1
小鹏第二代 VLA Lite 蒸馏版 9 月面向单图灵芯片 Max 车型推送，G9L Max 版首发搭载
IT之家 8 月 27 日消息，在今日的小鹏物理 AI 分享暨第二代 VLA 全新版本体验日活动中，小鹏汽车宣布第二代 VLA Lite 蒸馏版将于 9 月开启推送 ，面向单图灵芯片的 Max 车型。 小鹏 G9L Max 版本 将首发搭载图灵智驾第二代 VLA Lite 蒸馏版。 小鹏汽车表示，第…
2
SelFusion: Self-distillation for Diffusion Language Models
扩散语言模型的自蒸馏新方法登上ACL 2026，生成效率与质量或迎来新突破。
arXiv:2608.22898v1 Announce Type: new Abstract: Diffusion language models (DLMs) alleviate the inherent latency bottleneck of autoregressive (AR) larg…
3
Thinking at the Right Size: Amortized Distillation Across Post-Trained LLMs
不同尺寸模型间摊销蒸馏，让推理成本与思考深度灵活匹配，EMNLP 2026 前沿成果
arXiv:2608.22854v1 Announce Type: new Abstract: Practical deployment of large language models (LLMs) requires families of post-trained variants---inst…
4
Knowing but Not Saying: Preventing Factual Access Failures in LLM SFT via Recall-Anchored Distillation
大模型微调后“知道却说不出口”的难题，用召回锚定蒸馏可精准破解。
arXiv:2608.20794v1 Announce Type: new Abstract: Supervised fine-tuning (SFT) can degrade factual behavior outside the target domain. This degradation …
5
Beyond FLOPs: Energy-Aware Knowledge Distillation for Sustainable LLMs on Code-Related Task
不只盯算力，用能源感知蒸馏为代码大模型瘦身，兼顾性能与可持续性。
arXiv:2608.17515v1 Announce Type: cross Abstract: Background: Large Language Models (LLMs) are increasingly being applied to Software Engineering (SE)…
6
The Distributional View of Knowledge Distillation
用分布视角重新拆解知识蒸馏，带你理解模型压缩背后的数学本质，适合深度学习进阶读者。
arXiv:2608.15215v1 Announce Type: cross Abstract: Token-level knowledge distillation (KD) matches two conditional distributions per position, yet the …
7
Mitigating Rubric Interference in LLM Judges via On-Policy Self-Distillation
单次推理评估多细则易受干扰，新方法用自蒸馏提升LLM裁判准确率与效率
arXiv:2608.14684v1 Announce Type: cross Abstract: LLM judges increasingly evaluate responses against fine-grained rubric checklists. When a sample req…
8
蒸馏《基层中国的运行逻辑》后，让 AI 帮你分析求学、考公、投资与创业
把《基层中国的运行逻辑》炼成 AI 方法论工具箱，帮你看懂地方权力结构，求学考公投资创业都有参考。
这个项目，把 @聂辉华 老师写的《基层中国的运行逻辑》这本书总结成了一个 skill，它把县乡村治理框架（条块、含权量、双均衡、三座大山…）提炼成可被 Cursor / Claude Code / Codex / Grok 反复调用的方法论工具箱，用来解释地方新闻与权力结构，也用来做选择：求学、考公
9
A New Trick Reveals AI Models’ Inner Thoughts
新技巧让AI的“内心独白”无所遁形，还疑似抓到模型间蒸馏痕迹
Researchers devised a way to extract “reasoning traces” from Claude, GPT, and Gemini. What they found, they say, indicates that some Chinese AI may be…
10
Out-of-Distribution Federated Distillation with Domain-Aware Proxy
联邦蒸馏遇上数据分布偏移？域感知代理方案给出稳健新解，跨域协作更可靠。
arXiv:2608.08525v1 Announce Type: new Abstract: Federated Learning is a distributed machine learning paradigm that trains a global model by aggregatin…
11
On-Policy Delta Distillation for Multilingual Math Reasoning
新方法让大模型多语言数学推理更强，策略蒸馏带来显著提升，值得一看。
arXiv:2608.05802v1 Announce Type: cross Abstract: On-Policy Distillation (OPD) is emerging as a promising alternative to reinforcement learning for LL…
12
On-Policy Self-Distillation without Any Supervision
没有外部监督的在线自蒸馏方案，为LLM后训练省去人工标注与奖励模型依赖。
arXiv:2608.06296v1 Announce Type: new Abstract: On-policy (Self-)Distillation (OPD / OPSD) has shown strong potential for post-training large language…
13
消息称张一鸣内部下死命令：就算模型暂时落后，字节跳动也不会依赖 AI 蒸馏技术
张一鸣定调：字节不靠蒸馏追赶，宁愿暂时落后也要自研大模型
IT之家 8 月 6 日消息，据 The Information 昨日（8 月 5 日）报道，字节跳动创始人张一鸣在上个月的 Seed 团队全体会议上明确表示，即使意味着暂时落后于竞争对手， 公司也不会依赖 AI 蒸馏技术来改进其模型 。 IT之家注：“蒸馏”指的是利用更先进的前沿模型（通常由其他公…
14
Efficient Knowledge Distillation for LLMs: Offline Top-K Logits and a Fused Chunked KL Loss
用离线Top-K logits和融合分块KL损失，大幅降低大模型蒸馏成本，兼顾效率与质量。
arXiv:2608.03796v1 Announce Type: cross Abstract: Small language models are often the only option for deployment under tight latency, cost, and on-pre…
15
Weak-to-Strong On-Policy Distillation
挑战传统认知：弱教师如何高效指导强学生？这项研究提出了弱到强的在线策略蒸馏新范式。
arXiv:2607.26246v1 Announce Type: new Abstract: On-policy distillation (OPD), which aligns a student with the teacher's token-level distribution on th…
16
TabRank: Chain-of-Thought Distillation for Table Re-Rankers
用思维链蒸馏提升表格重排序效果，TabRank创新方法亮相。
arXiv:2607.25182v1 Announce Type: cross Abstract: The ability to retrieve relevant tables for answering questions is a key task for structured informa…
17
RM-Distiller: Exploiting Generative LLM for Reward Model Distillation
生成式大模型蒸馏奖励模型新方法，解决偏好标注昂贵难题。
arXiv:2601.14032v2 Announce Type: replace Abstract: Reward models (RMs) play a pivotal role in aligning large language models (LLMs) with human prefer…
18
科技行业力挺开源 AI，Anthropic 因保持沉默遭同行批评
硅谷巨头集体签公开信力挺开源AI，唯独Anthropic因沉默遭同行炮轰，揭开放权之争新波澜。
IT之家 7 月 27 日消息，科技行业正集体支持开放权重（Open-weight）AI，但有一家重量级 AI 实验室却始终保持沉默。 随着美国政府考虑限制部分中国 AI 模型，Anthropic 因未签署支持开放权重 AI 的公开信，正受到硅谷同行的广泛批评。 上周五，包括英伟达（Nvidia）、…
19
Did Chinese AI Steal From Anthropic, and OpenAI Loses Control of Two Models
中美AI竞赛再起波澜：白宫指控Moonshot AI蒸馏Anthropic模型，同时OpenAI两大模型失控。
On this episode of Uncanny Valley, we dive into accusations that China’s Moonshot AI stole from Anthropic, and how the US Army needs to cut back on AI…
20
Semi-Supervised Text-Attributed Graph Distillation
图半监督蒸馏新方法被KDD2026录用，利用文本属性提升图表示学习效率。
arXiv:2607.20477v1 Announce Type: new Abstract: {\em Text-Attributed Graphs} (TAGs) have emerged as an expressive data model for integrating graph top…