fine-T2I warmup used by the STCA (Semantic-Temporal Credit Assignment) pipeline.
| Field | Value |
|---|---|
| Base | stabilityai/stable-diffusion-3.5-medium |
| Component | transformer only (~9.3 GiB) |
| Step | 7000 (Fine-T2I SFT warmup) |
| Format | safetensors + Diffusers config.json |
Load
from diffusers import SD3Transformer2DModel, StableDiffusion3Pipeline
import torch
repo = "RogerYao/stca-sd35-warmup-7000"
transformer = SD3Transformer2DModel.from_pretrained(
repo, subfolder="transformer", torch_dtype=torch.bfloat16
)
pipe = StableDiffusion3Pipeline.from_pretrained(
"stabilityai/stable-diffusion-3.5-medium",
transformer=transformer,
torch_dtype=torch.bfloat16,
)
For STCA-release RL training, set:
SFT_WARMUP_PATH=<local_dir>/transformer
# or download:
# hf download RogerYao/stca-sd35-warmup-7000 --local-dir outputs/warmup/checkpoint-7000
License
Derived from Stability AI SD3.5 Medium. You must accept the Stability Community License before use. Redistribution follows that license.
- Downloads last month
- -
Model tree for RogerYao/stca-sd35-warmup-7000
Base model
stabilityai/stable-diffusion-3.5-medium