Extend the fix to v3 and LoRA training scripts: - s2_train_v3.py: skip dist.init_process_group() + DummyDDP for Windows single-GPU - s2_train_v3_lora.py: same fix applied to LoRA fine-tuning script
install.sh