Inference 资料

1. P010
   PromptTuner: SLO-Aware Elastic System for LLM Prompt TuningarXiv:2603.05087 · Scholar 收录
2. P013
   DAWN: Dependency-Aware Fast Inference for Diffusion LLMsarXiv:2602.06953 · Scholar 收录
3. P018
   Inference-time Alignment via Sparse Junction SteeringarXiv:2602.21215 · Scholar 收录
4. P022
   Semantic-Aware Scheduling for GPU Clusters with Large Language ModelsarXiv:2510.03334 · Scholar 收录
5. P031
   DDiT: Dynamic Resource Allocation for Diffusion Transformer Model ServingarXiv:2506.13497 · Scholar 收录
6. P047
   UniTG: A Unified System for Efficient and Seamless Textual Graph LearningInternational Conference on Very Large Data Bases (VLDB), August, 2026
7. P054
   SpecForge: A Flexible and Efficient Open-Source Training Framework for Speculative DecodingInternational Conference on Machine Learning (ICML), July, 2026
8. P055
   CONCUR: High-Throughput Agentic Batch Inference of LLM via Congestion-Based Concurrency ControlInternational Conference on Machine Learning (ICML), July, 2026
9. P056
   DSB: Dynamic Sliding Block Scheduling for Diffusion LLMsInternational Conference on Machine Learning (ICML), July, 2026
10. P064
    SPPO: Making Million-Token LLM Training Practical on Modest GPU ClustersACM International Conference on Supercomputing (ICS), July, 2026
11. P070
    Di-PS: System-Algorithm Co-Design for Asynchronous and Heterogeneous Cross-cluster LLM Training at ScaleUSENIX Symposium on Networked Systems Design and Implementation (NSDI), May, 2026
12. P072
    ReSpec: Towards Optimizing Speculative Decoding in Reinforcement Learning SystemsAnnual Conference on Machine Learning and Systems (MLSys), May, 2026
13. P077
    DSA: Efficient Inference For Video Generation Models via Distributed Sparse AttentionInternational Conference on Learning Representations (ICLR), April, 2026
14. P097
    Impact-driven Context Filtering For Cross-file Code CompletionConference on Language Modeling (COLM), October, 2025
15. P114
    Rethinking Key-Value Cache Compression Techniques for Large Language Model ServingAnnual Conference on Machine Learning and Systems (MLSys), May, 2025
16. P129
    TorchGT: A Holistic System for Large-scale Graph Transformer TrainingInternational Conference for High Performance Computing, Networking, Storage, and Analysis (SC), November, 2024
17. P151
    Lins: Reducing Communication Overhead of ZeRO for Efficient LLM TrainingIEEE/ACM International Symposium on Quality of Service (IWQoS), June, 2024
18. P152
    Ymir: A Scheduler for Foundation Model Fine-tuning Workloads in DatacentersACM International Conference on Supercomputing (ICS), June, 2024
19. P153
    AutoSched: An Adaptive Self-configured Framework for Scheduling Deep Learning Training WorkloadsACM International Conference on Supercomputing (ICS), June, 2024
20. P158
    Sylvie: 3D-adaptive and Universal System for Large-scale Graph Neural Network TrainingIEEE International Conference on Data Engineering (ICDE), May, 2024
21. P164
    Characterization of Large Language Model Development in the DatacenterUSENIX Symposium on Networked Systems Design and Implementation (NSDI), April, 2024
22. P177
    Hydro: Surrogate-based Hyperparameter Tuning Service in DatacentersUSENIX Symposium on Operating Systems Design and Implementation (OSDI), July, 2023
23. P187
    Lucid: A Non-Intrusive, Scalable and Interpretable Scheduler for Deep Learning Training JobsACM International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS), Distinguished Paper Award , March, 2023
24. P193
    Titan: A Scheduler for Foundation Model Fine-tuning WorkloadsACM Symposium on Cloud Computing (SoCC), November, 2022
25. P194
    Tear Up the Bubble Boom: Lessons Learned From a Deep Learning Research and Development ClusterIEEE International Conference on Computer Design (ICCD), October, 2022
26. P208
    CHRONUS: A Novel Deadline-aware Scheduler for Deep Learning Training JobsACM Symposium on Cloud Computing (SoCC), November, 2021
27. P209
    Characterization and Prediction of Deep Learning Workloads in Large-Scale GPU DatacentersInternational Conference for High Performance Computing, Networking, Storage, and Analysis (SC), November, 2021
28. P239
    Collaborative Inference and Learning between Edge SLMs and Cloud LLMs: A Survey of Algorithms, Execution, and Open ChallengesAccepted by ACM Computing Surveys
29. P264
    ICEFROG: A Layer-Elastic Scheduling System for Deep Learning Training in GPU ClustersIEEE Transactions on Parallel and Distributed Systems, Volume: 36, Issue: 6, June 2025
30. P275
    UniSched: A Unified Scheduler for Deep Learning Training Jobs with Different User DemandsIEEE Transactions on Computers, Volume: 73, Issue: 6, June 2024
31. P282
    Deep Learning Workload Scheduling in GPU Datacenters: A SurveyACM Computing Surveys, Volume: 56, Issue: 6, January 2024
32. P300
    ASTRAEA: A Fair Deep Learning Scheduler for Multi-tenant GPU ClustersIEEE Transactions on Parallel and Distributed Systems, Volume: 33, Issue: 11, November 2022
33. 001
    Distributed Deep Neural Networks Over the Cloud, the Edge and End Devices (ICDCS 2017)Cloud-Edge Collaboration · Automatic Graph Partitioning
34. 002
    Dynamic Adaptive DNN Surgery for Inference Acceleration on the Edge (INFOCOM 2019)Cloud-Edge Collaboration · Automatic Graph Partitioning
35. 003
    Neurosurgeon: Collaborative Intelligence Between the Cloud and Mobile EdgeCloud-Edge Collaboration · Automatic Graph Partitioning
36. 004
    SPINN: Synergistic Progressive Inference of Neural Networks over Device and Cloud (MobiCom 2020)Cloud-Edge Collaboration · Automatic Graph Partitioning
37. 005
    Walle: An End-to-End, General-Purpose, and Large-Scale Production System for Device-Cloud Collaborative Machine LearningCloud-Edge Collaboration · Framework
38. 006
    A case for disaggregation of ML data processing (arXiv 2210.14826)Data Processing · Uncategorized
39. 007
    Disaggregating ML Input Data Processing at ScaleData Processing · Uncategorized
40. 008
    GoldMiner: Elastic Scaling of Training Data Pre-Processing Pipelines for Deep Learning (SIGMOD 2023)Data Processing · Uncategorized
41. 009
    Pecan: Cost-Efficient ML Data Preprocessing with Automatic Transformation Ordering and Hybrid PlacementData Processing · Uncategorized
42. 010
    Understanding Data Storage and Ingestion for Large-Scale Deep Recommendation Model Training (ISCA 2022)Data Processing · Uncategorized
43. 011
    DSA: Domain-Specific ArchitectureDeep Learning Compiler · Acronyms
44. 012
    MLIR: Scaling Compiler Infrastructure for Domain Specific Computation (CGO 2021)Deep Learning Compiler · System Architecture
45. 013
    TVM: An Automated End-to-End Optimizing Compiler for Deep LearningDeep Learning Compiler · System Architecture
46. 014
    AStitch: Enabling a New Multi-dimensional Optimization Space for Memory-Intensive ML Training and Inference on Modern SIMT Architectures (ASPLOS 2022)Deep Learning Compiler · Tensor Program Generation / General Tensor Program Generation
47. 015
    Ansor: Generating High-Performance Tensor Programs for Deep LearningDeep Learning Compiler · Tensor Program Generation / General Tensor Program Generation
48. 016
    Cocktailer: Analyzing and Optimizing Dynamic Control Flow in Deep LearningDeep Learning Compiler · Tensor Program Generation / General Tensor Program Generation
49. 017
    EINNET: Optimizing Tensor Programs with Derivation-Based TransformationsDeep Learning Compiler · Tensor Program Generation / General Tensor Program Generation
50. 018
    Effectively Scheduling Computational Graphs of Deep Neural Networks toward Their Domain-Specific AcceleratorsDeep Learning Compiler · Tensor Program Generation / General Tensor Program Generation
51. 019
    Welder: Scheduling Deep Learning Memory Access via Tile-graphDeep Learning Compiler · Tensor Program Generation / General Tensor Program Generation
52. 020
    Event Tensor: A Unified Abstraction for Compiling Dynamic MegakernelDeep Learning Compiler · Tensor Program Generation / Megakernel Compilation
53. 021
    Mirage Persistent Kernel: A Compiler and Runtime for Mega-Kernelizing Tensor Programs (arXiv:2512.22219)Deep Learning Compiler · Tensor Program Generation / Megakernel Compilation
54. 022
    An Introduction to Computational Networks and the Computational Network Toolkit (MSR-TR-2014-112)Deep Learning Framework · Uncategorized
55. 023
    Caffe: Convolutional Architecture for Fast Feature Embedding (arXiv 1408.5093)Deep Learning Framework · Uncategorized
56. 024
    Jittor: a novel deep learning framework with meta-operators and unified graph execution (Science China Information Sciences 2020)Deep Learning Framework · Uncategorized
57. 025
    MXNet: A Flexible and Efficient Machine Learning Library for Heterogeneous Distributed Systems (NIPS 2016 Workshop on MLSys)Deep Learning Framework · Uncategorized
58. 026
    OneFlow: Redesign the Distributed Deep Learning Framework from Scratch (arXiv 2110.15032)Deep Learning Framework · Uncategorized
59. 027
    Pathways: Asynchronous Distributed Dataflow for ML (MLSys 2022)Deep Learning Framework · Uncategorized
60. 028
    PyTorch: An Imperative Style, High-Performance Deep Learning Library (NeurIPS 2019)Deep Learning Framework · Uncategorized
61. 029
    TensorFlow: A System for Large-Scale Machine Learning (OSDI 2016)Deep Learning Framework · Uncategorized
62. 030
    XDL: An Industrial Deep Learning Framework for High-dimensional Sparse Data (DLP-KDD 2019)Deep Learning Framework · Uncategorized
63. 031
    EasyScale: Elastic Training with Consistent Accuracy and Improved Utilization on GPUsDeep Learning Training · Elastic Training
64. 032
    A Unified Architecture for Accelerating Distributed DNN Training in Heterogeneous GPU/CPU ClustersDeep Learning Training · Optimizing Network Communication
65. 033
    Hanayo: Harnessing Wave-like Pipeline Parallelism for Enhanced Large Model Training EfficiencyDeep Learning Training · Parallelism
66. 034
    One weird trick for parallelizing convolutional neural networks (arXiv 1404.599)Deep Learning Training · Parallelism
67. 035
    Supporting Very Large Models using Automatic Dataflow Graph PartitioningDeep Learning Training · Parallelism
68. 036
    Echo: Compiler-based GPU Memory Footprint Reduction for LSTM RNN Training (ISCA 2020)Deep Learning Training · Reduce GPU Memory Footprints / Compression
69. 037
    Gist: Efficient Data Encoding for Deep Neural Network Training (ISCA 2018)Deep Learning Training · Reduce GPU Memory Footprints / Compression
70. 038
    Gandiva: Introspective Cluster Scheduling for Deep LearningDeep Learning Training · Reduce GPU Memory Footprints / GPU Sharing
71. 039
    Salus: Fine-Grained GPU Sharing Primitives for Deep Learning ApplicationsDeep Learning Training · Reduce GPU Memory Footprints / GPU Sharing
72. 040
    Zico: Efficient GPU Memory Sharing for Concurrent DNN TrainingDeep Learning Training · Reduce GPU Memory Footprints / GPU Sharing
73. 041
    Capuchin: Tensor-based GPU Memory Management for Deep LearningDeep Learning Training · Reduce GPU Memory Footprints / Tensor Swapping / Recomputation
74. 042
    Checkmate: Breaking the Memory Wall with Optimal Tensor RematerializationDeep Learning Training · Reduce GPU Memory Footprints / Tensor Swapping / Recomputation
75. 043
    SuperNeurons: Dynamic GPU Memory Management for Training Deep Neural Networks (PPoPP 2018)Deep Learning Training · Reduce GPU Memory Footprints / Tensor Swapping / Recomputation
76. 044
    SwapAdvisor: Pushing Deep Learning Beyond the GPU Memory Limit via Smart SwappingDeep Learning Training · Reduce GPU Memory Footprints / Tensor Swapping / Recomputation
77. 045
    Training Deep Nets with Sublinear Memory Cost (arXiv 1604.06174)Deep Learning Training · Reduce GPU Memory Footprints / Tensor Swapping / Recomputation
78. 046
    vDNN: Virtualized Deep Neural Networks for Scalable, Memory-Efficient Neural Network Design (MICRO 2016)Deep Learning Training · Reduce GPU Memory Footprints / Tensor Swapping / Recomputation
79. 047
    DiT: Diffusion TransformerDiffusion Models · Acronyms
80. 048
    CAT-DM: Controllable Accelerated Virtual Try-on with Diffusion ModelDiffusion Models · Diffusion Model Serving / Image Editing
81. 049
    FlashPS: Efficient Generative Image Editing with Mask-aware Caching and SchedulingDiffusion Models · Diffusion Model Serving / Image Editing
82. 050
    Approximate Caching for Efficiently Serving Text-to-Image Diffusion ModelsDiffusion Models · Diffusion Model Serving / Image Generation
83. 051
    Cache Me if You Can: Accelerating Diffusion Models through Block CachingDiffusion Models · Diffusion Model Serving / Image Generation
84. 052
    DeepCache: Accelerating Diffusion Models for FreeDiffusion Models · Diffusion Model Serving / Image Generation
85. 053
    DiFlow: A System for Micro-Serving Text-to-Image Diffusion WorkflowsDiffusion Models · Diffusion Model Serving / Image Generation
86. 054
    DistriFusion: Distributed Parallel Inference for High-Resolution Diffusion ModelsDiffusion Models · Diffusion Model Serving / Image Generation
87. 055
    Katz: Efficient Workflow Serving for Diffusion Models with Many AdaptersDiffusion Models · Diffusion Model Serving / Image Generation
88. 056
    MixFusion: A Patch-Level Parallel Serving System for Mixed-Resolution Diffusion ModelsDiffusion Models · Diffusion Model Serving / Image Generation
89. 057
    PipeFusion: Displaced Patch Pipeline Parallelism for Inference of Diffusion Transformer Models (arXiv:2405.14430)Diffusion Models · Diffusion Model Serving / Image Generation
90. 058
    xDiT: an Inference Engine for Diffusion Transformers (DiTs) with Massive Parallelism (arXiv:2411.01738)Diffusion Models · Diffusion Model Serving / Image Generation
91. 059
    Fast Video Generation with Sliding Tile AttentionDiffusion Models · Diffusion Model Serving / Video Generation
92. 060
    FlexCache: Flexible Approximate Cache System for Video Diffusion (arXiv:2501.04012)Diffusion Models · Diffusion Model Serving / Video Generation
93. 061
    DiffusionPipe: Training Large Diffusion Models with Efficient PipelinesDiffusion Models · Diffusion Model Training
94. 062
    Cambricon-D: Full-Network Differential Acceleration for Diffusion ModelsDiffusion Models · Domain-Specific Accelerator (DSA)
95. 063
    X-Adapter: Adding Universal Compatibility of Plugins for Upgraded Diffusion ModelDiffusion Models · Supporting Add-on Modules
96. 064
    DLRM: Deep Learning Recommendation ModelDeep Learning Recommendation Model (DLRM) · Acronyms
97. 065
    DisaggRec: Architecting Disaggregated Systems for Large-Scale Personalized Recommendation (arXiv 2212.00939)Deep Learning Recommendation Model (DLRM) · DLRM Inference
98. 066
    Accelerating Neural Recommendation Training with Embedding SchedulingDeep Learning Recommendation Model (DLRM) · DLRM Training
99. 067
    Bagpipe: Accelerating Deep Recommendation Model TrainingDeep Learning Recommendation Model (DLRM) · DLRM Training
100. 068
     Heterogeneous Acceleration Pipeline for Recommendation System TrainingDeep Learning Recommendation Model (DLRM) · DLRM Training
101. 069
     EVStore: Storage and Caching Capabilities for Scaling Embedding Tables in Deep Recommendation SystemsDeep Learning Recommendation Model (DLRM) · GPU Cache
102. 070
     UGache: A Unified GPU Cache for Embedding-based Deep LearningDeep Learning Recommendation Model (DLRM) · GPU Cache
103. 071
     Ekko: A Large-Scale Deep Learning Recommender System with Low-Latency Model UpdateDeep Learning Recommendation Model (DLRM) · Model Update
104. 072
     AdaEmbed: Adaptive Embedding for Large-Scale Recommendation ModelsDeep Learning Recommendation Model (DLRM) · Pruning
105. 073
     HPO: Hyper-Parameter TuningHyper-Parameter Tuning (HPO) · Acronyms
106. 074
     CherryPick: Adaptively Unearthing the Best Cloud Configurations for Big Data AnalyticsHyper-Parameter Tuning (HPO) · HPO for Systems
107. 075
     Google Vizier: A Service for Black-Box Optimization (KDD 2017)Hyper-Parameter Tuning (HPO) · HPO for Systems
108. 076
     Morphling: Fast, Near-Optimal Auto-Configuration for Cloud-Native Model ServingHyper-Parameter Tuning (HPO) · HPO for Systems
109. 077
     Selecting the Best VM across Multiple Public Clouds: A Data-Driven Performance Modeling ApproachHyper-Parameter Tuning (HPO) · HPO for Systems
110. 078
     Elastic Hyperparameter Tuning on the CloudHyper-Parameter Tuning (HPO) · Optimizing HPO Workloads
111. 079
     Hydro: Surrogate-Based Hyperparameter Tuning Service in DatacentersHyper-Parameter Tuning (HPO) · Optimizing HPO Workloads
112. 080
     RubberBand: Cloud-based Hyperparameter TuningHyper-Parameter Tuning (HPO) · Optimizing HPO Workloads
113. 081
     LLM: Large Language ModelLarge Language Model (LLM) · Acronyms
114. 082
     LoRA: Low-Rank AdaptationLarge Language Model (LLM) · Acronyms
115. 083
     RL: Reinforcement LearningLarge Language Model (LLM) · Acronyms
116. 084
     RLHF: Reinforcement Learning from Human FeedbackLarge Language Model (LLM) · Acronyms
117. 085
     SDC: Silent Data CorruptionLarge Language Model (LLM) · Acronyms
118. 086
     PUZZLE: Efficiently Aligning Large Language Models through Light-Weight Context SwitchLarge Language Model (LLM) · LLM Alignment
119. 087
     ThunderAgent: A Fast, Simple, and Program-Aware Agentic Inference SystemLarge Language Model (LLM) · LLM Inference / Agentic Inference
120. 088
     LMPrefill: An Inference Engine for Prefill-only Workloads in Large Language Model ApplicationsLarge Language Model (LLM) · LLM Inference / Chunked Prefill
121. 089
     Taming Throughput-Latency Tradeoff in LLM Inference with Sarathi-ServeLarge Language Model (LLM) · LLM Inference / Chunked Prefill
122. 090
     Tender: Accelerating Large Language Models via Tensor Decomposition and Runtime RequantizationLarge Language Model (LLM) · LLM Inference / Compression
123. 091
     Fairness in Serving Large Language ModelsLarge Language Model (LLM) · LLM Inference / Fairness
124. 092
     Locality-aware Fair Scheduling in LLM Serving (arXiv:2501.14312)Large Language Model (LLM) · LLM Inference / Fairness
125. 093
     Cauchy: A Cost-Efficient LLM Serving System through Adaptive Heterogeneous Deployment (SoCC 2025)Large Language Model (LLM) · LLM Inference / Heterogeneous Deployment
126. 094
     Coral: Cost-Efficient Multi-LLM Serving over Heterogeneous Cloud GPUs (arXiv:2605.04357)Large Language Model (LLM) · LLM Inference / Heterogeneous Deployment
127. 095
     Demystifying Cost-Efficiency in LLM Serving over Heterogeneous GPUs (arXiv:2502.00722)Large Language Model (LLM) · LLM Inference / Heterogeneous Deployment
128. 096
     HexGen-2: Disaggregated Generative Inference of LLMs in Heterogeneous Environment (ICLR 2025)Large Language Model (LLM) · LLM Inference / Heterogeneous Deployment
129. 097
     HexGen: Generative Inference of Foundation Model over Heterogeneous Decentralized EnvironmentLarge Language Model (LLM) · LLM Inference / Heterogeneous Deployment
130. 098
     SageServe: Optimizing LLM Serving on Cloud Data Centers with Forecast Aware Auto-Scaling (SIGMETRICS Abstracts 2026)Large Language Model (LLM) · LLM Inference / Heterogeneous Deployment
131. 099
     SpotServe: Serving Generative Large Language Models on Preemptible InstancesLarge Language Model (LLM) · LLM Inference / Heterogeneous Deployment
132. 100
     ALISA: Accelerating Large Language Model Inference via Sparsity-Aware KV CachingLarge Language Model (LLM) · LLM Inference / KV Cache Management
133. 101
     CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge FusionLarge Language Model (LLM) · LLM Inference / KV Cache Management
134. 102
     CacheGen: KV Cache Compression and Streaming for Fast Large Language Model ServingLarge Language Model (LLM) · LLM Inference / KV Cache Management
135. 103
     DroidSpeak: KV Cache Sharing Across Fine-tuned Model VariantsLarge Language Model (LLM) · LLM Inference / KV Cache Management
136. 104
     Efficient Memory Management for Large Language Model Serving with PagedAttentionLarge Language Model (LLM) · LLM Inference / KV Cache Management
137. 105
     Jenga: Effective Memory Management for Serving LLM with HeterogeneityLarge Language Model (LLM) · LLM Inference / KV Cache Management
138. 106
     Prompt Cache: Modular Attention Reuse for Low-Latency InferenceLarge Language Model (LLM) · LLM Inference / KV Cache Management
139. 107
     Parrot: Efficient Serving of LLM-based Applications with Semantic VariableLarge Language Model (LLM) · LLM Inference / LLM-based Applications
140. 108
     SGLang: Efficient Execution of Structured Language Model ProgramsLarge Language Model (LLM) · LLM Inference / LLM-based Applications
141. 109
     Teola: Towards End-to-End Optimization of LLM-based ApplicationsLarge Language Model (LLM) · LLM Inference / LLM-based Applications
142. 110
     CaraServe: CPU-Assisted and Rank-Aware LoRA Serving for Generative LLM Inference (arXiv:2401.11240)Large Language Model (LLM) · LLM Inference / LoRA Serving
143. 111
     Punica: Multi-Tenant LoRA ServingLarge Language Model (LLM) · LLM Inference / LoRA Serving
144. 112
     S-LoRA: Serving Thousands of Concurrent LoRA AdaptersLarge Language Model (LLM) · LLM Inference / LoRA Serving
145. 113
     dLoRA: Dynamically Orchestrating Requests and Adapters for LoRA LLM ServingLarge Language Model (LLM) · LLM Inference / LoRA Serving
146. 114
     FlexGen: High-Throughput Generative Inference of Large Language Models with a Single GPULarge Language Model (LLM) · LLM Inference / Offloading
147. 115
     LLM in a flash: Efficient Large Language Model Inference with Limited Memory (arXiv 2312.11514)Large Language Model (LLM) · LLM Inference / Offloading
148. 116
     AlpaServe: Statistical Multiplexing with Model Parallelism for Deep Learning ServingLarge Language Model (LLM) · LLM Inference / Parallelism and Partitioning
149. 117
     DeepSpeed-Inference: Enabling Efficient Inference of Transformer Models at Unprecedented ScaleLarge Language Model (LLM) · LLM Inference / Parallelism and Partitioning
150. 118
     Efficiently Scaling Transformer InferenceLarge Language Model (LLM) · LLM Inference / Parallelism and Partitioning
151. 119
     EPIC: Efficient Position-Independent Context Caching for Serving Large Language Models (ICML 2025)Large Language Model (LLM) · LLM Inference / Position-Independent Caching (PIC)
152. 120
     DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model ServingLarge Language Model (LLM) · LLM Inference / Prefill-Decode (PD) Disaggregation
153. 121
     Inference without Interference: Disaggregate LLM Inference for Mixed Downstream Workloads (arXiv:2401.11181)Large Language Model (LLM) · LLM Inference / Prefill-Decode (PD) Disaggregation
154. 122
     Mooncake: A KVCache-centric Disaggregated Architecture for LLM Serving (FAST 2025)Large Language Model (LLM) · LLM Inference / Prefill-Decode (PD) Disaggregation
155. 123
     Prefill-as-a-Service: KVCache of Next-Generation Models Could Go Cross-Datacenter (arXiv:2604.15039)Large Language Model (LLM) · LLM Inference / Prefill-Decode (PD) Disaggregation
156. 124
     Splitwise: Efficient Generative LLM Inference Using Phase SplittingLarge Language Model (LLM) · LLM Inference / Prefill-Decode (PD) Disaggregation
157. 125
     FastServe: Iteration-Level Preemptive Scheduling for Large Language Model InferenceLarge Language Model (LLM) · LLM Inference / Request Scheduling
158. 126
     Llumnix: Dynamic Scheduling for Large Language Model ServingLarge Language Model (LLM) · LLM Inference / Request Scheduling
159. 127
     Orca: A Distributed Serving System for Transformer-Based Generative ModelsLarge Language Model (LLM) · LLM Inference / Request Scheduling
160. 128
     Cache-Craft: Managing Chunk-Caches for Efficient Retrieval-Augmented Generation (SIGMOD 2025)Large Language Model (LLM) · LLM Inference / Retrieval-Augmented Generation (RAG)
161. 129
     CacheFocus: Dynamic Cache Re-Positioning for Efficient Retrieval-Augmented Generation (arXiv:2502.11101)Large Language Model (LLM) · LLM Inference / Retrieval-Augmented Generation (RAG)
162. 130
     RAGCache: Efficient Knowledge Caching for Retrieval-Augmented Generation (arXiv:2404.12457)Large Language Model (LLM) · LLM Inference / Retrieval-Augmented Generation (RAG)
163. 131
     FaaScale: Unlocking Fast LLM Scaling for Serverless InferenceLarge Language Model (LLM) · LLM Inference / Serverless Inference
164. 132
     HydraServe: Minimizing Cold Start Latency for Serverless LLM Serving in Public CloudsLarge Language Model (LLM) · LLM Inference / Serverless Inference
165. 133
     ServerlessLLM: Low-Latency Serverless Inference for Large Language ModelsLarge Language Model (LLM) · LLM Inference / Serverless Inference
166. 134
     Deja Vu: Contextual Sparsity for Efficient LLMs at Inference TimeLarge Language Model (LLM) · LLM Inference / Sparsity
167. 135
     InfiniGen: Efficient Generative Inference of Large Language Models with Dynamic KV Cache ManagementLarge Language Model (LLM) · LLM Inference / Sparsity
168. 136
     PowerInfer: Fast Large Language Model Serving with a Consumer-grade GPULarge Language Model (LLM) · LLM Inference / Sparsity
169. 137
     Fast Inference from Transformers via Speculative DecodingLarge Language Model (LLM) · LLM Inference / Speculative Decoding
170. 138
     Online Speculative DecodingLarge Language Model (LLM) · LLM Inference / Speculative Decoding
171. 139
     SpecInfer: Accelerating Generative LLM Serving with Speculative Inference and Token Tree VerificationLarge Language Model (LLM) · LLM Inference / Speculative Decoding
172. 140
     Speculative Decoding with Big Little Decoder (NeurIPS 2023)Large Language Model (LLM) · LLM Inference / Speculative Decoding
173. 141
     Measuring Agents in ProductionLarge Language Model (LLM) · LLM Inference / Workload Characterization
174. 142
     TraceLab: Characterizing Coding Agent Workloads for LLM Serving (arXiv:2606.30560)Large Language Model (LLM) · LLM Inference / Workload Characterization
175. 143
     Accelerating the Training of Large Language Models using Efficient Activation Rematerialization and Optimal Hybrid ParallelismLarge Language Model (LLM) · LLM Training / Hybrid Parallelism
176. 144
     Alpa: Automating Inter- and Intra-Operator Parallelism for Distributed Deep LearningLarge Language Model (LLM) · LLM Training / Hybrid Parallelism
177. 145
     Beat the long tail: Distribution-Aware Speculative Decoding for RL TrainingLarge Language Model (LLM) · LLM Training / RL Post-Training
178. 146
     Disaggregated RL systems**: split rollout, inference, environment, reward, and training stages across best-fit resources.Large Language Model (LLM) · LLM Training / RL Post-Training
179. 147
     DynaRL: Flexible and Dynamic Scheduling of Large-Scale Reinforcement Learning TrainingLarge Language Model (LLM) · LLM Training / RL Post-Training
180. 148
     Fault tolerance**: isolate and recover failures across trainer, rollout, and control-plane roles.Large Language Model (LLM) · LLM Training / RL Post-Training
181. 149
     HetRL: Efficient Reinforcement Learning for LLMs in Heterogeneous EnvironmentsLarge Language Model (LLM) · LLM Training / RL Post-Training
182. 150
     Heterogeneous environments**: make RL training efficient across mixed GPU generations and hardware capabilities.Large Language Model (LLM) · LLM Training / RL Post-Training
183. 151
     History Doesn't Repeat Itself but Rollouts Rhyme: Accelerating Reinforcement Learning with RhymeRLLarge Language Model (LLM) · LLM Training / RL Post-Training
184. 152
     RLinf: Flexible and Efficient Large-Scale Reinforcement Learning via Macro-to-Micro Flow TransformationLarge Language Model (LLM) · LLM Training / RL Post-Training
185. 153
     ReSpec: Towards Optimizing Speculative Decoding in Reinforcement Learning SystemsLarge Language Model (LLM) · LLM Training / RL Post-Training
186. 154
     ReaL: Efficient RLHF Training of Large Language Models with Parameter ReallocationLarge Language Model (LLM) · LLM Training / RL Post-Training
187. 155
     RobustRL: Role-Based Fault Tolerance System for RL Post-TrainingLarge Language Model (LLM) · LLM Training / RL Post-Training
188. 156
     RollArt: Disaggregated Multi-Task Agentic RL Training at ScaleLarge Language Model (LLM) · LLM Training / RL Post-Training
189. 157
     RollPacker: Taming Long-Tail Rollouts for RL Post-Training with Tail BatchingLarge Language Model (LLM) · LLM Training / RL Post-Training
190. 158
     Rollout latency and long-tail mitigation**: predict, batch, reuse, or otherwise reduce long-tail rollout work.Large Language Model (LLM) · LLM Training / RL Post-Training
191. 159
     Seer: Online Context Learning for Fast Synchronous LLM Reinforcement LearningLarge Language Model (LLM) · LLM Training / RL Post-Training
192. 160
     Speculative decoding for RL**: adapt draft-and-verify generation to RL training constraints such as drafter staleness and rollout distribution shift.Large Language Model (LLM) · LLM Training / RL Post-Training
193. 161
     Taming the Long-Tail: Efficient Reasoning RL Training with Adaptive DrafterLarge Language Model (LLM) · LLM Training / RL Post-Training
194. 162
     Weave: Efficient Co-Scheduling for Disaggregated RL Post-TrainingLarge Language Model (LLM) · LLM Training / RL Post-Training
195. 163
     Workflow scheduling and resource reallocation**: reshape RL pipelines or dynamically move compute, memory, communication, and parameters across roles.Large Language Model (LLM) · LLM Training / RL Post-Training
196. 164
     Bamboo: Making Preemptible Instances Resilient for Affordable Training of Large DNNsLarge Language Model (LLM) · LLM Training / Reliability and Fault Tolerance
197. 165
     Empirical reliability studies**: characterize production failure modes and operational mitigations from large training runs.Large Language Model (LLM) · LLM Training / Reliability and Fault Tolerance
198. 166
     Gemini: Fast Failure Recovery in Distributed Training with In-Memory CheckpointsLarge Language Model (LLM) · LLM Training / Reliability and Fault Tolerance
199. 167
     Holmes: Localizing Irregularities in LLM Training with Mega-scale GPU ClustersLarge Language Model (LLM) · LLM Training / Reliability and Fault Tolerance
200. 168
     Large-Scale AI Infra Reliability: Challenges, Strategies, and Llama 3 Training Experience (DSN-S 2025)Large Language Model (LLM) · LLM Training / Reliability and Fault Tolerance
201. 169
     MegaScale-Omni: A Hyper-Scale, Workload-Resilient System for MultiModal LLM Training in ProductionLarge Language Model (LLM) · LLM Training / Reliability and Fault Tolerance
202. 170
     MegaScale: Scaling Large Language Model Training to More Than 10,000 GPUsLarge Language Model (LLM) · LLM Training / Reliability and Fault Tolerance
203. 171
     Oobleck: Resilient Distributed Training of Large Models Using Pipeline TemplatesLarge Language Model (LLM) · LLM Training / Reliability and Fault Tolerance
204. 172
     Production reliability infrastructure**: make failures observable, diagnosable, and routinely recoverable at 10K+ GPU scale.Large Language Model (LLM) · LLM Training / Reliability and Fault Tolerance
205. 173
     Recovery mechanisms**: reduce lost work after failures through redundancy, checkpoint placement, or pre-planned reconfiguration.Large Language Model (LLM) · LLM Training / Reliability and Fault Tolerance
206. 174
     Robust LLM Training Infrastructure at ByteDanceLarge Language Model (LLM) · LLM Training / Reliability and Fault Tolerance
207. 175
     SDCs in the Wild: Characterizing and Diagnosing SDC-Defective GPUs in Production LLM TrainingLarge Language Model (LLM) · LLM Training / Reliability and Fault Tolerance
208. 176
     Safeguarding LLM Training at Scale: Online SDC Detection and Insights from 35 Million GPU HoursLarge Language Model (LLM) · LLM Training / Reliability and Fault Tolerance
209. 177
     Workload resilience**: absorb dynamic workload variation before it turns into large efficiency loss or training instability.Large Language Model (LLM) · LLM Training / Reliability and Fault Tolerance
210. 178
     Morphling: Fast, Near-Optimal Auto-Configuration for Cloud-Native Model ServingModel Serving · Auto-Configuration for Model Serving
211. 179
     Serving Unseen Deep Learning Models with Near-Optimal Configurations: a Fast Adaptive Search ApproachModel Serving · Auto-Configuration for Model Serving
212. 180
     Clipper: A Low-Latency Online Prediction Serving SystemModel Serving · Model Serving Systems
213. 181
     INFaaS: Automated Model-less Inference ServingModel Serving · Model Serving Systems
214. 182
     Microsecond-scale Preemption for Concurrent GPU-accelerated DNN InferencesModel Serving · Model Serving Systems
215. 183
     Paella: Low-latency Model Serving with Software-defined GPU SchedulingModel Serving · Model Serving Systems
216. 184
     TensorFlow-Serving: Flexible, High-Performance ML Serving (NIPS 2017 Workshop on ML Systems)Model Serving · Model Serving Systems
217. 185
     Usher: Holistic Interference Avoidance for Resource Optimized ML InferenceModel Serving · Model Serving Systems
218. 186
     A Survey of Large-Scale Deep Learning Serving System Optimization: Challenges and Opportunities (arXiv 2111.14247)Model Serving · Survey
219. 187
     A Survey of Multi-Tenant Deep Learning Inference on GPU (MLSys 2022 Workshop on Cloud Intelligence / AIOps)Model Serving · Survey
220. 188
     Accelerating Distributed MoE Training and Inference with LinaMixture of Experts (MoE) · MoE Inference
221. 189
     Optimizing Dynamic Neural Networks with BrainstormMixture of Experts (MoE) · MoE Inference
222. 190
     Accelerating Distributed MoE Training and Inference with LinaMixture of Experts (MoE) · MoE Training
223. 191
     Janus: A Unified Distributed Training Framework for Sparse Mixture-of-Experts ModelsMixture of Experts (MoE) · MoE Training
224. 192
     SmartMoE: Efficiently Training Sparsely-Activated Models through Combining Offline and Online ParallelizationMixture of Experts (MoE) · MoE Training
225. 193
     Mixtral-8x7BMixture of Experts (MoE) · Models
226. 194
     DL: Deep LearningResource Scheduler · Acronyms
227. 195
     ML: Machine LearningResource Scheduler · Acronyms
228. 196
     AlloX: Compute Allocation in Hybrid ClustersResource Scheduler · Scheduling for DL Training Workloads
229. 197
     AntMan: Dynamic Scaling on GPU Clusters for Deep LearningResource Scheduler · Scheduling for DL Training Workloads
230. 198
     Astraea: A Fair Deep Learning Scheduler for Multi-Tenant GPU Clusters (TPDS 2021)Resource Scheduler · Scheduling for DL Training Workloads
231. 199
     Balancing Efficiency and Fairness in Heterogeneous GPU Clusters for Deep LearningResource Scheduler · Scheduling for DL Training Workloads
232. 200
     Blox: A Modular Toolkit for Deep Learning SchedulersResource Scheduler · Scheduling for DL Training Workloads
233. 201
     CASSINI: Network-Aware Job Scheduling in Machine Learning ClustersResource Scheduler · Scheduling for DL Training Workloads
234. 202
     Gandiva: Introspective Cluster Scheduling for Deep LearningResource Scheduler · Scheduling for DL Training Workloads
235. 203
     Heterogeneity-Aware Cluster Scheduling Policies for Deep Learning WorkloadsResource Scheduler · Scheduling for DL Training Workloads
236. 204
     HiveD: Sharing a GPU Cluster for Deep Learning with GuaranteesResource Scheduler · Scheduling for DL Training Workloads
237. 205
     Interference-aware Multiplexing for Deep Learning in GPU Clusters: A Middleware ApproachResource Scheduler · Scheduling for DL Training Workloads
238. 206
     Looking Beyond GPUs for DNN Scheduling on Multi-Tenant ClustersResource Scheduler · Scheduling for DL Training Workloads
239. 207
     Lucid: A Non-intrusive, Scalable and Interpretable Scheduler for Deep Learning Training JobsResource Scheduler · Scheduling for DL Training Workloads
240. 208
     Lyra: Elastic Scheduling for Deep Learning ClustersResource Scheduler · Scheduling for DL Training Workloads
241. 209
     MAPA: Multi-Accelerator Pattern Allocation Policy for Multi-Tenant GPU Servers (SC 2021)Resource Scheduler · Scheduling for DL Training Workloads
242. 210
     Multi-Resource Interleaving for Deep Learning TrainingResource Scheduler · Scheduling for DL Training Workloads
243. 211
     Optimus: An Efficient Dynamic Resource Scheduler for Deep Learning Clusters (EuroSys 2018)Resource Scheduler · Scheduling for DL Training Workloads
244. 212
     Pollux: Co-adaptive Cluster Scheduling for Goodput-Optimized Deep LearningResource Scheduler · Scheduling for DL Training Workloads
245. 213
     Shockwave: Fair and Efficient Cluster Scheduling for Dynamic Adaptation in Machine LearningResource Scheduler · Scheduling for DL Training Workloads
246. 214
     Sia: Heterogeneity-aware, goodput-optimized ML-cluster schedulingResource Scheduler · Scheduling for DL Training Workloads
247. 215
     Singularity: Planet-Scale, Preemptive and Elastic Scheduling of AI Workloads (arXiv 2202.07848)Resource Scheduler · Scheduling for DL Training Workloads
248. 216
     Themis: Fair and Efficient GPU Cluster Scheduling (NSDI 2020)Resource Scheduler · Scheduling for DL Training Workloads
249. 217
     Tiresias: A GPU Cluster Manager for Distributed Deep LearningResource Scheduler · Scheduling for DL Training Workloads
250. 218
     Topology-Aware GPU Scheduling for Learning Workloads in Cloud Environments (SC 2017)Resource Scheduler · Scheduling for DL Training Workloads
251. 219
     SLAQ: Quality-Driven Scheduling for Distributed Machine LearningResource Scheduler · Scheduling for General ML Training Workloads
252. 220
     Deep Learning Workload Scheduling in GPU Datacenters: Taxonomy, Challenges and Vision (arXiv 2205.11913)Resource Scheduler · Survey
253. 221
     Analysis of Large-Scale Multi-Tenant GPU Clusters for DNN Training Workloads (ATC 2019)Resource Scheduler · Trace Analysis
254. 222
     Characterization and Prediction of Deep Learning Workloads in Large-Scale GPU Datacenters (SC 2021)Resource Scheduler · Trace Analysis
255. 223
     Characterizing Deep Learning Training Workloads on Alibaba-PAI (IISWC 2019)Resource Scheduler · Trace Analysis
256. 224
     MLaaS in the Wild: Workload Analysis and Scheduling in Large-Scale Heterogeneous GPU ClustersResource Scheduler · Trace Analysis
257. 225
     MSRL: Distributed Reinforcement Learning with Dataflow FragmentsReinforcement Learning (RL) · Uncategorized
258. 226
     Ray: A Distributed Framework for Emerging AI ApplicationsReinforcement Learning (RL) · Uncategorized
259. 2026
     [随笔]什么是AI Native2026-07-11 · 随笔/观点
260. 2026
     详细谈谈DSpark投机解码的原理2026-07-04 · 完整技术章节
261. 2026
     谈谈2026年ScaleUP标准的变化2026-06-26 · 完整系统章节
262. 2026
     用Agent来分析Nvidia RTX系列芯片2026-06-22 · 完整研究工作流章节
263. 2026
     谈谈如何使用Agentic Workflow进行大模型全栈研究2026-06-08 · 完整研究工作流章节
264. 2026
     谈谈 AWS 的随机图网络拓扑2026-06-03 · 完整网络系统章节
265. 2026
     使用Agent做Nvidia GPU芯片微架构分析2026-05-29 · 完整研究工作流章节
266. 2026
     [大模型中的数学-103] Grassmann流形视角下的Attention算法2026-05-25 · 待人工标注
267. 2026
     谈谈GFD vs cudaMemcpBatchAsync2026-05-25 · 完整技术章节
268. 2026
     [大模型中的数学-102] 范畴论视角下的Attention算法2026-05-24 · 待人工标注
269. 2026
     [大模型中的数学-101] 大语言模型中的“语言”: 是符号的幻影, 还是意义的载体?2026-05-24 · 待人工标注
270. 2026
     Agent写的一个高性能Host-to-Device 传输库2026-05-23 · 完整研究工作流章节
271. 2026
     DeepSeek-V4详细分析(2): MegaMoE2026-05-14 · 模型架构解读
272. 2026
     谈谈OpenAI 发布的MRC2026-05-07 · 待人工标注
273. 2026
     DeepEPv2分析(3)-EP Direct Dispatch/Combine Kernel2026-05-05 · 完整技术章节
274. 2026
     DeepEPv2分析(4)-EP Hybrid Dispatch Combine Kernel2026-05-05 · 完整技术章节
275. 2026
     DeepEPv2分析(1)2026-05-04 · 完整技术章节
276. 2026
     DeepEPv2分析(2)-EP Overview2026-05-04 · 完整技术章节
277. 2026
     DeepSeek-V4详细分析(1): 算法和模型结构2026-05-01 · 技术结构解读
278. 2026
     NCCL Gin & Symmetric Memory2026-05-01 · 完整技术章节
279. 2026
     谈谈Agentic时代的架构及互连2026-04-05 · 完整研究工作流章节
280. 2026
     Token交易所专题研究：AI能力商品化驱动新型交易所基础设施形成2026-03-22 · 待人工标注
281. 2026
     谈谈Kimi Attention Res的工作2026-03-21 · 技术机制解读
282. 2026
     Groq LPU架构详解2026-03-19 · 待人工标注
283. 2026
     详细谈谈Rubin + Groq 3 LPU架构2026-03-19 · 待人工标注
284. 2026
     AI需要新的数学语言?2026-03-16 · 待人工标注
285. 2026
     谈谈Feynman+LPU的架构2026-03-14 · 待人工标注
286. 2026
     谈谈DeepSeek DualPath以及ScaleOut直接接入存储2026-02-27 · 完整系统章节
287. 2026
     谈谈光互连的一些问题2026-02-26 · 完整系统章节
288. 2026
     谈谈Attention SInk及未来Attention算法设计2026-02-19 · 待人工标注
289. 2026
     从芯片架构的视角谈谈ScaleUP, smart or dumb2026-02-13 · 完整系统章节
290. 2026
     大模型推理系统(2)--体系结构的视角2026-02-09 · 完整系统章节
291. 2026
     从Moltbook社交到Agent Swarm2026-02-01 · 研究方法与协作讨论
292. 2026
     基于可训练先验的广义最优传输注意力机制2026-01-31 · 待人工标注
293. 2026
     再来谈谈Sparse Attention2026-01-26 · 技术机制解读
294. 2026
     谈谈基于ScaleUP的存储扩展和三网融合2026-01-25 · 待人工标注
295. 2026
     世界终究是属于草台班子的-22026-01-23 · 待人工标注
296. 2026
     猜一猜DeepSeek-V4的模型架构2026-01-18 · 推测性分析
297. 2026
     “漫”谈RDMA现代化2026-01-16 · 待人工标注
298. 2026
     谈谈DeepSeek Engram: Conditional Memory2026-01-14 · 技术机制解读
299. 2026
     谈谈mHC-Lite:无需使用Sinkhorn-Knopp迭代的算法2026-01-13 · 待人工标注
300. 2026
     大语言模型推理硬件的挑战与研究方向2026-01-12 · 完整系统章节
301. 2026
     Nvidia如何整合吸收Groq的技术2026-01-07 · 待人工标注
302. 2026
     谈谈DeepSeek mHC2026-01-03 · 待人工标注
303. 2025
     谈谈那个被NV看上值20B的Groq2025-12-30 · 待人工标注
304. 2025
     谈谈字节的veRoCE, 但为啥不叫VeiWARP呢?2025-12-21 · 待人工标注
305. 2025
     谈谈AWS Trainium3/4 及未来的AI Infra演进2025-12-19 · 待人工标注
306. 2025
     旧文重发: 谈谈GPU云的经营风险和流动性管理2025-12-11 · 待人工标注
307. 2025
     谈谈Nested Learning2025-12-08 · 待人工标注
308. 2025
     谈谈Scale Across网络2025-12-04 · 待人工标注
309. 2025
     真的该看空NV么? 可能并不是!2025-11-26 · 待人工标注
310. 2025
     英伟达已经做的很好了? 吗?2025-11-22 · 待人工标注
311. 2025
     谈谈GPU微架构及系统架构的演进2025-11-11 · 待人工标注
312. 2025
     谈谈未来Attention算法的选择, Full, Sparse or Linear ?2025-11-08 · 技术比较
313. 2025
     Inside Nvidia GPU: 谈谈Blackwell的不足并预测一下Rubin的微架构2025-11-05 · 待人工标注
314. 2025
     Tensor-103.4 Blackwell GEMM2025-11-01 · 待人工标注
315. 2025
     Tensor-103.3 Hopper Persistent Kernel2025-10-27 · 待人工标注
316. 2025
     Tensor-103.2: Hopper GEMM2025-10-22 · 待人工标注
317. 2025
     Tensor-103.1: Basic GEMM2025-10-18 · 待人工标注
318. 2025
     谈谈ESUN, SUE和UALink2025-10-18 · 完整系统章节
319. 2025
     Tilelang-2: 基本操作2025-10-12 · 待人工标注
320. 2025
     Tensor-102: GEMV2025-10-09 · 待人工标注
321. 2025
     谈谈 Hierarchical Sparse Attention (HSA)2025-10-06 · 待人工标注
322. 2025
     Tensor-101: Element-Wise Add2025-10-05 · 待人工标注
323. 2025
     Tilelang-1: Introduction2025-10-02 · 待人工标注
324. 2025
     CuteDSL-2: 基本操作2025-09-30 · 待人工标注
325. 2025
     学习一下DeepSeek-V3.22025-09-30 · 模型解读
326. 2025
     谈谈阿里的AI Infra磐久128卡超节点2025-09-25 · 完整系统章节
327. 2025
     CuTe Layout代数-1: Overview2025-09-24 · 待人工标注
328. 2025
     CuteDSL-1: Introduction2025-09-21 · 待人工标注
329. 2025
     谈谈UB, UALink以及NV和Intel的合作2025-09-20 · 完整系统章节
330. 2025
     详细分析一下Nvidia Rubin CPX2025-09-17 · 待人工标注
331. 2025
     学习一下Linear Layout2025-09-15 · 待人工标注
332. 2025
     从GPU缓存的视角看芯片设计和互连2025-09-13 · 完整系统章节
333. 2025
     谈谈芯片设计的一些取舍2025-09-07 · 待人工标注
334. 2025
     Firefly: 数据中心内的高精度时钟同步2025-09-05 · 待人工标注
335. 2025
     谈谈Google Falcon的可靠传输论文并对比分析CIPU eRDMA2025-09-05 · 完整系统章节
336. 2025
     谈谈RDMA和ScaleUP的可靠传输2025-08-31 · 完整系统章节
337. 2025
     Hotchip-2025:Day2 AI芯片和光互连2025-08-27 · 完整系统章节
338. 2025
     Hotchip-2025: Day1网络篇2025-08-26 · 完整系统章节
339. 2025
     Hotchip-2025: Day0 Tutorials记录2025-08-25 · 待人工标注
340. 2025
     谈谈Transformer的一些演进: UT,MoD,MoR...2025-08-25 · 待人工标注
341. 2025
     谈谈AutoGLM 2.0以及背后的Agent Infra2025-08-22 · 完整研究工作流章节
342. 2025
     大模型时代的数学基础(9)- SDPA和最优传输, 强化学习及信息几何的联系2025-08-21 · 待人工标注
343. 2025
     再来谈谈UltraEthernet的设计原则2025-08-17 · 完整系统章节
344. 2025
     谈谈搜广推场景的RDMA2025-08-11 · 待人工标注
345. 2025
     从GPT-OSS谈谈大模型算法和Infra演进2025-08-07 · 独立系统解释
346. 2025
     从一个基金经理的视角谈谈架构师的素养2025-08-06 · 待人工标注
347. 2025
     再来谈谈AFD2025-08-05 · 待人工标注
348. 2025
     从算法-系统-硬件协同的视角谈谈AFD2025-08-03 · 待人工标注
349. 2025
     谈谈Agentic AI对Infra的需求2025-07-22 · 基础设施研究讨论
350. 2025
     同构视角下的异构计算: 谈谈AI的软硬件交付界面, GPGPU or DSA2025-07-19 · 完整系统章节
351. 2025
     全球首个A股金融智能体?2025-07-18 · 待人工标注
352. 2025
     谈谈Kimi K2, 第一个开源的1T参数模型2025-07-12 · 独立系统解释
353. 2025
     谈谈AWS GB200实例, 顺便谈谈各种自研NPU的困境2025-07-10 · 待人工标注
354. 2025
     谈谈GPU云的经营风险和流动性管理2025-07-05 · 待人工标注
355. 2025
     从ODCC夏季会议再来谈谈ScaleUP的一些进展2025-07-01 · 完整系统章节
356. 2025
     学习华为CloudMatrix 384上LLM推理2025-06-18 · 待人工标注
357. 2025
     关于3D Stacking DRAM GPU的一些想法2025-06-14 · 完整系统章节
358. 2025
     谈谈SDR-RDMA, 所谓的软件定义的可靠性2025-06-09 · 完整系统章节
359. 2025
     从AI落地的视角看看Infra的需求2025-06-08 · 待人工标注
360. 2025
     再来谈谈ScaleUP和ScaleOut2025-05-31 · 完整系统章节
361. 2025
     锐评某友商说传统云还在卖铁: 从金融的视角谈云计算及其流动性管理2025-05-28 · 待人工标注
362. 2025
     一个关于AI编程/Agent的讨论2025-05-25 · 研究方法与编程讨论
363. 2025
     谈谈NV在ComputeX Keynote2025-05-20 · 待人工标注
364. 2025
     谈谈DeepSeek对AI架构硬件的思考2025-05-16 · 待人工标注
365. 2025
     CUDA Green Context2025-05-07 · 完整系统章节
366. 2025
     现代NVidia GPU架构2025-05-05 · 待人工标注
367. 2025
     民科: 预测一下未来五年大模型的架构?2025-05-03 · 待人工标注
368. 2025
     谈谈以太网GPU Scale-UP的工作EthLink2025-04-30 · 完整系统章节
369. 2025
     谈谈大模型可解释性(2)2025-04-30 · 待人工标注
370. 2025
     Test-Time ScalingLaw又失效了? 算力约束下的国产模型突破在何方?2025-04-27 · 待人工标注
371. 2025
     高带宽处理器, 又在基于PNM卷访存密集型算子?2025-04-26 · 待人工标注
372. 2025
     CUDA-Next: 基于任务的张量计算的DSL?2025-04-26 · 待人工标注
373. 2025
     谈谈腾讯DeepEP的优化2025-04-24 · 待人工标注
374. 2025
     再来谈谈GPU体系结构及互联2025-04-20 · 完整系统章节
375. 2025
     从Sutton苦涩的教训来谈谈AI基础设施2025-04-15 · 待人工标注
376. 2025
     谈谈GPU的内存模型及互联网络设计2025-04-13 · 完整系统章节
377. 2025
     谈谈AI时代的CIPU 2.02025-04-10 · 完整系统章节
378. 2025
     谈谈Llama4 和DeepSeek GRM2025-04-06 · 待人工标注
379. 2025
     谈谈字节的Attention/Expert分离2025-04-05 · 独立系统解释
380. 2025
     大模型推理系统(1)--先从分布式系统DEC VAXCluster谈起2025-04-02 · 完整系统章节
381. 2025
     从GTC25谈谈GPU互联(修正版)2025-03-31 · 完整系统章节
382. 2025
     从GTC25谈谈GPU互联2025-03-30 · 待人工标注
383. 2025
     从3FS性能谈谈数据密集型应用上云的挑战和机会2025-03-29 · 完整系统章节
384. 2025
     Tensor-011 Blackwell TensorCore2025-03-26 · 待人工标注
385. 2025
     从GTC25谈谈大模型架构的演进2025-03-23 · 待人工标注
386. 2025
     谈谈3FS的写性能优化的难点2025-03-21 · 完整系统章节
387. 2025
     DeepSeek-V3/R1推理效率分析(2)2025-03-21 · 独立系统解释
388. 2025
     再以eRDMA的视角谈谈DeepEP和3FS2025-03-19 · 完整系统章节
389. 2025
     谈谈三万亿的破绽(6): 从GTC25上分析2025-03-19 · 待人工标注
390. 2025
     DeepSeek-V3/R1推理效率分析-Blackwell性能估计(V0.4)2025-03-18 · 独立系统解释
391. 2025
     DeepSeek 3FS解读与源码分析（4）：Meta Service解读2025-03-17 · 完整系统章节
392. 2025
     DeepSeek 3FS解读与源码分析（5）：客户端解读2025-03-17 · 完整系统章节
393. 2025
     DeepSeek-V3/R1推理效率分析(v0.17)2025-03-16 · 独立系统解释
394. 2025
     DeepSeek-V3/R1推理效率分析(v0.15)2025-03-15 · 独立系统解释
395. 2025
     DeepSeek-V3/R1推理效率分析2025-03-14 · 独立系统解释
396. 2025
     从DeepSeek MoE专家负载均衡谈起2025-03-10 · 待人工标注
397. 2025
     DeepSeek 3FS解读与源码分析（3）：Storage模块解读2025-03-09 · 完整系统章节
398. 2025
     谈谈三万亿的破绽(5): 以基金经理的视角2025-03-07 · 待人工标注
399. 2025
     RDMA这十年的反思4: 从DeepSeek的3FS和DeepEP谈起2025-03-07 · 完整系统章节
400. 2025
     谈谈字节的COMET, 另一个细粒度的MoE通信和计算Overlap方案2025-03-06 · 独立系统解释
401. 2025
     DeepSeek 3FS解读与源码分析（2）：网络通信模块分析2025-03-02 · 完整系统章节
402. 2025
     DeepSeek 3FS解读与源码分析（1）：高效训练之道2025-03-02 · 完整系统章节
403. 2025
     谈谈DeepSeek原厂推理方案2025-03-02 · 待人工标注
404. 2025
     基于eRDMA实测DeepSeek开源的3FS2025-03-01 · 完整系统章节
405. 2025
     [一个脑洞]MoE专家权重是否可以预测并Prefetch?2025-02-28 · 待人工标注
406. 2025
     分析一下EP并行和DeepSeek开源的DeepEP代码2025-02-27 · 独立系统解释
407. 2025
     谈谈Kimi的LLM相关的工作, K1.5/MOBA/MUON2025-02-23 · 待人工标注
408. 2025
     基于强化学习的模型商品化推荐系统及MaaS盈利模式分析2025-02-22 · 待人工标注
409. 2025
     从Deepseek R1和NSA算法谈谈个人的一些反思2025-02-19 · 待人工标注
410. 2025
     谈谈DeepSeek Native Sparse Attention2025-02-18 · 待人工标注
411. 2025
     谈谈微信+DeepSeek2025-02-16 · 待人工标注
412. 2025
     谈谈DeepSeek MoE模型优化和未来演进以及字节Ultra-Sparse Memory相关的工作2025-02-15 · 待人工标注
413. 2025
     详细谈谈DeepSeek MoE相关的技术发展2025-02-13 · 待人工标注
414. 2025
     谈谈DeepSeek-R1满血版推理部署和优化2025-02-13 · 待人工标注
415. 2025
     谈谈国产算力支持大模型和MoE/RL算法协同演进方向2025-02-13 · 待人工标注
416. 2025
     详细谈谈DeepSeek MoE相关的技术发展2025-02-10 · 待人工标注
417. 2025
     谈谈国产算力支持大模型和MoE/RL算法协同演进方向2025-02-06 · 待人工标注
418. 2025
     AI的中国资本叙事2025-01-31 · 待人工标注
419. 2025
     [去年旧文重发]谈谈AISys架构师的基本素养2025-01-28 · 待人工标注
420. 2025
     关于Blackwell架构的一些更正2025-01-27 · 待人工标注
421. 2025
     谈谈三万亿的破绽(4)2025-01-27 · 待人工标注
422. 2025
     Blackwell TensorCore架构2025-01-26 · 待人工标注
423. 2025
     一些关于Blackwell微架构的分析2025-01-25 · 待人工标注
424. 2025
     谈谈大模型架构的演进之路, The Art of memory.2025-01-23 · 待人工标注
425. 2025
     一个关于MoE的猜想2025-01-14 · 待人工标注
426. 2025
     Pretrain ScalingLaw真的终结了么?2025-01-05 · 待人工标注
427. 2025
     新年: 但愿那月落重生灯再红2025-01-01 · 待人工标注
428. 2025
     谈谈一个新的MoA模型架构DeepSick-4.1T2025-04-01 · 待人工标注

当前分类里没有匹配条目。

Systems for ML 的题名、分类和外链来自 Lingyun Yang 维护的公开目录快照；公众号部分只保存标题、日期、原始链接和本站导航标注，不转载正文。核验规则和收录范围见证据与来源。