Reg-DPO: SFT-Regularized Direct Preference Optimization with GT-Pair for Improving Video Generation
"训练稳定性" 相关文章
VL Norm：让强化学习更稳、更快的关键一步
VL Norm：让强化学习更稳、更快的关键一步
AMiD: Knowledge Distillation for LLMs with $\alpha$-mixture Assistant Distribution
Taming the Judge: Deconflicting AI Feedback for Stable Reinforcement Learning
小米 AI 新论文，雷军千万年薪要挖的 DeepSeek“天才少女”罗福莉署名
Stabilizing MoE Reinforcement Learning by Aligning Training and Inference Routers
Stabilizing MoE Reinforcement Learning by Aligning Training and Inference Routers
Stability of Transformers under Layer Normalization
AMAQ: Adaptive Mixed-bit Activation Quantization for Collaborative Parameter Efficient Fine-tuning
Why Low-Precision Transformer Training Fails: An Analysis on Flash Attention
Interactive Training: Feedback-Driven Neural Network Optimization
Stable Forgetting: Bounded Parameter-Efficient Unlearning in LLMs
A Unified Noise-Curvature View of Loss of Trainability
SimpleTIR：让大模型“边写代码边思考”不再崩溃
螞蟻國產GPU訓練大模型細節曝光！Ling模型研發負責人發文詳解背後故事
GAN归来：模型大幅简化，训练更稳定，逆袭扩散模型，AI社区疯传
Papers I’ve read this week, Mixture of Experts edition
Analyzing the Impact of Flash Attention on Numeric Deviation and Training Stability in Large-Scale Machine Learning Models
