Set PYTORCH_CUDA_ALLOC_CONF=expandable_segments to fight fragmentation; explicit low_cpu_mem_usage=False on TAEF1 + add accelerate to requirements to silence the warning

Update requirements.txt

Use upstream environment.yml pip deps verbatim (avoid piecemeal additions)

Add boto3/botocore (imaginaire easy_io imports them eagerly)

Add wandb (imaginaire utils import it eagerly)

Initial PiD + Z-Image step-by-step denoising demo for ZeroGPU