Unverified50% confidenceFactExact time
随着MoE架构的普及，激活参数量与总参数量的分离使得FLOP/s阈值失去准确性，DeepSeek-V3训练消耗算力远低于其参数规模所暗示的量级
1
Sources
50%
Confidence
Long-term
Relevance
7/25/2026
First Seen
Sources
中国拟限制顶尖AI模型出海，DeepSeek同步布局自研推理芯片
bilibiliAI-seeker7/7/2026
Related Claims
Unverified算法效率的快速提升使得同等能力的模型所需计算量下降，以FLOP为唯一阈值的监管标准可能很快失去精准性75% similarUnverifiedThe application of the Mixture of Experts (MoE) architecture has broken the traditional trade-off between parameter scale and inference speed.72% similarUnverifiedThe core idea of PEFT is to not update all parameters, but instead introduce a small number of trainable adapter parameters (such as LoRA's low-rank matrix decomposition), reducing VRAM and computation requirements while retaining performance close to full fine-tuning.71% similarUnverified低精度引入的数值扰动会在RL的多轮迭代中通过策略梯度的乘积形式被放大，可能导致训练崩溃或模型性能退化70% similarUnverified稀疏激活架构下推理阶段的FLOPs与激活参数规模线性相关而非与总参数规模相关，从而降低单次推理的计算资源消耗和延迟70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/611318API
curl https://kongchang.com/api/v1/knowledge/claims/611318MCP
get_claim(id=611318)