TEMPO Integrates Time-Series Data via Discrete Tokenization

TEMPO extends LLM vocabularies using a context-aware Transformer tokenizer and finite scalar quantization to process time-series signals. Using a Qwen3-4B backbone, it achieves macro-F1 scores up to 89.0% on activity recognition and 86.0% on bearing-fault diagnosis.