NEW by DeepSeek: SPCT w/ DeepSeek-GRM-27B скачать в хорошем качестве

NEW by DeepSeek: SPCT w/ DeepSeek-GRM-27B

DeepSeek published a NEW learning method and a NEW model for the next generation of Reasoning models, called DeepSeek-GRM-27B. In this video I explain the new learning method "Self-Principled Critique Tuning" (SPCT) and its implementation in the new DeepSeek-GRM-27B. Maybe the core for the next DeepSeek R2? all rights w/ authors: Inference-Time Scaling for Generalist Reward Modeling Zijun Liu1,2, Peiyi Wang1, Runxin Xu1, Shirong Ma1, Chong Ruan1, Peng Li3, Yang Liu2,3, Yu Wu1 from 1DeepSeek-AI, 2 Dept. of Computer Sci. & Tech., Tsinghua University, 3 Institute for AI Industry Research (AIR), Tsinghua University #deepseek #airesearch #reward #aiexplained