LLM & SLM 研究日报
算法·训练·推理 —— 大语言模型与小语言模型的前沿研究
生成时间: 2026/7/15 09:00:05
📊 今日概况
| 方向 | 论文数 |
|---|---|
| 🧮 算法与架构 | 8 |
| 🏋️ 训练方法 | 4 |
| ⚡ 推理优化 | 4 |
| 总计扫描 | 50 |
📝 论文列表
🧮 算法与架构 (8 篇)
1. MET: Theory-Grounded and Culture-Aware Multilingual Moral Reasoning
- arXiv: 2607.11736
- 摘要: moral,met,culture,multilingual,mclash,qwen3,reasoning,grounded,grounds,theory
- 关键词: moral,met,culture,multilingual,mclash,qwen3,reasoning,grounded,grounds,theory
2. LightMem-Ego: Your AI Memory for Everyday Life
- arXiv: 2607.11487
- 摘要: lightmem,ego,memory,continuously,multimodal,everyday,life,assistance,audio,streams
- 关键词: lightmem,ego,memory,continuously,multimodal,everyday,life,assistance,audio,streams
3. UMoE:Unlocking Every Expert in Domain-Specific Training
- arXiv: 2607.11444
- 摘要: umoe,sft,expert,a3b,domain,qwen3,pool,30b,experts,moe
- 关键词: umoe,sft,expert,a3b,domain,qwen3,pool,30b,experts,moe
4. Direct Image-to-Modern Vietnamese Translation of Han-Nom Manuscripts via Multimodal RLHF Preference Alignment
- arXiv: 2607.11434
- 摘要: nom,vietnamese,han,preference,kto,manuscripts,rlhf,multimodal,ppo,dpo
- 关键词: nom,vietnamese,han,preference,kto,manuscripts,rlhf,multimodal,ppo,dpo
5. Invariant Learning Dynamics of Transformers in Inductive Reasoning Tasks
- arXiv: 2607.11875
- 摘要: inductive,tasks,reasoning,transformers,manifold,dynamics,interpretable,invariant,transformer,circuit
- 关键词: inductive,tasks,reasoning,transformers,manifold,dynamics,interpretable,invariant,transformer,circuit
6. Transformer-Guided Swarm Intelligence for Frugal Neural Architecture Search
- arXiv: 2607.11826
- 摘要: frugal,architecture,search,nas,transformer,gpu,3060,bloat,swarm,memetic
- 关键词: frugal,architecture,search,nas,transformer,gpu,3060,bloat,swarm,memetic
7. An Exact Instrument for State Usage in Selective State-Space Models, and the Input-Driven Migration It Reveals
- arXiv: 2607.11796
- 摘要: mamba,instrument,input,deployed,130m,falcon,exact,modes,mode,error
- 关键词: mamba,instrument,input,deployed,130m,falcon,exact,modes,mode,error
8. AutoMatBench: An Automatic Optimization Toolkit for the Acceleration of Material Properties Prediction Benchmarking
- arXiv: 2607.11526
- 摘要: matbench,benchmarking,automatbench,mpp,material,performances,prediction,ood,optimization,toolkit
- 关键词: matbench,benchmarking,automatbench,mpp,material,performances,prediction,ood,optimization,toolkit
🏋️ 训练方法 (4 篇)
1. Metacognition in LLMs: Foundations, Progress, and Opportunities
- arXiv: 2607.11881
- 摘要: metacognition,llms,metacognitive,abilities,progress,taxonomize,foundations,intelligence,yale,opportunities
- 关键词: metacognition,llms,metacognitive,abilities,progress,taxonomize,foundations,intelligence,yale,opportunities
2. RAGU: A Multi-Step GraphRAG Engine with a Compact Domain-Adapted LLM
- arXiv: 2607.11683
- 摘要: graphrag,ragu,meno,lite,llm,skills,extraction,factoid,hipporag2,knowledge
- 关键词: graphrag,ragu,meno,lite,llm,skills,extraction,factoid,hipporag2,knowledge
3. UMoE:Unlocking Every Expert in Domain-Specific Training
- arXiv: 2607.11444
- 摘要: umoe,sft,expert,a3b,domain,qwen3,pool,30b,experts,moe
- 关键词: umoe,sft,expert,a3b,domain,qwen3,pool,30b,experts,moe
4. Direct Image-to-Modern Vietnamese Translation of Han-Nom Manuscripts via Multimodal RLHF Preference Alignment
- arXiv: 2607.11434
- 摘要: nom,vietnamese,han,preference,kto,manuscripts,rlhf,multimodal,ppo,dpo
- 关键词: nom,vietnamese,han,preference,kto,manuscripts,rlhf,multimodal,ppo,dpo
⚡ 推理优化 (4 篇)
1. PaperRouter-Agent: A Content-Grounded LLM Agent for Personalized Hierarchical Paper Routing
- arXiv: 2607.11564
- 摘要: folder,paperrouter,agent,folders,routing,papers,hierarchy,venue,recall,hierarchical
- 关键词: folder,paperrouter,agent,folders,routing,papers,hierarchy,venue,recall,hierarchical
2. HiFi-LLP: High-Fidelity, Low-Cost Latency Predictors with Confidence for Robust HW-NAS
- arXiv: 2607.11746
- 摘要: nas,llp,hifi,predictors,hil,latency,confidence,hardware,aware,latbench
- 关键词: nas,llp,hifi,predictors,hil,latency,confidence,hardware,aware,latbench
3. Fundamental Limitations of Fixed-Budget Best-Arm Identification
- arXiv: 2607.11635
- 摘要: arm,oracle,budget,proportions,identification,static,fixed,best,sampling,decay
- 关键词: arm,oracle,budget,proportions,identification,static,fixed,best,sampling,decay
4. Advancing Optimal Subset Oracle via Learning Relaxation of Neural Set Functions
- arXiv: 2607.11555
- 摘要: variational,set,subset,relaxation,sampling,gradients,overhead,surrogate,advancing,oracle
- 关键词: variational,set,subset,relaxation,sampling,gradients,overhead,surrogate,advancing,oracle
分析生成失败
📚 附录
筛选关键词
算法: attention mechanism, mixture of experts, MoE, sparse attention, flash attention, rotary position, RoPE, grouped query, GQA, KV cache …
训练: pre-training, pretraining, post-training, fine-tuning, finetuning, supervised fine-tuning, SFT, alignment, RLHF, DPO …
推理: inference, serving, latency, throughput, speculative decoding, batching, continuous batching, PagedAttention, vLLM, quantization …
本报告由 OpenClaw 自动生成 | LLM & SLM Research Daily