RouteLLM (2024年)(ルートエルエルエム)
2024年6月UC Berkeley LMSYS発表RouteLLM・Industry-leading LLM router strong-weak + Industry-leading 85% cost reduction + Industry-leading preference-based routing + Industry-leading UC Berkeley LMSYS RouteLLM Apache 2.0 LLM routing framework 2024。
概要
RouteLLM は2024年6月UC Berkeley LMSYS発表・Industry-leading LLM router strong-weak 2024年UC Berkeley LMSYS RouteLLM Apache 2.0 LLM routing framework position確立。Industry-leading 85% cost reduction + preference-based routing + matrix factorization + BERT classifier + GPT-4 strong vs Mixtral-8x7B weak routing + 2024年6月 ArXiv paper。
主な特徴・仕組み
- Industry-leading UC Berkeley LMSYS (Sky Computing Lab) + Anyscale + Canva 2024年6月 release
- Industry-leading RouteLLM LLM routing framework signature
- Industry-leading LLM router strong-weak signature (GPT-4 + Mixtral-8x7B routing)
- Industry-leading 85% cost reduction signature (matched GPT-4 quality 95%)
- Industry-leading preference-based routing signature (Chatbot Arena Elo trained)
- Industry-leading matrix factorization + BERT classifier + causal LLM signature (3 router types)
- Industry-leading 1.65M+ Chatbot Arena preference data trained
- Industry-leading 2024年6月 ArXiv paper "RouteLLM Learning to Route LLMs with Preference Data"
- Industry-leading Apache 2.0 license signature
- Industry-leading GitHub native + Python SDK
- Industry-leading UC Berkeley LMSYS 2023→2024 Chatbot Arena creators
まとめ
RouteLLM = 2024年6月UC Berkeley LMSYS発表のpremium Apache 2.0 LLM routing framework。Industry-leading LLM router strong-weak GPT-4 + Mixtral-8x7B routing + 85% cost reduction matched GPT-4 quality 95% + preference-based routing Chatbot Arena Elo trained + matrix factorization + BERT classifier + causal LLM 3 router types + 1.65M+ Chatbot Arena preference data trained + 2024年6月 ArXiv paper + Apache 2.0 + GitHub + Python SDK + UC Berkeley LMSYS + Anyscale + Canva 2024年6月 release + LMSYS Chatbot Arena creators position確立。