import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("heli-stand/image-rm-sd3.5-medium-pickapicv2", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Selected RM Checkpoint
This is the selected Image RM checkpoint trained on liuhuohuo2/pick-a-pic-v2, using stabilityai/stable-diffusion-3.5-medium as the backbone.
Configuration
- Checkpoint:
checkpoint-final - T5 maximum sequence length:
256 - Learning rate:
3e-5 - Guidance scale:
5 - Effective batch size:
128 - Resolution:
512 - Seed:
42
Final Validation Metrics
| Split | Accuracy | Loss | Reward margin |
|---|---|---|---|
eval |
0.6993 | 0.6054 | 0.6142 |
eval_unique |
0.6573 | 0.5874 | 0.5162 |
eval was used as the primary selection metric and eval_unique as the complementary metric. This run provided the best overall balance among the completed T5-256 RM runs, with only mild late drift in eval loss and no late rise in eval_unique loss.
Use this checkpoint for reward labeling and reward-based alignment.
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support