Refactor and Optimize DeepSeek Coder Training Script

[Imadnajam]

---

Key Changes

1. Improved Tokenization and Data Preprocessing

We created a clear and modular tokenization function _tokenize_fn, which processes the input strings by padding and truncating them appropriately.

def _tokenize_fn(strings: Sequence[str], tokenizer: transformers.PreTrainedTokenizer) -> Dict:
    tokenized_list = [
        tokenizer(
            text,
            return_tensors="pt",
            padding="longest",
            max_length=tokenizer.model_max_length,
            truncation=True,
        )
        for text in strings
    ]