Introduces continue_training.sh to automatically resume interrupted training stages by detecting existing checkpoints and proceeding as needed. Updates README_MACOS.md with instructions and troubleshooting for using the new script, including manual continuation steps and improved guidance for memory, architecture, and performance issues.
Added the --device_batch_size argument to the base_loss evaluation command in runmac_overnight.sh to ensure batch size is configurable during evaluation.
Introduces automatic memory detection and batch size optimization for Apple Silicon Macs in runcpu.sh and runmac_overnight.sh scripts. Adds a comprehensive README_MACOS.md with usage instructions, performance profiles, environment variable overrides, troubleshooting, and expected training times. Updates scripts to allow manual overrides and improve usability for various Mac configurations. Also switched python to arm64 for 2-3x improvement
Added dev/runmac_overnight.sh for optimized Mac training. Updated device-specific logic throughout dataloader, GPT, Muon optimizer, and training scripts to avoid CUDA-only features on MPS/CPU (e.g., torch.compile, pin_memory, non_blocking, bfloat16). Relaxed torch version constraints in pyproject.toml and removed Linux/CUDA-specific PyTorch config for better macOS support.
