| """Formal V4.2 LM+chat pretraining entry point; optional W&B tracking.""" |
| import argparse |
| from pathlib import Path |
| import sys |
| import tomllib |
| ROOT=Path(__file__).resolve().parents[2] |
| sys.path.insert(0,str(ROOT/'src')) |
|
|
|
|
| def main(): |
| parser=argparse.ArgumentParser(description=__doc__) |
| parser.add_argument('--config',type=Path,default=ROOT/'configs/train-v42.toml') |
| parser.add_argument('--resume',action='store_true') |
| parser.add_argument('--dry-run',action='store_true', |
| help='Verify configuration, code and data identities, then exit') |
| args=parser.parse_args() |
| config=tomllib.loads(args.config.read_text('utf-8')) |
| output=ROOT/config['output_dir'] |
| logs=ROOT/config['log_dir'] |
| if args.dry_run: |
| from vimeml.training.joint_v42 import build_plan |
| plan=build_plan(config,ROOT,verify_files=True) |
| print({'stage':plan['stage'],'parameters':plan['parameters'], |
| 'max_steps':plan['settings']['max_steps'], |
| 'updates_per_epoch':plan['settings']['updates_per_epoch'][:3]+['...'], |
| 'required_files':len(plan['required_files'])}) |
| return |
| if not args.resume and ((output.exists() and any(output.iterdir())) or (logs.exists() and any(logs.iterdir()))): |
| parser.error('Fresh V4.2 output/log directories must be empty') |
| if args.resume and not (output/'last.pt').exists(): |
| parser.error('No V4.2 checkpoint to resume') |
| from vimeml.training.joint_v42 import run |
| tracking=config.get('tracking',{}) |
| if not tracking.get('enabled'): |
| result=run(config,args.resume) |
| print(result,flush=True) |
| return |
| import wandb |
| directory=ROOT/'artifacts/tracking' |
| directory.mkdir(parents=True,exist_ok=True) |
| with wandb.init(project=tracking.get('project','vimeml'),name=output.name,group=output.name, |
| mode=tracking.get('mode','online'),dir=str(directory),sync_tensorboard=True, |
| save_code=False,force=True,config=config, |
| settings=wandb.Settings(disable_git=True,init_timeout=30)) as cloud: |
| from vimeml.training.data import write_json |
| write_json(directory/f'{output.name}-live.json',{'id':cloud.id,'url':cloud.url,'resume':args.resume}) |
| print(f'W&B run: {cloud.url}',flush=True) |
| result=run(config,args.resume) |
| cloud.summary.update({'training_result':result}) |
|
|
|
|
| if __name__=='__main__': |
| main() |
|
|
