在慢速电脑上运行744B参数的GLM-5.2模型，实测不同精度下多token预测头的性能表现。
A few days ago I found myself trying out GLM 5.2 and was really positively impressed. The capabilities and security I was getting from this LLM are si…
熵引导多token预测方法，加速LLM推理并提升生成质量。
arXiv:2606.27550v1 Announce Type: cross Abstract: Multi-token prediction has been shown to increase data density during training, improve downstream t…
美团原生多模态LongCat-Next，用统一Token预测让AI看懂、听懂、说话，不再拼接模块。
LongCat-Next 是我们在通往物理世界 AI 道路上的一次探索。今天，我们把研究思路的核心——LongCat-Next 模型和它的离散分词器全部开源，希望更多开发者能基于它，构建真正能感知、理解并作用于真实世界的 AI。
新型LLM推理加速方法，成对输入输出实现潜在多token预测，显著提升效率。
arXiv:2605.27255v1 Announce Type: cross Abstract: Long chain-of-thought reasoning has made autoregressive decoding the dominant inference cost of mode…
探索LLM核心迷思：所有输出真的只是概率分布游戏吗？Hacker News上的一场硬核讨论。
Sorry I know this has been discussed to death sort of, but I feel like it's not always explicitly stated in some discussions and people have different…
提出NITP隐式Token预测新方法，革新LLM预训练范式，已被ICML 2026接收。
arXiv:2605.24956v1 Announce Type: new Abstract: Standard next-token prediction (NTP) supervises language models solely through discrete labels in the …