Group Relative Policy Optimization (GRPO) eliminates the value model used in Proximal Policy Optimization (PPO), reducing active memory footprint from 42 GB to 28 GB for a 7B parameter model in FP16 b…
DeepSeek-AI released DeepSpec, an open-source codebase for training and evaluating draft models for speculative decoding, supporting three draft model algorithms (DSpark, DFlash, Eagle3) and requiring…
DeepSeek-AI published DeepSeek-V3.2, an open large language model featuring Sparse Attention that reduces computational complexity from O(L²) to O(Lk). An interactive visual explainer of the paper is …
DeepSeek-AI published DeepSeek-OCR: Contexts Optical Compression on arXiv in 2025, achieving ~10× context compression with 97% decode accuracy by rendering long text as images. An interactive visual e…