LLM Vocabulary Compression for Low-Compute Environments