6. LLM Research Papers: The 2026 List (Jan–May) — Sebastian Raschka

类型：arxiv
标识：2604.12374
链接：https://arxiv.org/abs/2604.12374
主题：agent, engineering, evaluation, llm-infra, multimodal, rag, risk
主分类：llm-infra
形态：method
被引：18
被引来源：Semantic Scholar
S2被引：18
OpenAlex被引：0
影响力被引：2
TLDR：Nemotron 3 Super is the first model in the Nemotron 3 family to be pre-trained in NVFP4, leverage LatentMoE, a new Mixture-of-Experts architecture that optimizes for both accuracy per FLOP and accuracy per parameter, and include MTP layers for inference acceleration through native speculative decoding.
OpenAlex ID：W7154499146
OpenAlex DOI：10.48550/arxiv.2604.12374
DOI：10.48550/arxiv.2604.12374
DOI来源：OpenAlex
开放获取：green
开放获取链接：https://doi.org/10.48550/arxiv.2604.12374
OpenAlex更新：2026-07-19
标题中文：6. LLM Research Papers：2026 清单（1月–5月） — Sebastian Raschka
TLDR中文：Nemotron 3 Super 是 Nemotron 3 系列中首个采用 NVFP4 进行预训练的模型，借助 LatentMoE（一种同时优化精度 per FLOP 与精度 per parameter 的新型 Mixture-of-Experts 架构），并集成 MTP 层以通过原生 speculative decoding 加速推理。
待LLM分类：否
来源文件：
/inbox/jay/2026-06-16-afternoon-briefing-csdn-backend-agents-moe-substack.md
[S2 enrich]
[OpenAlex backfill]

可复用信息

- 来源: https://magazine.sebastianraschka.com/p/llm-research-papers-2026-part1
- 核心论文时间线:
- Jan 29: Scaling Embeddings Outperforms Scaling Experts (arXiv:2601.21204)
- Feb 8: ViT-5: Vision Transformers for the Mid-2020s (arXiv:2602.08071)
- Feb 17: GLM-5: From Vibe Coding to Agentic Engineering (arXiv:2602.15763)
- Mar 15: Attention Residuals (arXiv:2603.15031)
- Mar 16: Mamba-3: Improved Sequence Modeling Using State Space Principles (arXiv:2603.15569)