| |
| """Quick launcher for Pear1.5 hardware-agnostic training. |
| |
| Usage: |
| python launch.py --platform colab |
| python launch.py --platform kaggle-t4 |
| python launch.py --platform kaggle-tpu |
| python launch.py --platform local |
| python launch.py --platform all |
| """ |
| import argparse |
| import asyncio |
| import os |
| import sys |
| import subprocess |
| from pathlib import Path |
|
|
| |
| sys.path.insert(0, str(Path(__file__).parent)) |
|
|
| from hardware import get_hardware_info, get_training_config, print_hardware_summary |
|
|
|
|
| def launch_colab(notebook_path: str = None): |
| """Launch on Colab via browser.""" |
| url = "https://colab.research.google.com/drive/1BCLulJMl1xEErqRPbzn9caxH4wk56wBB" |
| print(f"Opening Colab: {url}") |
| print("1. Connect to runtime (Runtime -> Connect)") |
| print("2. Run the first cell to install dependencies") |
| print("3. Run the training cell") |
| |
| if sys.platform == "win32": |
| os.system(f'start "" "{url}"') |
| elif sys.platform == "darwin": |
| os.system(f'open "{url}"') |
| else: |
| os.system(f'xdg-open "{url}"') |
|
|
|
|
| def launch_kaggle(gpu_type: str = "t4"): |
| """Launch on Kaggle.""" |
| url = "https://www.kaggle.com/code/new" |
| print(f"Opening Kaggle: {url}") |
| print(f"1. Set accelerator to: {'TPU v5e-8' if gpu_type == 'tpu' else 'GPU T4 x2'}") |
| print("2. Add dataset: pear15-expanded-data") |
| print("3. Add HF_TOKEN secret") |
| print("4. Paste train_agnostic.py content into a cell and run") |
| |
| if sys.platform == "win32": |
| os.system(f'start "" "{url}"') |
| elif sys.platform == "darwin": |
| os.system(f'open "{url}"') |
| else: |
| os.system(f'xdg-open "{url}"') |
|
|
|
|
| def launch_lightning(): |
| """Launch on Lightning AI.""" |
| url = "https://lightning.ai/tk8679a/studios" |
| print(f"Opening Lightning AI: {url}") |
| print("1. Create new Studio with A100") |
| print("2. Open terminal") |
| print("3. Run: pip install -r requirements.txt") |
| print("4. Run: python train_agnostic.py --data-path /data --output-dir /checkpoints") |
| |
| if sys.platform == "win32": |
| os.system(f'start "" "{url}"') |
| elif sys.platform == "darwin": |
| os.system(f'open "{url}"') |
| else: |
| os.system(f'xdg-open "{url}"') |
|
|
|
|
| def launch_local(args): |
| """Launch training locally.""" |
| print("Launching local training...") |
| |
| cmd = [ |
| sys.executable, "scripts/train_agnostic.py", |
| "--data-path", args.data_path, |
| "--output-dir", args.output_dir, |
| "--resume-from", args.resume_from, |
| ] |
| |
| if args.max_steps: |
| cmd.extend(["--max-steps", str(args.max_steps)]) |
| if args.hf_model_path: |
| cmd.extend(["--hf-model-path", args.hf_model_path]) |
| if args.seed: |
| cmd.extend(["--seed", str(args.seed)]) |
| |
| print(f"Running: {' '.join(cmd)}") |
| subprocess.run(cmd, cwd=Path(__file__).parent.parent) |
|
|
|
|
| def launch_all(): |
| """Launch on all available platforms.""" |
| print("=" * 60) |
| print("LAUNCHING ON ALL PLATFORMS") |
| print("=" * 60) |
| |
| |
| hw = get_hardware_info() |
| config = get_training_config(hw) |
| print_hardware_summary(hw, config) |
| |
| |
| print("\n1. Starting local training...") |
| import subprocess |
| local_proc = subprocess.Popen([ |
| sys.executable, "scripts/train_agnostic.py", |
| "--data-path", "/content/pear15-expanded-data", |
| "--output-dir", "/content/checkpoints/local", |
| "--resume-from", "latest", |
| ], cwd=Path(__file__).parent.parent) |
| |
| |
| print("\n2. Opening Colab A100...") |
| launch_colab() |
| |
| |
| print("\n3. Opening Kaggle T4...") |
| launch_kaggle("t4") |
| |
| |
| print("\n4. Opening Kaggle TPU v5e-8...") |
| launch_kaggle("tpu") |
| |
| |
| print("\n5. Opening Lightning AI...") |
| launch_lightning() |
| |
| print("\n" + "=" * 60) |
| print("ALL PLATFORMS LAUNCHED") |
| print("=" * 60) |
| print("Monitor progress:") |
| print(" - Colab: Check notebook output") |
| print(" - Kaggle: Check notebook output") |
| print(" - Local: Check terminal output") |
| print(" - Checkpoints saved to respective output dirs") |
| print("\nPress Ctrl+C to stop local training") |
| |
| try: |
| local_proc.wait() |
| except KeyboardInterrupt: |
| local_proc.terminate() |
| print("Stopped local training") |
|
|
|
|
| def main(): |
| parser = argparse.ArgumentParser(description="Pear1.5 Hardware-Agnostic Training Launcher") |
| parser.add_argument("--platform", choices=["colab", "kaggle-t4", "kaggle-tpu", "lightning", "local", "all"], |
| default="local", help="Platform to launch on") |
| parser.add_argument("--data-path", default="/content/pear15-expanded-data", |
| help="Path to expanded dataset") |
| parser.add_argument("--output-dir", default="/content/checkpoints", |
| help="Output directory for checkpoints") |
| parser.add_argument("--resume-from", default="latest", |
| help="Resume from checkpoint (step, 'latest', or path)") |
| parser.add_argument("--max-steps", type=int, default=None, |
| help="Maximum steps to run") |
| parser.add_argument("--hf-model-path", default=None, |
| help="HF model path for fresh init") |
| parser.add_argument("--seed", type=int, default=0) |
| args = parser.parse_args() |
| |
| if args.platform == "colab": |
| launch_colab() |
| elif args.platform == "kaggle-t4": |
| launch_kaggle("t4") |
| elif args.platform == "kaggle-tpu": |
| launch_kaggle("tpu") |
| elif args.platform == "lightning": |
| launch_lightning() |
| elif args.platform == "local": |
| launch_local(args) |
| elif args.platform == "all": |
| launch_all() |
|
|
|
|
| if __name__ == "__main__": |
| main() |
