| """ |
| Download scGPT whole-human pretrained model. |
| Run on login node: python scripts/download_scgpt.py |
| """ |
|
|
| import os |
| import sys |
|
|
|
|
| def download_scgpt_model(output_dir: str = "transfer/data/scGPT_pretrained"): |
| """Download scGPT whole-human model files.""" |
| os.makedirs(output_dir, exist_ok=True) |
|
|
| required_files = ["best_model.pt", "vocab.json", "args.json"] |
| missing = [f for f in required_files if not os.path.exists(os.path.join(output_dir, f))] |
|
|
| if not missing: |
| print(f"All scGPT pretrained files already exist in {output_dir}") |
| return |
|
|
| print(f"Missing files: {missing}") |
| print(f"Please download the scGPT whole-human pretrained model to: {output_dir}") |
| print() |
| print("Option 1: From the scGPT release (if available):") |
| print(" Download from: https://github.com/bowang-lab/scGPT/releases") |
| print(f" Place files in: {output_dir}/") |
| print() |
| print("Option 2: Manual setup:") |
| print(" Required files:") |
| for f in required_files: |
| print(f" - {f}") |
| print() |
| print(" args.json example content:") |
| print(' {"embsize": 512, "nheads": 8, "d_hid": 512, "nlayers": 12,') |
| print(' "n_layers_cls": 3, "dropout": 0.2, "pad_token": "<pad>",') |
| print(' "pad_value": 0, "mask_value": -1}') |
|
|
|
|
| if __name__ == "__main__": |
| |
| repo_root = os.path.dirname(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))) |
| default_dir = os.path.join(repo_root, "data", "scGPT_pretrained") |
|
|
| import argparse |
| parser = argparse.ArgumentParser() |
| parser.add_argument("--output_dir", default=default_dir) |
| args = parser.parse_args() |
|
|
| download_scgpt_model(args.output_dir) |
|
|
