Increase model size (n_embd, n_layer, n_head) for the multi-GPU configuration. Explicitly set AdamW betas to (0.9, 0.99).
Add train_multigpu.py for distributed data parallel training. Update train.py to save the training configuration to a JSON file. Generalize .gitignore to exclude all *.pt checkpoint files. Delete obsolete train_dpp.py file.
