Learning to Translate from Soft to Hard LLM Prompts

In three linesMethod to translate soft prompts into natural language prompts using a dedicated translation model. Translations outperform InSPEcT across multiple benchmarks. Application: soft prompts optimized on small open-source models convert to portable text prompts that exceed original performance when deployed on closed-API models.

Summary generated by Claude — human-verified