import os

import torch

DATA_DIR = "data"
TOKENIZER_PATH = "tokenizer.json"
CHECKPOINT_DIR = "checkpoints"
TRAIN_TOKENS_CACHE = "data/train_tokens.npy"
VAL_TOKENS_CACHE = "data/val_tokens.npy"

DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
DTYPE = torch.float16 if DEVICE == "cuda" else torch.float32


def load_dotenv(path=".env"):
    """Tiny dependency-free .env loader. Populates os.environ (does not overwrite
    variables already set in the real environment). Lets HF_TOKEN live in a
    gitignored .env file instead of being hardcoded or pasted into the shell."""
    if not os.path.exists(path):
        return
    with open(path, encoding="utf-8") as f:
        for line in f:
            line = line.strip()
            if not line or line.startswith("#") or "=" not in line:
                continue
            key, _, value = line.partition("=")
            key, value = key.strip(), value.strip().strip('"').strip("'")
            os.environ.setdefault(key, value)


load_dotenv()

# HuggingFace token (optional for public datasets like wikitext-103; used if present).
HF_TOKEN = os.environ.get("HF_TOKEN") or os.environ.get("HUGGINGFACE_TOKEN")