train_scripts/load_data.py

import os
from datasets import load_dataset

def load_hf_jsonl_dataset(dataset_id: str, train_glob: str, val_file: str):
    """
    Loads JSONL files stored in a HF *dataset repo*.
    Example:
      dataset_id="Chamaka8/serendip-cpt-sinhala"
      train_glob="data/raw/train_shards/train_*.jsonl"
      val_file="data/raw/val.jsonl"
    """
    train_files = [f"hf://datasets/{dataset_id}/{train_glob}"]
    val_files = [f"hf://datasets/{dataset_id}/{val_file}"]

    ds_train = load_dataset("json", data_files={"train": train_files}, split="train")
    ds_val   = load_dataset("json", data_files={"validation": val_files}, split="validation")
    return ds_train, ds_val

def load_local_shards(train_glob: str, val_path: str):
    ds_train = load_dataset("json", data_files={"train": [train_glob]}, split="train")
    ds_val   = load_dataset("json", data_files={"validation": [val_path]}, split="validation")
    return ds_train, ds_val