| """Publish a trained DINNNER checkpoint and its model code to the Hugging Face Hub. |
| |
| The checkpoint is checked against the model code (weights/dinnner_core.py by |
| default) first, then both are uploaded with a short model card. The repo is |
| private unless --public is given: keep it private until the paper is |
| published. Run it after `hf auth login` (or with an HF_TOKEN environment |
| variable that has write access): |
| |
| python scripts/upload_weights.py \ |
| --checkpoint path/to/checkpoint.pth \ |
| --repo faridan/dinnner-pubmedbert-bionlp13pc |
| |
| Leave out --checkpoint to update only the model code and the model card. |
| """ |
|
|
| import argparse |
| import os |
| import sys |
| import tempfile |
|
|
| sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__)))) |
|
|
| import torch |
| from huggingface_hub import HfApi |
|
|
| from dinnner.config import MODELS |
| from dinnner.model import load_model |
|
|
| CARD = """--- |
| license: {license} |
| language: en |
| library_name: pytorch |
| pipeline_tag: token-classification |
| tags: [biomedical, named-entity-recognition, ner, bionlp13pc, dinnner] |
| datasets: [bionlp13pc] |
| base_model: {encoder} |
| --- |
| |
| # DINNNER + PubMedBERT (BioNLP13PC) |
| |
| Weights for **DINNNER** (Distribution-Informed Neural Network for Named Entity |
| Recognition), trained on the BioNLP 2013 Pathway Curation corpus. |
| |
| The paper describing DINNNER is under review. Details of the method and |
| results will be added here once it is published. |
| |
| * Encoder: `{encoder}` |
| * Labels: {labels} |
| * Max subword length: {max_length} |
| |
| Try it online: {demo} |
| |
| The checkpoint is a plain PyTorch `state_dict` for the `DINNNER` class in |
| `dinnner_core.py`. |
| """ |
|
|
|
|
| def main(): |
| ap = argparse.ArgumentParser() |
| ap.add_argument("--checkpoint", help="local .pth file (omit to update only the code and card)") |
| ap.add_argument("--code", help="model code to upload (default: the model's local_code_path)") |
| ap.add_argument("--repo", required=True, help="e.g. your-username/dinnner-pubmedbert-bionlp13pc") |
| ap.add_argument("--model", default="pubmedbert-bionlp13pc", choices=list(MODELS)) |
| ap.add_argument("--demo", default="https://huggingface.co/spaces/faridan/dinnner") |
| ap.add_argument("--license", default="other", help="licence id for the card, e.g. mit or cc-by-4.0") |
| ap.add_argument("--public", action="store_true", help="make the repo public (after publication)") |
| args = ap.parse_args() |
|
|
| spec = MODELS[args.model] |
| code = args.code or spec.local_code_path |
| print("Checking the checkpoint against the model code...") |
| local = {"local_code_path": code, **({"local_path": args.checkpoint} if args.checkpoint else {})} |
| load_model(spec.__class__(**{**spec.__dict__, **local}), torch.device("cpu")) |
| print("OK") |
|
|
| api = HfApi() |
| api.create_repo(args.repo, repo_type="model", private=not args.public, exist_ok=True) |
| api.update_repo_settings(args.repo, private=not args.public) |
| if args.checkpoint: |
| print(f"Uploading {args.checkpoint} as {spec.weights_file} ...") |
| api.upload_file(path_or_fileobj=args.checkpoint, path_in_repo=spec.weights_file, repo_id=args.repo, |
| commit_message="Add DINNNER checkpoint") |
| api.upload_file(path_or_fileobj=code, path_in_repo=spec.code_file, repo_id=args.repo, |
| commit_message="Add DINNNER model code") |
|
|
| card = CARD.format(license=args.license, encoder=spec.encoder, max_length=spec.max_length, |
| demo=args.demo, labels=", ".join(spec.id2label.values())) |
| with tempfile.NamedTemporaryFile("w", suffix=".md", delete=False) as f: |
| f.write(card) |
| api.upload_file(path_or_fileobj=f.name, path_in_repo="README.md", repo_id=args.repo, |
| commit_message="Add model card") |
| print(f"Done: https://huggingface.co/{args.repo}") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
