example.py

"""Minimal end-to-end example: load the HF-hosted weights and translate.

Run from the parent project directory (so `src` is importable):
    python example.py --text "Hello world, how are you?"
"""

import argparse
import json
from pathlib import Path

import sentencepiece as spm
import torch

# Requires: src/ from https://github.com/Euswbnix/Machine_translation on the path
from src.model import Transformer
from src.inference.translate import batched_beam_search
from src.data.tokenizer import BOS_ID, EOS_ID, PAD_ID


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--weights", default="pytorch_model.bin")
    ap.add_argument("--spm", default="sentencepiece.model")
    ap.add_argument("--config", default="config.json")
    ap.add_argument("--text", required=True, help="English sentence to translate.")
    ap.add_argument("--beam", type=int, default=5)
    ap.add_argument("--device", default="cuda" if torch.cuda.is_available() else "cpu")
    args = ap.parse_args()

    cfg = json.loads(Path(args.config).read_text())

    model = Transformer(
        vocab_size=cfg["vocab_size"],
        d_model=cfg["d_model"],
        n_heads=cfg["n_heads"],
        n_encoder_layers=cfg["n_encoder_layers"],
        n_decoder_layers=cfg["n_decoder_layers"],
        d_ff=cfg["d_ff"],
        dropout=0.0,
        max_seq_len=cfg["max_seq_len"],
        share_embeddings=cfg["share_embeddings"],
        pad_idx=PAD_ID,
    ).to(args.device)
    model.load_state_dict(torch.load(args.weights, map_location=args.device))
    model.eval()

    sp = spm.SentencePieceProcessor()
    sp.load(args.spm)

    # Encode: wrap with BOS/EOS the same way the trainer does
    ids = [BOS_ID] + sp.encode(args.text, out_type=int) + [EOS_ID]
    src = torch.tensor([ids], dtype=torch.long, device=args.device)

    hyp_ids = batched_beam_search(
        model, src, beam_size=args.beam, max_len=cfg["max_seq_len"], length_penalty=1.0
    )[0]

    # Strip BOS/EOS and decode
    hyp_ids = [t for t in hyp_ids if t not in (BOS_ID, EOS_ID, PAD_ID)]
    print(sp.decode(hyp_ids))


if __name__ == "__main__":
    main()