📚 论文索引

A curated collection of research papers on AI systems, compilers, architecture, and systems software.

---

Table of Contents

• Deep Learning Compiler
• LLM Inference
• LLM Training
• Deep Learning
• LLM Evaluation & Safety
• LLM for Kernel Optimization
• Agent Systems
• GPU Microarchitecture
• Math Foundations
• Compiler
• Operating Systems
• Hypervisor & Virtualization
• RISC-V

Legend: ✅ = Read  |  ⬜ = To Read  |  📝 = Note Available

Metadata: Venue 链接指向正式论文集、出版社、作者页面或预印本来源；arXiv 表示已核实的预印本版本，博客、开源项目与技术报告按实际类型标注。Why It Matters 说明阅读价值，不代表该论文已读。核对日期：2026-09-15；数量按分类条目统计，保留跨分类重复收录。

Learning roadmap: LLM Training, Inference & Algorithms — Recommended Reading Path

---

🔧 Deep Learning Compiler

Status	Paper	Venue	Why It Matters	Links
✅	The Deep Learning Compiler: A Comprehensive Survey	TPDS'21	梳理多层 IR、图优化与后端调度，建立深度学习编译器的整体知识框架	Paper / Note
✅	MLIR: Scaling Compiler Infrastructure for Domain Specific Computation	CGO'21	通过可扩展 dialect 与逐层 lowering 复用编译基础设施，理解现代 AI 编译器架构	Paper / Note
✅	TIRAMISU: A Polyhedral Compiler for Expressing Fast and Portable Code	CGO'19	分离算法、循环变换、数据布局和通信，理解多面体调度与跨硬件代码生成	Paper / Note
✅	Rammer: Enabling Holistic Deep Learning Compiler Optimizations with rTasks	OSDI'20	用 rTask 统一算子内外并行性，在编译期生成静态时空调度以减少运行时开销	Paper / Note
✅	ROLLER: Fast and Efficient Tensor Compilation for Deep Learning	OSDI'22	用硬件感知 rTile 和性能模型构造候选 kernel，缩短传统大规模自动调优时间	Paper / Note
✅	BOLT: Bridging The Gap Between Auto-Tuners and Hardware-Native Performance	MLSys'22	利用 CUTLASS 等硬件原生模板进行搜索，连接自动调优的灵活性与库级性能	Paper / Note
✅	AStitch: Enabling a New Multi-dimensional Optimization Space for Memory-Intensive ML Training and Inference on Modern SIMT Architectures	ASPLOS'22	联合探索线程映射、片上复用与算子融合，优化访存密集型训练和推理算子	Paper / Note
✅	AMOS: Enabling Automatic Mapping for Tensor Computations On Spatial Accelerators with Hardware Abstraction	ISCA'22	将硬件 intrinsic 抽象成可分析的计算与访存语义，自动生成张量到加速器的映射	Paper / Note
✅	Welder: Scheduling Deep Learning Memory Access via Tile-graph	OSDI'23	用 tile-graph 与数据流量模型统一算子内外数据复用，降低端到端访存开销	Paper / Note
✅	Effectively Scheduling Computational Graphs of Deep Neural Networks toward Their Domain-Specific Accelerators	OSDI'23	将硬件资源纳入子图划分和跨层指令调度，减少片外数据搬运并提高加速器利用率	Paper
✅	Cocktailer: Analyzing and Optimizing Dynamic Control Flow in Deep Learning	OSDI'23	以 uTask 统一控制流和数据流，把动态分支与循环下沉到加速器以减少 CPU 同步	Paper / Note
✅	Chimera: An Analytical Optimizing Framework for Effective Compute-intensive Operators Fusion	HPCA'23	用分析模型选择计算块执行顺序，并结合硬件微内核优化计算密集型算子链融合	Paper / Note
✅	Graphene: An IR for Optimized Tensor Computations on GPUs	ASPLOS'23	将数据和线程都表示为可分块张量，显式表达 GPU tensor 指令要求的复杂映射	Paper / Note
✅	Uncovering Nested Data Parallelism and Data Reuse in DNN Computation with FractalTensor	SOSP'24	通过嵌套张量与数组算子表达跨循环、跨算子的并行性和复用，突破平面计算图限制	Paper
✅	ThunderKittens: Simple, Fast, and Adorable AI Kernels	ICLR'25	以 tile 数据结构、异步流水线模板和网格调度简化高性能 GPU kernel 编写	Paper
✅	Mirage: A Multi-Level Superoptimizer for Tensor Programs	OSDI'25	用多层 µGraph 搜索代数和调度变换，并验证等价性，探索超越既有算子库的融合	Paper
✅	PipeThreader: Software-Defined Pipelining for Efficient DNN Execution	OSDI'25	用 sTask-graph 和软件定义流水线协调 GPU 专用单元，自动发现计算与数据搬运的重叠方案	Paper
✅	TileLang: A Composable Tiled Programming Model for AI Systems	arXiv'25	将 tile 数据流与布局、线程映射和流水线调度解耦，降低高性能 AI kernel 编程负担	Paper
✅	Tawa: Automatic Warp Specialization for Modern GPUs with Asynchronous References	CGO'26	以异步引用表达依赖并自动生成 warp specialization，减少手工管理并发和同步的负担	Paper
✅	KPerfIR: Towards an Open and Compiler-centric Ecosystem for GPU Kernel Performance Tooling on Modern AI Workloads	OSDI'25	将性能测量实现为编译器 pass，观察 GPU kernel 内部的细粒度执行与重叠瓶颈	Paper
⬜	Optimal Software Pipelining and Warp Specialization for Tensor Core GPUs	OSDI'26	将软件流水线和 warp specialization 联合建模为约束优化，推导模型假设下的最优调度	Paper
⬜	GraCE: Unlocking CUDA Graphs with Compiler Support for ML Workloads	OSDI'26	自动变换程序并分析成本收益，扩大 CUDA Graph 覆盖率并减少参数拷贝和启动开销	Paper

🚀 LLM Inference

Decoding Algorithms

Status	Paper	Venue	Why It Matters	Links
⬜	The Curious Case of Neural Text Degeneration	ICLR'20	系统比较 greedy、beam、top-k 与 nucleus sampling，理解生成质量、随机性和退化	Paper
⬜	Fast Inference from Transformers via Speculative Decoding	ICML'23	draft–verify 与拒绝采样；在保持目标分布不变的前提下并行生成多个 token	Paper
⬜	Accelerating Large Language Model Decoding with Speculative Sampling	arXiv'23	speculative sampling 的独立推导与工程验证，适合和上一论文对照阅读	Paper
⬜	SpecInfer: Accelerating Large Language Model Serving with Tree-based Speculative Inference and Verification	ASPLOS'24	用候选树提高并行验证宽度，连接解码算法与 serving batch	Paper
⬜	Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads	ICML'24	不依赖独立 draft model 的多头预测与 tree attention	Paper
⬜	Break the Sequential Dependency of LLM Inference Using Lookahead Decoding	ICML'24	从 Jacobi iteration 理解无额外模型的并行候选生成	Paper
⬜	EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty	ICML'24	在 feature space 自回归预测，理解高接受率 drafter 的训练方法	Paper
⬜	Better & Faster Large Language Models via Multi-token Prediction	ICML'24	训练时预测多个未来 token，将训练目标与推理解码并行性连接起来	Paper

General

Status	Paper	Venue	Why It Matters	Links
✅	A Survey of LLM Inference Systems	arXiv'25	贯通 kernel、批处理、调度与 KV 管理，理解单副本、分离式和分布式推理系统的设计取舍	Paper / Note
⬜	WaferLLM: Large Language Model Inference at Wafer Scale	OSDI'25	研究晶圆级互连和存储条件下的 LLM 映射，理解不同硬件形态如何改变推理并行策略	Paper

Long Context Inference

Status	Paper	Venue	Why It Matters	Links
✅	Training-Free Long-Context Scaling of Large Language Models	ICML'24	用 Dual Chunk Attention 调整位置关系，无继续训练地扩展模型可处理的上下文长度	Paper / Note
✅	Efficient Streaming Language Models with Attention Sinks	ICLR'24	保留 attention sink 与滑动窗口，实现有界 KV 缓存的流式生成并明确历史信息保留边界	Paper / 讨论笔记
✅	Quest: Query-Aware Sparsity for Efficient Long-Context LLM Inference	ICML'24	根据当前 query 和 KV 页元数据选择重要页面，减少长上下文解码中的 KV 读取	Paper
✅	DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads	ICLR'25	区分 retrieval heads 与 streaming heads，为不同 head 分配不同 KV 保留策略	Paper
✅	MiniMax Sparse Attention	arXiv'26	按 GQA group 选择 KV 块并协同设计稀疏 kernel，将块级稀疏转化为长上下文加速	Paper / Note
✅ 📝	HiSparse: Scaling Sparse-Attention Decoding with Hierarchical KV Cache Management	arXiv'26	将完整 KV 放在主存、热点放在有界 GPU 缓存，并用跨层预取隐藏稀疏解码缺页开销	Paper / Note
✅ 📝	IndexCache: Accelerating Sparse Attention via Cross-Layer Index Reuse	arXiv'26	跨层复用 Top-k 位置，用 LM-loss 搜索或多层 KL 蒸馏减少 indexer 计算，并区分索引缓存与主 KV 缓存	Paper / Note
⬜	Unifying Sparse Attention with Hierarchical Memory for Scalable Long-Context LLM Serving (SPIN)	arXiv'26	统一不同稀疏粒度与分层 KV 存储，通过局部性感知缓存和元数据设计降低 GPU–CPU 传输开销	Paper

LLM Serving

Status	Paper	Venue	Why It Matters	Links
⬜	Orca: A Distributed Serving System for Transformer-Based Generative Models	OSDI'22	提出迭代级调度和选择性批处理，使不同长度请求能动态进入和离开生成批次	Paper
⬜	Efficient Memory Management for Large Language Model Serving with PagedAttention	SOSP'23	用分页 KV 管理消除大块连续分配需求，并支持缓存共享，奠定高吞吐 LLM serving 基础	Paper
⬜	FlexGen: High-Throughput Generative Inference of Large Language Models with a Single GPU	ICML'23	联合规划 GPU、CPU 和磁盘上的权重与 KV 放置，以卸载和大批处理换取单 GPU 高吞吐	Paper
⬜	AlpaServe: Statistical Multiplexing with Model Parallelism for Deep Learning Serving	OSDI'23	利用模型并行实现统计复用，理解多个模型共用集群时的资源配置和尾延迟取舍	Paper
⬜	FastServe: Fast Distributed Inference Serving for Large Language Models	arXiv'23	以 token 级抢占和多级反馈队列缓解队首阻塞，并配合 KV 卸载改善请求完成延迟	Paper
⬜	InferCept: Efficient Intercept Support for Augmented Large Language Model Inference	ICML'24	为调用外部工具时暂停的生成管理 KV 保留、卸载与重算，减少恢复推理的资源浪费	Paper
⬜	Punica: Multi-Tenant LoRA Serving	MLSys'24	用专门的批量 LoRA kernel 共同服务不同适配器，提高多租户微调模型的 GPU 利用率	Paper
⬜	S-LoRA: Serving Thousands of Concurrent LoRA Adapters	MLSys'24	统一管理适配器权重与 KV 的分页内存，并批量执行异构 LoRA 请求以扩大并发规模	Paper
⬜	Splitwise: Efficient Generative LLM Inference Using Phase Splitting	ISCA'24	将 prefill 与 decode 放到独立资源池，匹配两阶段不同的算力、带宽和功耗需求	Paper
⬜	Taming Throughput-Latency Tradeoff in LLM Inference with Sarathi-Serve	OSDI'24	用 chunked prefill 和混合批处理控制解码停顿，改善吞吐与逐 token 延迟的取舍	Paper
⬜	Llumnix: Dynamic Scheduling for Large Language Model Serving	OSDI'24	通过运行时请求迁移协调多实例负载，在动态请求与资源变化下控制延迟	Paper
⬜	Preble: Efficient Distributed Prompt Scheduling for LLM Serving	ICLR'25	联合优化跨 GPU 前缀 KV 复用与负载均衡，避免局部缓存命中率牺牲整体服务延迟	Paper
✅	SGLang: Efficient Execution of Structured Language Model Programs	NeurIPS'24	以结构化程序前端、RadixAttention 前缀复用和约束解码优化多调用 LLM 应用	Paper
✅	FlashInfer: Efficient and Customizable Attention Engine for LLM Inference Serving	MLSys'25	统一 KV 格式、可定制 attention kernel 与动态负载调度，连接算子优化和实际 serving	Paper
⬜	DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model Serving	OSDI'24	按 TTFT 与 TPOT 目标独立配置 prefill、decode 资源，优化满足服务目标的有效吞吐	Paper
⬜	LoongServe: Efficiently Serving Long-Context Large Language Models with Elastic Sequence Parallelism	SOSP'24	用弹性序列并行动态调整长请求占用的 GPU，协调长上下文 prefill 与 decode	Paper
⬜	Mooncake: Trading More Storage for Less Computation — A KVCache-centric Architecture for Serving LLM Chatbot	FAST'25	围绕分离式 KV 缓存池组织推理，利用更多存储和跨请求复用减少重复 prefill	Paper
⬜	NanoFlow: Towards Optimal Large Language Model Serving Throughput	OSDI'25	用 nano-batch 流水线重叠计算、访存和通信，突破逐算子执行对服务吞吐的限制	Paper
⬜	From Tokens to Layers: Redefining Stall-Free Scheduling for MoE Serving with Layered Prefill	MLSys'26	沿层组而非 token 切分 prefill，减少 MoE 专家权重重复加载，同时保持解码连续性	Paper
⬜	CRAFT: Fine-Grained Cost-Aware Expert Replication for Efficient Mixture-of-Experts Serving	MLSys'26	按层估计专家复制收益，在显存预算内平衡路由负载，避免过度复制挤占 KV 空间	Paper
⬜	Strata: Hierarchical Context Caching for Long Context Language Model Serving	OSDI'26	结合大块 KV I/O 与缓存感知调度，缓解 GPU、主存和 SSD 分层上下文缓存的加载停顿	Paper
⬜	No Buffer, No Bottleneck: Efficient Zero-Copy KV Cache Offloading for Long-Context LLMs (DirectKV)	OSDI'26	在 NVLink-C2C 平台让 GPU 直接访问主存 KV，消除 staging buffer 与额外拷贝	Paper

Quantization & Compression

Status	Paper	Venue	Why It Matters	Links
⬜	LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale	NeurIPS'22	混合精度分解 activation outlier，是理解 LLM INT8 的起点	Paper
⬜	GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers	ICLR'23	基于近似二阶信息的 one-shot weight-only PTQ	Paper
⬜	SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models	ICML'23	用等价缩放把 activation 量化难度迁移到 weight，形成 W8A8 路线	Paper
⬜	AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration	MLSys'24	用 activation 识别显著通道并保护关键权重，形成主流 W4A16 路线	Paper
⬜	SpQR: A Sparse-Quantized Representation for Near-Lossless LLM Weight Compression	ICLR'24	将少量 outlier 与低比特权重分离，理解稀疏异常值处理	Paper
⬜	OmniQuant: Omnidirectionally Calibrated Quantization for Large Language Models	ICLR'24	通过可学习的 clipping 与等价变换改善低比特 PTQ	Paper
⬜	AQLM: Extreme Compression of Large Language Models via Additive Quantization	ICML'24	additive codebook 与极低 bit/weight，连接压缩算法和解码 kernel	Paper
⬜	KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache	ICML'24	区分 K/V 的通道与 token 统计特性，理解 KV cache 低比特化	Paper
⬜	QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMs	NeurIPS'24	用 Hadamard rotation 消除 outlier，适合从算法追到融合 kernel	Paper
⬜	The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits	arXiv'24	BitNet b1.58 与三值权重，理解量化感知训练的架构路线	Paper

MegaKernel

Status	Paper	Venue	Why It Matters	Links
✅	Look Ma, No Bubbles! Designing a Low-Latency Megakernel for Llama-1B	Blog'25	以 Llama-1B 实例解释 megakernel 如何消除 kernel 边界气泡，并管理细粒度同步和共享内存	Paper
✅	MPK: A Compiler and Runtime for Mega-Kernelizing Tensor Programs	OSDI'26	将多 GPU 推理编译成 SM 级任务图，在持久化 kernel 内调度计算与通信以降低端到端延迟	Paper
✅ 📝	Event Tensor: A Unified Abstraction for Compiling Dynamic Megakernel	MLSys'26	以 Event Tensor 统一 tile 任务依赖、动态形状和数据相关控制流，支持动态 megakernel 编译	Paper / Note
✅	TileRT: Tile-Based Runtime for Ultra-Low-Latency LLM Inference	GitHub project'25	用 tile 级运行时重排并重叠计算、I/O 与跨设备通信，探索大模型低延迟推理的工程路径	Paper
✅	SonicMoE: Accelerating MoE with IO and Tile-aware Optimizations	arXiv'25	减少 MoE 前后向激活缓存和 I/O，并用 token rounding 降低 grouped GEMM 的填充浪费	Paper
✅ 📝	MegaMoE (fused MoE megakernel, DeepGEMM)	DeepGEMM PR'26	将专家分发、两次线性层、SwiGLU 和合并融合，展示 NVLink 通信与 Tensor Core 计算重叠	Source / Note
✅	Compiling LLMs into a MegaKernel: A Path to Low-Latency Inference	Blog	从编译器和运行时角度解释全模型 megakernel，连接任务依赖、跨算子调度与低延迟推理	Paper
⬜	Ada-MK: Adaptive MegaKernel Optimization via Automated DAG-based Search for LLM Inference	arXiv'26	用离线 DAG 搜索固定 megakernel 执行路径，减少资源受限 GPU 上动态调度和共享内存开销	Paper

MegaMoE 摘要（非正式论文，见 DeepGEMM PR #304）：把 MoE 前向中分发、两层分组 GEMM、SwiGLU、合并压进单一持久化 CUDA 核；用对称显存布局与 NVLink 在核内做专家并行词元交换，并以波次调度、L1/L2 词元池上的细粒度到达计数 / 掩码，把通信与计算流水重叠；SM100 上按分发、TMA+MMA、尾声与合并划分线程束角色与寄存器预算。

🏋️ LLM Training

Training Numerics & Memory

Status	Paper	Venue	Why It Matters	Links
⬜	Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour	arXiv'17	global batch、线性学习率缩放与 warmup；理解数据并行扩展为何会改变优化行为	Paper
⬜	Mixed Precision Training	ICLR'18	FP16 计算与 FP32 master weights、loss scaling 配合；连接 Tensor Core 吞吐和收敛稳定性	Paper
⬜	Training Deep Nets with Sublinear Memory Cost	arXiv'16	activation checkpointing/rematerialization 的经典计算–显存交换	Paper
⬜	Reducing Activation Recomputation in Large Transformer Models	MLSys'23	sequence parallelism 与 selective recomputation，解释 Megatron 的 activation 内存优化	Paper
⬜	GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection	ICML'24	对梯度做低秩投影以降低 optimizer state 和训练显存	Paper

Distributed Training

Status	Paper	Venue	Why It Matters	Links
⬜	PyTorch Distributed: Experiences on Accelerating Data Parallel Training	VLDB'20	剖析 DDP 梯度分桶、通信计算重叠与同步机制，理解数据并行的实际扩展瓶颈	Paper
⬜	GPipe: Efficient Training of Giant Neural Networks using Pipeline Parallelism	NeurIPS'19	用 microbatch 和重计算搭建同步流水线训练，理解大模型分层切分与 bubble 开销	Paper
⬜	PipeDream: Generalized Pipeline Parallelism for DNN Training	SOSP'19	通过流水线调度和权重版本管理提高训练利用率，理解流水线并行中的参数一致性取舍	Paper
⬜	Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism	arXiv'19	在 Transformer 内进行张量并行切分，以较少通信支撑多十亿参数语言模型训练	Paper
⬜	ZeRO: Memory Optimizations Toward Training Trillion Parameter Models	SC'20	分片优化器状态、梯度和参数，解释数据并行如何消除显存冗余并扩展模型规模	Paper
⬜	ZeRO-Offload: Democratizing Billion-Scale Model Training	USENIX ATC'21	将优化器计算和状态卸载到 CPU，并与 GPU 训练配合，让有限显存支持更大模型	Paper
⬜	Memory-Efficient Pipeline-Parallel DNN Training	ICML'21	用 PipeDream-2BW 的双版本权重和流水线调度降低内存需求，兼顾吞吐与训练语义	Paper
⬜	ZeRO-Infinity: Breaking the GPU Memory Wall for Extreme Scale Deep Learning	SC'21	联合 GPU、CPU 与 NVMe 分层卸载和带宽管理，突破超大模型训练的 GPU 显存容量限制	Paper
⬜	Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM	SC'21	组合张量、流水线与数据并行，并用交错流水线减少大模型训练气泡	Paper
⬜	GSPMD: General and Scalable Parallelization for ML Computation Graphs	arXiv'21	用少量张量分片标注自动传播并行策略，将单设备计算图扩展到多设备	Paper
⬜	Alpa: Automating Inter- and Intra-Operator Parallelism for Distributed Deep Learning	OSDI'22	分层搜索算子内与算子间并行策略，自动规划分布式训练执行方案	Paper
⬜	DeepSpeed Ulysses: System Optimizations for Enabling Training of Extreme Long Sequence Transformer Models	arXiv'23	通过 all-to-all 转换序列与 attention head 分片，支持极长序列训练	Paper
⬜	Ring Attention with Blockwise Transformers for Near-Infinite Context	ICLR'24	用环形 KV 块传输重叠分块 attention 计算，将长上下文分摊到多设备	Paper
⬜	Oobleck: Resilient Distributed Training of Large Models Using Pipeline Templates	SOSP'23	以异构流水线模板和冗余模型状态实现故障后的快速恢复与重配置	Paper
⬜	MegaScale: Scaling Large Language Model Training to More Than 10,000 GPUs	NSDI'24	结合训练算法、通信与故障监控优化，展示万卡规模 LLM 训练的工程瓶颈	Paper
⬜	LoongTrain: Efficient Training of Long-Sequence LLMs with Head-Context Parallelism	arXiv'24	结合 head 与 context 二维并行及双环通信，突破长序列训练的扩展限制	Paper
✅ 📝	PithTrain: A Compact and Agent-Native MoE Training System	arXiv'26	以紧凑模块和显式训练结构降低 MoE 系统修改成本，支持编码代理迭代优化	Paper / Note / Code Guide

Mixture-of-Experts Training

Status	Paper	Venue	Why It Matters	Links
⬜	Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer	ICLR'17	现代稀疏 MoE 的起点：top-k gate、稀疏激活与负载均衡	Paper
⬜	GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding	ICLR'21	将 MoE、SPMD sharding 和大规模 Transformer 训练结合	Paper
⬜	BASE Layers: Simplifying Training of Large, Sparse Models	ICML'21	用 balanced assignment 避免额外负载均衡损失	Paper
⬜	GLaM: Efficient Scaling of Language Models with Mixture-of-Experts	ICML'22	大规模稀疏语言模型的质量、计算与能耗权衡	Paper
⬜	DeepSpeed-MoE: Advancing Mixture-of-Experts Inference and Training to Power Next-Generation AI Scale	ICML'22	expert parallel、通信与 MoE inference/training 系统化设计	Paper
⬜	FasterMoE: Modeling and Optimizing Training of Large-Scale Dynamic Pre-Trained Models	PPoPP'22	动态路由下的 shadowing 与拓扑感知通信优化	Paper
⬜	Tutel: Adaptive Mixture-of-Experts at Scale	MLSys'23	自适应并行、all-to-all 与 kernel 优化的完整 MoE 系统	Paper
⬜	MegaBlocks: Efficient Sparse Training with Mixture-of-Experts	MLSys'23	将 token dropping 问题转成 block-sparse GEMM，最贴近算子视角	Paper

RL Training

Status	Paper	Venue	Why It Matters	Links
✅	Seer: Online Context Learning for Fast Synchronous LLM Reinforcement Learning	OSDI'26	利用同一 prompt 下响应长度和模式的相似性，减少同步 RL rollout 长尾与资源浪费	Paper

Fine-Tuning & Alignment

Status	Paper	Venue	Why It Matters	Links
⬜	Fine-Tuning Language Models from Human Preferences	arXiv'19	将 reward model 与 PPO 用于语言模型偏好优化的早期完整方案	Paper
⬜	Learning to Summarize from Human Feedback	NeurIPS'20	展示偏好数据、reward model 和 RL 在真实生成任务中的规模化	Paper
⬜	Prefix-Tuning: Optimizing Continuous Prompts for Generation	ACL'21	冻结主模型，仅训练可学习 prefix 的参数高效微调	Paper
⬜	The Power of Scale for Parameter-Efficient Prompt Tuning	EMNLP'21	soft prompt 与模型规模关系，适合理解 PEFT 的表达能力	Paper
⬜	LoRA: Low-Rank Adaptation of Large Language Models	ICLR'22	冻结权重并注入低秩更新，连接矩阵秩、训练显存和多租户推理	Paper
⬜	Proximal Policy Optimization Algorithms	arXiv'17	RLHF 所需的最低限度策略优化基础：ratio、advantage 与 clipping	Paper
⬜	Finetuned Language Models Are Zero-Shot Learners	ICLR'22	FLAN 与 instruction tuning，说明任务混合如何产生泛化能力	Paper
⬜	Training Language Models to Follow Instructions with Human Feedback	NeurIPS'22	InstructGPT 的 SFT → reward model → PPO 三阶段链路	Paper
⬜	Constitutional AI: Harmlessness from AI Feedback	arXiv'22	self-critique、revision 与 RLAIF，把安全原则引入后训练	Paper
⬜	Self-Instruct: Aligning Language Models with Self-Generated Instructions	ACL'23	自生成、过滤和扩增指令数据的经典流程	Paper
⬜	The Flan Collection: Designing Data and Methods for Effective Instruction Tuning	ICML'23	系统研究任务混合、模板、CoT 数据与 instruction tuning 配方	Paper
⬜	QLoRA: Efficient Finetuning of Quantized LLMs	NeurIPS'23	NF4、double quantization、paged optimizer 与 LoRA 的组合	Paper
⬜	LIMA: Less Is More for Alignment	NeurIPS'23	少量高质量监督数据与大规模预训练知识之间的分工	Paper
⬜	Direct Preference Optimization: Your Language Model is Secretly a Reward Model	NeurIPS'23	将显式 reward model + PPO 化为稳定的 pairwise classification loss	Paper
⬜	RRHF: Rank Responses to Align Language Models with Human Feedback	NeurIPS'23	用候选排序损失统一多种反馈来源	Paper
⬜	A General Theoretical Paradigm to Understand Learning from Human Preferences	AISTATS'24	IPO 与偏好优化的理论视角，理解 DPO 类方法的过拟合和正则化	Paper
⬜	KTO: Model Alignment as Prospect Theoretic Optimization	ICML'24	只需 desirable/undesirable 标签的非成对偏好学习	Paper
⬜	ORPO: Monolithic Preference Optimization without Reference Model	EMNLP'24	将 SFT 与偏好约束合并，移除独立 reference model	Paper
⬜	SimPO: Simple Preference Optimization with a Reference-Free Reward	NeurIPS'24	reference-free、长度归一化 reward 与 margin objective	Paper
⬜	DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models	arXiv'24	GRPO、数学数据与可验证奖励，是 reasoning RL 的关键前置	Paper
⬜	DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning	Nature'25	大规模 reasoning RL、冷启动数据与蒸馏路线	Paper

Compute-Communication Overlap

Status	Paper	Venue	Why It Matters	Links
✅	Flux: Fast Software-based Communication Overlap on GPUs through Kernel Fusion	arXiv'24	将通信融入 GEMM 内核并按 tile 细粒度重叠，降低张量并行的暴露通信耗时	Paper
✅	DeepEP: An Efficient Expert-Parallel Communication Library	GitHub'25（开源库）	提供面向 MoE token dispatch/combine 的高吞吐与低延迟专家并行通信实现	Paper
⬜	Centauri: Enabling Efficient Scheduling for Communication-Computation Overlap in Large Model Training via Communication Partitioning	ASPLOS'24	通过通信分区和分层调度扩大重叠空间，协调大模型训练中的通信与计算资源	Paper
⬜	Comet: Fine-grained Computation-communication Overlapping for Mixture-of-Experts	MLSys'25	分析 MoE 数据依赖并重排计算通信任务，实现细粒度重叠以降低专家并行开销	Paper
⬜	TileLink: Generating Efficient Compute-Communication Overlapping Kernels using Tile-Centric Primitives	MLSys'25	用 tile 级通信原语和编译调度生成重叠内核，减少分布式算子的手工优化工作	Paper
⬜	Triton-distributed: Programming Overlapping Kernels on Distributed AI Systems with the Triton Compiler	arXiv'25	将 OpenSHMEM 通信原语引入 Triton，使分布式计算通信重叠可用统一语言表达	Paper
⬜	Efficient and Adaptable Overlapping for Computation and Communication via Signaling and Reordering (FlashOverlap)	EuroSys'26	通过 tile 就绪信号和数据重排实现细粒度重叠，复用通信库并减少对计算的干扰	Paper
⬜	TokenWeave: Efficient Compute-Communication Overlap for Distributed LLM Inference	MLSys'26	融合 AllReduce 与 RMSNorm，以少量 SM 支持较小 token 批次的推理通信重叠	Paper
⬜	UEP: Portable Expert-Parallel Communication	OSDI'26	用 GPU–CPU 控制通道与 CPU 代理发起 RDMA，实现跨 GPU/NIC 的可移植专家通信	Paper

🧠 Deep Learning

Foundations & Optimization

Status	Paper	Venue	Why It Matters	Links
⬜	Learning Representations by Back-Propagating Errors	Nature'86	从局部算子 backward 上升到计算图链式法则、梯度流和 activation 保存	Paper
⬜	Understanding the Difficulty of Training Deep Feedforward Neural Networks	AISTATS'10	Xavier initialization 与方差传播，理解初始化为何影响深层网络稳定性	Paper
⬜	Dropout: A Simple Way to Prevent Neural Networks from Overfitting	JMLR'14	经典正则化、train/eval 行为差异与随机 mask	Paper
⬜	Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift	ICML'15	对比 batch statistics 与 LayerNorm，理解同步 BN 和训练/推理差异	Paper
⬜	Deep Residual Learning for Image Recognition	CVPR'16	残差连接与深层梯度传播，是 Transformer residual stream 的结构前置	Paper
⬜	Adam: A Method for Stochastic Optimization	ICLR'15	一阶/二阶矩、bias correction 与 optimizer state 显存	Paper
⬜	Decoupled Weight Decay Regularization	ICLR'19	区分 L2 regularization 与 AdamW 的 decoupled weight decay	Paper
⬜	Layer Normalization	arXiv'16	单样本归一化、训练稳定性与 Transformer 中的 reduction/fusion	Paper
⬜	Root Mean Square Layer Normalization	NeurIPS'19	省去 re-centering 的 RMSNorm，连接现代 LLM 结构与高效 kernel	Paper
⬜	Adafactor: Adaptive Learning Rates with Sublinear Memory Cost	ICML'18	对二阶矩做 factored approximation，理解 optimizer memory 优化	Paper
⬜	Large Batch Optimization for Deep Learning: Training BERT in 76 Minutes	ICLR'20	LAMB 的 layer-wise scaling 与超大 batch 训练	Paper
⬜	Tensor Programs V: Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfer	NeurIPS'21	μP 与跨模型规模超参数迁移，连接 scaling experiment 和训练配方	Paper
⬜	Sophia: A Scalable Stochastic Second-order Optimizer for Language Model Pre-training	ICLR'24	低成本二阶曲率估计与 per-coordinate clipping	Paper

Language Modeling, Tokenization & Modern LLMs

Status	Paper	Venue	Why It Matters	Links
⬜	A Neural Probabilistic Language Model	JMLR'03	embedding、条件概率与 next-token language modeling 的早期完整形式	Paper
⬜	Sequence to Sequence Learning with Neural Networks	NeurIPS'14	自回归分解、teacher forcing、EOS 与 beam search 的基础	Paper
⬜	Neural Machine Translation by Jointly Learning to Align and Translate	ICLR'15	additive attention 与 encoder–decoder alignment，理解 Transformer 之前的问题	Paper
⬜	Neural Machine Translation of Rare Words with Subword Units	ACL'16	BPE、词表大小、序列长度和输出 softmax 成本之间的关系	Paper
⬜	SentencePiece: A Simple and Language Independent Subword Tokenizer and Detokenizer for Neural Text Processing	EMNLP'18（System Demonstrations）	从 raw text 训练 BPE/unigram tokenizer，理解现代 tokenizer pipeline	Paper
⬜	Improving Language Understanding by Generative Pre-Training	OpenAI Technical Report'18	GPT-1 与 decoder-only 预训练–微调范式	Paper
⬜	BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding	NAACL'19	encoder-only、masked LM 与 pretrain–finetune 范式	Paper
⬜	Language Models are Unsupervised Multitask Learners	OpenAI Technical Report'19	GPT-2、zero-shot transfer 与 WebText 数据路线	Paper
⬜	Language Models are Few-Shot Learners	NeurIPS'20	GPT-3、in-context learning 与 decoder-only scaling	Paper
⬜	Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer	JMLR'20	T5、span corruption、C4 与统一 text-to-text interface	Paper
⬜	GLU Variants Improve Transformer	arXiv'20	SwiGLU/GEGLU 与 gated FFN，解释现代 LLM 中三路投影和逐元素乘	Paper
⬜	RoFormer: Enhanced Transformer with Rotary Position Embedding	Neurocomputing'24	RoPE 的旋转与相对位置语义，以及 Q/K 融合实现	Paper
⬜	Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation	ICLR'22	ALiBi 与无需显式位置 embedding 的长度外推	Paper
⬜	PaLM: Scaling Language Modeling with Pathways	JMLR'23	大规模 dense LLM、Pathways 并行与训练行为分析	Paper
⬜	LLaMA: Open and Efficient Foundation Language Models	arXiv'23	将 RMSNorm、SwiGLU、RoPE、tokenizer、数据与训练配方串成现代 LLM	Paper
⬜	Llama 2: Open Foundation and Fine-Tuned Chat Models	arXiv'23	预训练、SFT、RLHF、安全评测与 chat model 的完整技术报告	Paper
⬜	Mistral 7B	arXiv'23	sliding-window attention、GQA 与 rolling buffer KV cache	Paper
⬜	DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models	ACL'24	shared expert、细粒度 expert segmentation 与稀疏计算配比	Paper
⬜	Mixtral of Experts	arXiv'24	实用 sparse MoE LLM 的结构、路由和质量–计算权衡	Paper
⬜	OLMo: Accelerating the Science of Language Models	ACL'24	开放数据、训练代码、checkpoint、日志和评测的端到端案例	Paper

Scaling Laws, Data & Pretraining

Status	Paper	Venue	Why It Matters	Links
⬜	Scaling Laws for Neural Language Models	arXiv'20	建立 loss 与参数量、数据量、训练算力之间的幂律关系	Paper
⬜	Training Compute-Optimal Large Language Models	NeurIPS'22	Chinchilla scaling 与 compute-optimal 参数/token 配比	Paper
⬜	Scaling Data-Constrained Language Models	NeurIPS'23	数据受限和重复 epoch 下的 scaling behavior	Paper
⬜	The Pile: An 800GB Dataset of Diverse Text for Language Modeling	arXiv'21	多域预训练语料组成、治理与 benchmark contamination	Paper
⬜	Deduplicating Training Data Makes Language Models Better	ACL'22	去重对记忆、评测污染、训练效率和质量的影响	Paper
⬜	Data Selection for Language Models via Importance Resampling	NeurIPS'23	DSIR：用目标分布重要性重采样选择预训练数据	Paper
⬜	DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretraining	NeurIPS'23	将多域数据配比转化为 group DRO 优化问题	Paper
⬜	The RefinedWeb Dataset for Falcon LLM: Outperforming Curated Corpora with Web Data, and Web Data Only	NeurIPS'23 (D&B)	大规模网页过滤与去重的生产级案例	Paper
⬜	Textbooks Are All You Need	arXiv'23	合成高质量数据、数据质量与小模型能力的交换关系	Paper
⬜	Dolma: An Open Corpus of Three Trillion Tokens for Language Model Pretraining Research	ACL'24	开放语料的来源、过滤、去重、PII 与治理流程	Paper
⬜	DataComp-LM: In Search of the Next Generation of Training Sets for Language Models	NeurIPS'24 (D&B)	受控比较过滤、去重和数据混合策略	Paper
⬜	The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale	NeurIPS'24 (D&B)	现代网页清洗、质量过滤与 FineWeb-Edu 配方	Paper

Attention Mechanisms & Variants

Status	Paper	Venue	Why It Matters	Links
✅	Attention Is All You Need	NeurIPS'17	用多头自注意力替代循环与卷积，建立可并行训练的 Transformer 架构	Paper / Note
✅	Big Bird: Transformers for Longer Sequences	NeurIPS'20	结合局部、随机与全局连接实现线性稀疏注意力，并分析其表达能力	Paper / Note
✅	FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness	NeurIPS'22	通过分块、算子融合与重计算减少 HBM 读写，实现不近似的高效 attention	Paper / Note
✅	FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning	ICLR'24	优化线程块与 warp 间工作划分，减少非矩阵计算和共享内存通信，提高 attention 吞吐	Paper / Note
✅	Flash-Decoding for Long-Context Inference	Blog'23	沿 KV 序列切分并合并部分 attention 结果，提高小 batch 长上下文 decode 的 GPU 利用率	Paper / Note
✅	A Survey of Efficient Attention Methods: Hardware-efficient, Sparse, Compact, and Linear Attention	Tech report'25	统一梳理硬件优化、稀疏、KV 压缩与线性 attention，区分计算和缓存优化的作用范围	Paper

Sparse Attention Algorithms & Training

这一分类覆盖稀疏注意力本身的架构、选择算法与训练方法。MSA 和 DSA 是其中的现代案例，而不是分类边界。

Architecture and Modern Sparse Attention

Status	Paper	Venue	Why It Matters	Links
✅ 📝	Fast Transformer Decoding: One Write-Head is All You Need	arXiv'19	MQA 原始论文；理解所有 Query heads 共享 KV 与 decode 带宽瓶颈	Paper / Note
⬜	GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints	EMNLP'23	建立 MHA–GQA–MQA 的连续关系；理解 MSA 为什么按 GQA group 独立选择	Paper
⬜	DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model	arXiv'24	MLA、latent KV compression、decoupled RoPE，以及 uk_proj / uo_proj 的矩阵吸收	Paper
⬜	SeerAttention: Learning Intrinsic Sparse Attention in Your LLMs	NeurIPS'25	可学习的 block gate、自蒸馏与 block-sparse kernel；最接近 MSA Indexer 的对照之一	Paper
⬜	Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention	ACL'25	同时研究压缩、选择、局部窗口与硬件对齐；理解 DSA 的算法–kernel 协同背景	Paper
⬜	MoBA: Mixture of Block Attention for Long-Context LLMs	NeurIPS'25	MoE 风格的 block routing；适合对比 MSA 的 block max-pooling 与 group-specific selection	Paper
⬜	MInference 1.0: Accelerating Pre-filling for Long-Context LLMs via Dynamic Sparse Attention	NeurIPS'24	无需重新训练的动态稀疏 prefill；对比学习式 Indexer 与预设 attention pattern	Paper
⬜	SpargeAttn: Accurate Sparse Attention Accelerating Any Model Inference	ICML'25	两阶段在线过滤与 softmax-aware pruning；理解推理期稀疏化的另一条路线	Paper
⬜	BLASST: Dynamic BLocked Attention Sparsity via Softmax Thresholding	MLSys'26	复用 online softmax 统计量跳过低贡献 blocks；无需训练并覆盖 MHA/GQA/MQA/MLA	Paper

现代架构阅读顺序： MQA → GQA → DeepSeek-V2/MLA → SeerAttention → Native Sparse Attention → MoBA → Quest → MInference → SpargeAttn → BLASST。

Algorithm and Training Foundations

这组论文用于从算子实现反向补齐稀疏注意力依赖的算法与训练概念：稀疏 pattern、内容路由、不可微 Top-k、知识蒸馏、Router 稳定性和稀疏归一化。

Sparse Pattern and Content Routing

Status	Paper	Venue	Why It Matters	Links
⬜	Generating Long Sequences with Sparse Transformers	arXiv'19	固定 factorized sparse pattern 与早期 block-sparse kernel；理解“规则但不自适应”的稀疏性	Paper
⬜	Longformer: The Long-Document Transformer	arXiv'20	local window + global token；理解局部先验和全局信息通路	Paper
⬜	Reformer: The Efficient Transformer	ICLR'20	LSH attention、bucket、排序和 reversible layers；理解动态候选集带来的数据重排成本	Paper
⬜	Efficient Content-Based Sparse Attention with Routing Transformers	TACL'21	online k-means 内容路由；连接固定 sparse pattern 与 learned Indexer	Paper

Big Bird 已在 Attention Mechanisms & Variants 中记录并标为已读，不在这里重复计数。

Differentiable Top-k and Discrete Selection

Status	Paper	Venue	Why It Matters	Links
⬜	Sparser is Faster and Less is More: Efficient Sparse Attention for Long-Range Transformers	arXiv'24	SparseK scoring network + differentiable Top-k；直接对比 MSA/DSA 的 hard Top-k + KL 路线	Paper
⬜	Differentiable Top-k Operator with Optimal Transport	NeurIPS'20	用熵正则 Optimal Transport 平滑 Top-k，理解连续松弛及其梯度	Paper
⬜	Fast, Differentiable and Sparse Top-k: A Convex Analysis Perspective	ICML'23	从凸优化构造可微且真正稀疏的 Top-k，并讨论 GPU/TPU-friendly 算法	Paper
⬜	Categorical Reparameterization with Gumbel-Softmax	ICLR'17	离散采样的经典连续松弛；理解 temperature、annealing 与 soft-train/hard-inference 差异	Paper

Distillation and Router Training

Status	Paper	Venue	Why It Matters	Links
✅ 📝	Distilling the Knowledge in a Neural Network	NeurIPS DL Workshop'15	soft target、temperature 与 KL；理解主 attention 如何作为 Indexer teacher	Paper / Note
⬜	Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity	JMLR'22	hard Top-1 routing、capacity 与 load-balancing loss；将 Indexer 理解成 memory router	Paper
⬜	ST-MoE: Designing Stable and Transferable Sparse Expert Models	arXiv'22	Router 稳定性、辅助损失与 router z-loss；理解小型路由器如何影响整个模型训练	Paper
⬜	Mixture-of-Experts with Expert Choice Routing	NeurIPS'22	expert 选择 token 而非 token 选择 expert；类比 q2k→k2q reverse index 与 KV-owner 调度	Paper

Sparse Normalization

Status	Paper	Venue	Why It Matters	Links
⬜	From Softmax to Sparsemax: A Sparse Model of Attention and Multi-Label Classification	ICML'16	在概率 simplex 上产生精确零值并保留可计算 Jacobian	Paper
⬜	Adaptively Sparse Transformers	EMNLP-IJCNLP'19	使用可学习的 $\alpha$-entmax 让不同 attention heads 自适应选择稠密或稀疏分布	Paper

算法/训练补课顺序： Sparse Transformer → Routing Transformer → SparseK Attention → Differentiable Top-k → Knowledge Distillation → Switch Transformer / ST-MoE → 回看 SeerAttention、NSA、MSA 与 DSA。

New Architectures

Status	Paper	Venue	Why It Matters	Links
⬜	Hyena Hierarchy: Towards Larger Convolutional Language Models	ICML'23	交替使用隐式长卷积与输入控制门控，探索低于二次复杂度的语言序列建模	Paper
⬜	RWKV: Reinventing RNNs for the Transformer Era	EMNLP Findings'23	将线性 attention 与循环计算结合，实现并行训练及常量状态的逐 token 推理	Paper
⬜	Retentive Network: A Successor to Transformer for Large Language Models	arXiv'23	用同一 retention 机制统一并行、循环和分块计算，兼顾训练效率与低成本解码	Paper
⬜	Mamba: Linear-Time Sequence Modeling with Selective State Spaces	COLM'24	以输入依赖的选择性状态空间和硬件感知扫描，提高线性时间模型的内容选择能力	Paper
⬜	Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality	ICML'24	建立 attention 与结构化状态空间的对偶关系，导出 Mamba-2 及高效分块算法	Paper
⬜	xLSTM: Extended Long Short-Term Memory	NeurIPS'24	通过指数门控与标量或矩阵记忆扩展 LSTM，探索可扩展的循环语言模型	Paper
✅	Gated Linear Attention Transformers with Hardware-Efficient Training	ICML'24	用数据依赖门控增强线性 attention，并以 IO 感知实现兼顾表达能力和训练吞吐	Paper / Note
✅	Kimi Linear: An Expressive, Efficient Attention Architecture	arXiv'25	用 Kimi Delta Attention 与混合架构改善线性 attention 的表达能力、长上下文效率和缓存成本	Paper
✅	DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models	arXiv'25	以 DSA 降低长上下文 attention 成本，并结合可扩展强化学习提升推理与工具使用能力	Paper
✅ 📝	DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence	Tech report'26	组合压缩稀疏与高度压缩 attention，降低百万 token 上下文的计算和 KV 缓存开销	Paper / Note

On-Device / Mobile

Status	Paper	Venue	Why It Matters	Links
✅	On-Device Training Under 256KB Memory	NeurIPS'22	通过量化感知梯度缩放、稀疏更新与编译期反向图裁剪，在 256KB 内存下训练模型	Paper
✅	PockEngine: Sparse and Efficient Fine-tuning in a Pocket	[MICRO'23](https://hanlab.mit.edu/projects/pockengine ; https://hanlab.mit.edu/projects/tinyml)	把稀疏反向传播与完整训练图编译结合，降低边缘设备微调的内存和运行时开销	[Paper](https://hanlab.mit.edu/projects/pockengine ; https://hanlab.mit.edu/projects/tinyml) / Note

📊 LLM Evaluation & Safety

Capability & Quality Evaluation

Status	Paper	Venue	Why It Matters	Links
⬜	Measuring Massive Multitask Language Understanding	ICLR'21	MMLU：用多学科考试题衡量知识与问题求解能力	Paper
⬜	Beyond the Imitation Game: Quantifying and Extrapolating the Capabilities of Language Models	TMLR'23	BIG-bench：异构任务与能力随规模变化	Paper
⬜	Holistic Evaluation of Language Models	TMLR'23	HELM：统一准确率、校准、鲁棒性、公平性、毒性和效率	Paper
⬜	TruthfulQA: Measuring How Models Mimic Human Falsehoods	ACL'22	将事实真实性与普通知识准确率区分开	Paper
⬜	Training Verifiers to Solve Math Word Problems	arXiv'21	GSM8K 与 verifier 路线，连接推理生成和可验证结果	Paper
⬜	Measuring Mathematical Problem Solving With the MATH Dataset	NeurIPS'21 (D&B)	竞赛数学、分步推理与严格答案评测	Paper
⬜	Evaluating Large Language Models Trained on Code	arXiv'21	HumanEval 与 pass@k，理解代码生成评测	Paper
⬜	GPQA: A Graduate-Level Google-Proof Q&A Benchmark	COLM'24	专家级科学问答，即使借助网络搜索仍具挑战；用于评估高难度推理能力	Paper
⬜	Instruction-Following Evaluation for Large Language Models	arXiv'23	IFEval：用可验证约束衡量 instruction following	Paper
⬜	Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena	NeurIPS'23 (D&B)	LLM judge、位置偏差、冗长偏差和 pairwise evaluation	Paper
⬜	Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference	ICML'24	真实用户盲测、成对偏好与 Elo/Bradley–Terry 排名	Paper
⬜	RewardBench: Evaluating Reward Models for Language Modeling	NAACL Findings'25	对话、推理、安全等场景的 reward model 基准	Paper
⬜	SWE-bench: Can Language Models Resolve Real-World GitHub Issues?	ICLR'24	真实仓库、issue、patch 与测试驱动的软件工程评测	Paper
⬜	LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code	ICLR'25	持续更新题目与时间切分，降低代码评测污染	Paper
⬜	Lessons from the Trenches on Reproducible Evaluation of Language Models	arXiv'24	prompt、tokenizer、版本和实现细节对评测复现的影响	Paper

Safety, Robustness & Red Teaming

Status	Paper	Venue	Why It Matters	Links
⬜	RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models	EMNLP Findings'20	开放式生成中的 toxicity 测量与 prompt 条件效应	Paper
⬜	Red Teaming Language Models with Language Models	EMNLP'22	用模型自动生成和筛选攻击，建立可扩展 red teaming	Paper
⬜	Universal and Transferable Adversarial Attacks on Aligned Language Models	arXiv'23	GCG 对抗后缀与可迁移 jailbreak 的基础工作	Paper
⬜	XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models	NAACL'24	同时衡量合理拒绝与过度拒绝	Paper
⬜	SafetyBench: Evaluating the Safety of Large Language Models	ACL'24	多类别、多语言的安全知识与行为评测	Paper
⬜	Towards Understanding Sycophancy in Language Models	ICLR'24	研究模型迎合用户观点的表现及训练信号来源	Paper
⬜	Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training	arXiv'24	研究条件触发的欺骗策略能否熬过安全训练	Paper
⬜	HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal	ICML'24	jailbreak、攻击方法与稳健拒绝的标准化评测	Paper
⬜	A StrongREJECT for Empty Jailbreaks	NeurIPS'24 (D&B)	jailbreak 评测必须衡量实际危害、完成度和拒绝质量	Paper
⬜	The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning	ICML'24	高风险双用途知识评测及其与 unlearning 的关系	Paper

🤖 LLM for Kernel Optimization

Status	Paper	Venue	Why It Matters	Links
✅	AVO: Agentic Variation Operators for Autonomous Evolutionary Search	arXiv'26	用自主编码 agent 替代固定变异与交叉算子，让进化搜索自适应生成和验证程序候选	Paper / Note
✅ 📝	CAKE: Compiler-Agent Co-Design for Frontier Kernel Evolution	arXiv'26	让 agent 编写显式硬件调度 IR，并结合验证、成本模型与局部诊断迭代优化 GPU kernel	Paper / Note
⬜	FlashInfer-Bench: Building the Virtuous Cycle for AI-driven LLM Systems	MLSys'26	以真实 LLM 工作负载、正确性验证和性能评测连接 AI kernel 生成与系统部署	Paper
⬜	Harness Engineering for LLM-Driven GPU Kernel Generation	arXiv'26	围绕约束、正确性验证、性能分析和候选筛选设计 harness，提高 LLM 生成 GPU kernel 的可靠性	Paper
⬜	Agentic Kernel Optimization: Generating State-of-the-Art GPU Kernels Without Hand-Written CUDA	Tech report'26	在正确性门控的 FlashInfer-Bench 流程中评估通用代码 agent 自动优化 MoE 与 DSA kernel 的能力	Paper

🧩 Agent Systems

Status	Paper	Venue	Why It Matters	Links
✅ 📝	SkVM: Revisiting Language VM for Skills across Heterogenous LLMs and Harnesses	arXiv'26	按模型与 harness 能力编译 skills，并用运行时代码固化和自适应重编译提高可移植性与执行效率	Paper / Note

🖥️ GPU Microarchitecture

Status	Paper	Venue	Why It Matters	Links
⬜	Understanding Latency Hiding on GPUs	PhD thesis'16	用简化微基准与性能模型解释 GPU 延迟隐藏，分析线程数、算术强度和吞吐的关系	Paper

📐 Math Foundations

Status	Paper	Venue	Why It Matters	Links
⬜	Categorical Foundations for CuTe Layouts	arXiv'26	用范畴与布局代数形式化 CuTe 的组合、乘积和除法，并验证其与实现一致	Paper

⚙️ Compiler

Status	Paper	Venue	Why It Matters	Links
✅	Honeycomb: Secure and Efficient GPU Executions via Static Validation	OSDI'23	在加载时静态验证 GPU 程序，结合 CPU TEE 将 OS 与驱动移出可信计算基	Paper / Note
✅	HIDA: A Hierarchical Dataflow Compiler for High-Level Synthesis	ASPLOS'24	用分层数据流表示与多级优化自动生成 FPGA 数据流架构，减少手工 HLS 调优	Paper / Note

🐧 Operating Systems

Status	Paper	Venue	Why It Matters	Links
✅	RedLeaf: Isolation and Communication in a Safe Operating System	OSDI'20	利用 Rust 类型与内存安全实现轻量隔离域、跨域零拷贝和驱动故障恢复	Paper / Note
✅	Theseus: an Experiment in Operating System Structure and State Management	OSDI'20	减少组件间状态耦合并用 Rust 编译器约束系统语义，支持 OS 在线演化与故障恢复	Paper
✅	Unikraft: Fast, Specialized Unikernels the Easy Way	EuroSys'21	以可组合微型库按应用裁剪 OS，降低构建高性能 unikernel 的移植和优化成本	Paper / Note
✅	The Demikernel Datapath OS Architecture for Microsecond-scale Datacenter Systems	SOSP'21	为 RDMA、DPDK 等异构 kernel-bypass 设备提供统一异步 API，降低微秒级数据中心 I/O 开销	Paper / Note

🛡️ Hypervisor & Virtualization

Status	Paper	Venue	Why It Matters	Links
✅	HyperBench: A Benchmark Suite for Virtualization Capabilities	SIGMETRICS / POMACS'19	用独立于特定 hypervisor 的微基准测量 VM 切换、地址转换等关键虚拟化能力	Paper / Note
✅	DuVisor: a User-level Hypervisor Through Delegated Virtualization	arXiv'22	通过硬件授权把 VM 退出等运行时操作下放用户态，减少内核攻击面与特权切换成本	Paper
✅	AvA: Accelerated Virtualization of Accelerators	ASPLOS'20	以 API 描述生成加速器虚拟化与远程调用支持，兼顾多种加速器的兼容性和性能	Paper
✅	Security and Performance in the Delegated User-level Virtualization	OSDI'23	分离 hypervisor 控制面与用户态 VM 处理面，用硬件授权同时降低攻击面和运行时开销	Paper / Note
✅	System Virtualization for Neural Processing Units	HotOS'23	提出 vNPU 抽象及分配、映射、调度机制，探索 NPU 多租户利用率与隔离的协同设计	Paper
✅	Nephele: Extending Virtualization Environments for Cloning Unikernel-based VMs	EuroSys'23	扩展虚拟化环境支持低开销克隆 unikernel VM，复用初始化状态并降低实例创建成本	Paper / Note
✅	Honeycomb: Secure and Efficient GPU Executions via Static Validation	OSDI'23	在加载时静态验证 GPU 程序，结合 CPU TEE 将 OS 与驱动移出可信计算基	Paper / Note

🔬 RISC-V

Status	Paper	Venue	Why It Matters	Links
✅	A First Look at RISC-V Virtualization from an Embedded Systems Perspective	IEEE TC'22 (online'21)	在 Rocket 上实现并评估 RISC-V H 扩展，分析嵌入式虚拟化中的中断、计时器与隔离开销	Paper
✅	CVA6 RISC-V Virtualization: Architecture, Microarchitecture, and Design Space Exploration	IEEE TVLSI'23	为 CVA6 实现硬件虚拟化并探索 GTLB、L2 TLB 等设计的性能、功耗和面积权衡	Paper

---

If you find this list helpful, feel free to ⭐ star this repo!