| """Load the public eight-class poem CNN and classify a UTF-8 poem text file. |
| |
| python scripts/load_hf_model.py --poem poem.txt --revision FULL_HF_COMMIT |
| """ |
|
|
| import argparse |
| from pathlib import Path |
|
|
| import torch |
| from huggingface_hub import HfApi |
| from transformers import AutoModelForSequenceClassification, AutoTokenizer |
|
|
|
|
| def main(argv=None): |
| parser = argparse.ArgumentParser(description=__doc__) |
| parser.add_argument("--repo_id", default="sprx7767/chandassu-recognition") |
| parser.add_argument("--revision", help="Immutable HF commit; if omitted, resolve main once and pin both loaders") |
| parser.add_argument("--subfolder", help="Optional explicit package: cnn-eight-class") |
| parser.add_argument("--poem", type=Path, required=True, help="Complete newline-separated poem in UTF-8") |
| parser.add_argument("--device", choices=["cpu", "cuda", "mps"], default="cpu") |
| args = parser.parse_args(argv) |
| revision = args.revision or HfApi().model_info(args.repo_id, token=False).sha |
| if len(revision) != 40 or any(c not in "0123456789abcdef" for c in revision): |
| parser.error("--revision must be a complete lowercase hexadecimal HF commit ID") |
| options = dict(revision=revision, trust_remote_code=True, token=False) |
| if args.subfolder: |
| options["subfolder"] = args.subfolder |
| tokenizer = AutoTokenizer.from_pretrained(args.repo_id, **options) |
| model = AutoModelForSequenceClassification.from_pretrained(args.repo_id, **options).to(args.device).eval() |
| if getattr(model.config, "prediction_unit", None) != "poem": |
| parser.error("This script requires the eight-class whole-poem model; original four-class models classify lines") |
| inputs = tokenizer(args.poem.read_text(encoding="utf-8"), padding=True, return_tensors="pt").to(args.device) |
| with torch.inference_mode(): |
| scores = model(**inputs).logits.double().softmax(-1)[0] |
| print(f"Hugging Face revision: {revision}") |
| print(f"Prediction: {model.config.id2label[int(scores.argmax())]} (uncalibrated score)") |
| for i in scores.argsort(descending=True).tolist(): |
| print(f"{model.config.id2label[i]}: {float(scores[i]):.4%}") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
