Koleshjr/dyu-fr-joeynmt-316-epochs_2_layers_8heads_128_384_plateau_2000_7_95_21_48

koleshjr/dyu-fr-joeynmt

An example of a machine translation model that translates Dyula to French using the JoeyNMT framework.

This following example is based on this Github repo that was kindly created by data354.

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Usage

Load and use for inference

import torch
from joeynmt.config import load_config, parse_global_args
from joeynmt.prediction import predict, prepare
from huggingface_hub import snapshot_download

# Download model
snapshot_download(
    repo_id="Koleshjr/dyu-fr-joeynmt-316-epochs_2_layers_8heads_128_384_plateau_2000_7_95_21_48",
    local_dir="/path/to/save/locally"
)

# Define model interface
class JoeyNMTModel:
    '''
    JoeyNMTModel which load JoeyNMT model for inference.

    :param config_path: Path to YAML config file
    :param n_best: return this many hypotheses, <= beam (currently only 1)
    '''
    def __init__(self, config_path: str, n_best: int = 1):
        seed = 42
        torch.manual_seed(seed)
        cfg = load_config(config_path)
        args = parse_global_args(cfg, rank=0, mode="translate")
        self.args = args._replace(test=args.test._replace(n_best=n_best))
        # build model
        self.model, _, _, self.test_data = prepare(self.args, rank=0, mode="translate")

    def _translate_data(self):
        _, _, hypotheses, trg_tokens, trg_scores, _ = predict(
            model=self.model,
            data=self.test_data,
            compute_loss=False,
            device=self.args.device,
            rank=0,
            n_gpu=self.args.n_gpu,
            normalization="none",
            num_workers=self.args.num_workers,
            args=self.args.test,
            autocast=self.args.autocast,
        )
        return hypotheses, trg_tokens, trg_scores

    def translate(self, sentence) -> list:
        '''
        Translate the given sentence.

        :param sentence: Sentence to be translated
        :return:
        - translations: (list of str) possible translations of the sentence.
        '''
        self.test_data.set_item(sentence.strip())
        translations, _, _ = self._translate_data()
        assert len(translations) == len(self.test_data) * self.args.test.n_best
        self.test_data.reset_cache()
        return translations

# Load model
config_path = "/path/to/lean_model/config_local.yaml" # Change this to the path to your model congig file
model = JoeyNMTModel(config_path=config_path, n_best=1)

# Translate
model.translate(sentence="i tɔgɔ bi cogodɔ")

Training procedure

Training hyperparameters

More information needed

Training results

More information needed

Framework versions

• JoeyNMT 2.3.0
• Torch 2.3.1+cu121