API Reference¶

Welcome to the TorchLingo API reference. This section documents all public classes, functions, and modules.

Module Overview¶

TorchLingo is organized into four main modules:

Config

Configuration management for hyperparameters, paths, and feature toggles.

Config Reference
Data Processing

Dataset classes, vocabulary builders, and data loaders.

Data Processing Reference
Models

Neural network architectures for sequence-to-sequence translation.

Models Reference
Preprocessing

Data loading, tokenization, and preprocessing utilities.

Preprocessing Reference

Quick Links¶

Most Used Classes¶

Class	Description
`Config`	Centralized configuration
`NMTDataset`	PyTorch dataset for parallel text
`SimpleVocab`	Simple vocabulary builder
`MeCabVocab`	Japanese morphological analysis
`JiebaVocab`	Chinese word segmentation
`SimpleTransformer`	Transformer encoder-decoder
`SimpleSeq2SeqLSTM`	LSTM encoder-decoder

Most Used Functions¶

Function	Description
`get_default_config()`	Get default configuration
`load_data()`	Load data from file
`collate_fn()`	Batch collation with padding
`create_dataloaders()`	Create train/val data loaders

Import Patterns¶

Recommended Imports¶

# Configuration
from torchlingo.config import Config, get_default_config

# Data processing
from torchlingo.data_processing import (
    NMTDataset,
    SimpleVocab,
    SentencePieceVocab,
    MeCabVocab,
    JiebaVocab,
    collate_fn,
    create_dataloaders,
)

# Models
from torchlingo.models import SimpleTransformer, SimpleSeq2SeqLSTM

# Preprocessing
from torchlingo.preprocessing import load_data, save_data, parallel_txt_to_dataframe

Full Module Import¶

import torchlingo

# Access submodules
cfg = torchlingo.config.get_default_config()
model = torchlingo.models.SimpleTransformer(...)

Conventions¶

Parameter Fallback¶

Most functions accept both explicit parameters and a config object. The priority is:

Explicit parameter (if provided)
Config object value (if provided)
Default config value

# These are equivalent:
dataset = NMTDataset("data.tsv", max_length=100)
dataset = NMTDataset("data.tsv", config=Config(max_seq_length=100))

Type Annotations¶

All public APIs are fully type-annotated:

def encode(self, sentence: str, add_special_tokens: bool = True) -> List[int]:
    ...

Docstring Style¶

We use Google-style docstrings:

def function(param1: int, param2: str) -> bool:
    """Short description.

    Longer description if needed.

    Args:
        param1: Description of param1.
        param2: Description of param2.

    Returns:
        Description of return value.

    Raises:
        ValueError: When something is wrong.
    """

Version Compatibility¶

Python: 3.10+
PyTorch: 2.0+
Pandas: 1.5+

Getting Help¶

📖 Check the Concepts section for explanations
🎓 Follow the Tutorials for step-by-step guides
🐛 Report issues on GitHub