万卡集群真实部署，已节省数百万 GPU 小时！MoE 通信优化技术 COMET 开源
"训练效率" 相关文章
Moonshot AI and UCLA Researchers Release Moonlight: A 3B/16B-Parameter Mixture-of-Expert (MoE) Model Trained with 5.7T Tokens Using Muon Optimizer
出人意料！DeepSeek-R1用的GRPO其实非最优？规模化强化学习训练用PPO就够了
DeepSeek新论文再次引发热议，它说了什么？
Max Tegmark组新工作：利用调和损失训练可解释的AI模型
Process Reinforcement through Implicit Rewards (PRIME): A Scalable Machine Learning Framework for Enhancing Reasoning Capabilities
谢赛宁新作：表征学习有多重要？一个操作刷新SOTA，DiT训练速度暴涨18倍
关于 RWKV 架构的一些谣言
谢赛宁新作：表征学习有多重要？一个操作刷新SOTA，DiT训练速度暴涨18倍
Nvidia AI Introduces the Normalized Transformer (nGPT): A Hypersphere-based Transformer Achieving 4-20x Faster Training and Improved Stability for LLMs
Revisiting Recurrent Neural Networks RNNs: Minimal LSTMs and GRUs for Efficient Parallel Training
腾讯推出新一代大模型“混元Turbo”：性能大幅提升，定价低50%
Sparse Maximal Update Parameterization (SμPar): Optimizing Sparse Neural Networks for Superior Training Dynamics and Efficiency
