LLM-Rosetta: Hub-and-Spoke IR Solves O(N²) Cross-Provider LLM API Translation

arXiv·high signal

Ding observes that despite substantial syntactic divergence, major LLM APIs (OpenAI, Anthropic, Google, etc.) share a common semantic core — the real challenge is the combinatorial surface of syntactic variants requiring O(N²) bilateral adapters. LLM-Rosetta introduces an intermediate representation that reduces this to O(N), enabling portable multi-provider architectures. Directly relevant to anyone building provider-agnostic agent systems or multi-model orchestration layers.