Zeroth-order fine-tuning runs 8x faster using inference infrastructure

Original: LLM Zeroth-Order Fine-Tuning is an Inference Workload

Source: arxiv.org ↗