parmarjatin4911@gmail.com𝐃𝐨𝐜𝐮𝐦𝐞𝐧𝐭 𝐑𝐞-𝐫𝐚𝐧𝐤𝐢𝐧𝐠 𝐢𝐬 𝐚𝐧 𝐢𝐦𝐩𝐨𝐫𝐭𝐚𝐧𝐭 𝐚𝐝𝐝𝐢𝐭𝐢𝐨𝐧 𝐭𝐨 𝐑𝐀𝐆 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞𝐬
The absence of a domain-specific Ranker can potentially diminish the quality of your initial ranked outputs. Additionally, obtaining sufficient training data for your ranker can be challenging in many cases ⬇️
This paper from Google Research proposes a 𝐏𝐚𝐢𝐫𝐰𝐢𝐬𝐞 𝐫𝐚𝐧𝐤𝐢𝐧𝐠 𝐩𝐫𝐨𝐦𝐩𝐭𝐢𝐧𝐠 (𝐏𝐑𝐏) 𝐩𝐚𝐫𝐚𝐝𝐢𝐠𝐦, which uses the query and a pair of documents as the prompt for LLMs to perform ranking tasks optimized to O(n) time 🤯
They also show that PRP based on the FLAN-UL2 model with 20B parameters is comparative with the previous best approach in the literature, based on the black box commercial GPT-4 that has (an estimated) 50X model size 😵
Interested? Then check the paper summary here https://lnkd.in/gKQbicDd
