同样的 token 预算，不同配方决定能力偏向——拆解 DoReMi、Data Mixing Laws 与 RegMix 三条配比优化路线。

Linear attention enables efficient long-context autoregressive decoding by compressing history into recurrent states, bu…

Reasoning language models have made substantial advances on a variety of complex tasks, yet their capabilities remain ov…

AI-generated videos often depict people within detailed scenes while leaving their small facial regions poorly resolved.…

Complex documents such as scientific papers contain rich information in tables, figures, and equations; however, existin…

双语育儿被焦虑裹挟得最深：错过「语言敏感期」就晚了？双语会让孩子说话更晚？外教和启蒙动画哪个管用？语言学与发展心理学的答案比营销号冷静得多——双语不会搞混孩子，敏感期的窗口远比传闻宽，而家庭里最重要的变量不是机构、不是教材，是语言「活着」。

!(https://www.gstatic.com/images/branding/googleg/1x/googleg_standard_color_128dp.png)Add as a preferredsource on Google…

LIMA 用 1000 条精选数据就调出了能打的模型，「数据质量压倒数量」从此有了实锤。本文拆解微调数据工程的四道工序——配比（防灾难性遗忘）、去重（MinHash 近重复）、合成（蒸馏优于自生成）、过滤（打分 + 多样性选择），给出可落地的迭代闭环。

Preference distillation typically treats a teacher response as preferred and the student's own response as rejected. Thi…

GGUF, GPTQ, AWQ, EXL2, and EXL3 solve the same problem in different ways. This guide separates file containers from quan…

Mixture-of-Experts (MoE) scales capacity, but existing designs cannot set three quantities independently. For a single t…

Nums AI has released Causilo, a pretrained tabular foundation model for classification and regression with a scikit-lear…

把 2026 年可核查的公开证据整理成一张六层能力路线图——预训练、后训练 RL、推理时计算、上下文与记忆、智能体与 Harness、世界模型。含 Meta ScaleRL 40 万 GPU 小时实验结论、RL 预算占比 10%–30% 口径、Chinchilla 对比、Meta-Harness 6x 差距等数据锚点，并给出优先级表与算法工程师/产品经理的行动建议。