toilaluan

New Experiment: intermediate semantic prediction

The baseline JiT predicts clean pixels (x_0) directly. REPA adds an intermediate DINO patch-alignment loss, but its projected features are not passed to later denoising blocks. Standard REG jointly denoises a DINO token, yet predicts its clean state only after the final block.

New Experiment instead jointly diffuses 17 DINOv3 tokens (CLS + pooled 4×4 spatial tokens) and predicts their clean semantic (x_0) after block 4. The clean CLS prediction conditions the remaining blocks, while a wide CLS latent is compressed into four continuation tokens. This

toilaluan

Update - March 26

For Validators

• Generation Volume Setting for Miners: Validators can now handle more synthetic requests in 10 minutes as miners have the option to set their generation volume.
• Integration with Weights & Biases: Validators can now log miner responses to Weights & Biases. To enable this feature, add the WANDB_API_KEY from your Weights & Biases account settings (https://wandb.ai/settings#api) and set the use_wandb argument to true. Example command to start the validator with Weights & Biases logging enabled:

WANDB_API_KEY=YourWandBApiKey pm2 start python --name "validator_nicheimage" \
-- -m neurons.validator.validator \
--netuid  \

toilaluan

toilaluan

toilaluan

toilaluan

toilaluan

toilaluan

import glob
import random
import numpy as np
import cv2
import albumentations as A
import random
import cv2


class DrawHorizontal: