- Added model tagging functionality to `runmps.sh`, allowing for dynamic model tagging based on the W&B run name.
- Updated `base_train.py`, `mid_train.py`, and `chat_sft.py` to utilize model tags for checkpoint management.
- Enhanced `base_eval.py` to accept model tags for loading models during evaluation.
- Improved handling of model tags to ensure proper checkpoint directory naming and logging.
- Introduced `kv_head_mult` to control the number of query heads sharing a key/value head in `base_train.py`, `mid_train.py`, and `runmps.sh`.
- Updated logging to include global token per second metrics during training.
- Added assertions to ensure `kv_head_mult` is valid and properly integrated into model calculations.
- Added `dev/runmps_evals.sh` for evaluating checkpoints and logging results to W&B.
- Introduced `dev/runmps.sh` for orchestrating training stages with W&B support.
- Updated `.gitignore` to include `wandb/` and `.runmps_wandb_ids`.
- Changed permissions for `dev/runcpu.sh` and added executable flag.
- Enhanced existing scripts to log metrics to W&B during training and evaluation processes.
