ml-systems-papers
github.com/byungsoo-oh/ml-systems-papers ↗Curated collection of papers in machine learning systems
657
GitHub Stars
800
Curated Resources
24
Categories
22 hours ago
Last Refreshed
Data ProcessingTraining SystemInference SystemAttention OptimizationMixture of Experts (MoE)Communication Optimization & Network Infrastructure for Distributed MLFault tolerance & Straggler mitigationGPU Memory Management & OptimizationGPU SharingCompilerGPU Kernel OptimizationLLM Long ContextModel CompressionFederated LearningPrivacy-Preserving MLML APIs & Application-Side OptimizationML for SystemsEnergy EfficiencyRetrieval-Augmented Generation (RAG)SimulationSystems for Agentic AIMultimodalHybrid LLMsOthers
Use this list with your AI agent
Add the Context Awesome MCP server to Claude, Cursor, or any MCP client, then ask:
"Show me distributed training resources from ml-systems-papers"
Installation instructions →What's inside
Model Compression
- 70% Size, 100% Accuracy: Lossless LLM Compression for Efficient GPU Inference via Dynamic-Length Float (DFloat11)
- AdaEmbed: Adaptive Embedding for Large-Scale Recommendation Models
- Any-Precision LLM: Low-Cost Deployment of Multiple, Different-Sized LLMs
- Approaching Shannon Bound with Lossless LLM Weight Compression
- BitNet: 1-bit Pre-training for Large Language Models
- DecDEC: A Systems Approach to Advancing Low-Bit LLM Quantization
Simulation
Training System
- Accelerating Distributed DLRM Training with Optimized TT Decomposition and Micro-BatchingDistributed training
- Accelerating the Training of Large Language Models using Efficient Activation Rematerialization and Optimal Hybrid ParallelismDistributed training
- Aceso: Efficient Parallel DNN Training through Iterative Bottleneck AlleviationDistributed training
- AdaGen: Workload-Adaptive Cluster Scheduler for Latency-Optimal LLM Inference ServingResource scheduling
- AdaPipe: Optimizing Pipeline Parallelism with Adaptive Recomputation and PartitioningDistributed training
- Adaptive Message Quantization and Parallelization for Distributed Full-graph GNN TrainingGNN training
Mixture of Experts (MoE)
- Accelerating Distributed MoE Training and Inference with Lina
- Accelerating MoE Model Inference with Expert Sharding
- Achieving Cloud-Grade SLOs for Local Mixture-of-Experts Inference through CPU-GPU Hybrid Design
- Ada-K Routing: Boosting the Efficiency of MoE-based LLMs
- AdaMOE: Token-Adaptive Routing with Null Experts for Mixture-of-Experts Language Models
- Adaptive Gating in Mixture-of-Experts based Language Models
Inference System
- Accelerating LLM Serving for Multi-turn Dialogues with Efficient Resource Management
- Accelerating Sparse Transformer Inference on GPU
- Aegaeon: Effective GPU Pooling for Concurrent LLM Serving on the Market
- A House United Within Itself: SLO-Awareness for On-Premises Containerized ML Inference Clusters via Faro
- AlignedServe: Orchestrating Prefix-aware Batching to Build a High-throughput and Computing-efficient LLM Serving System
- An LPDDR-based CXL-PNM Platform for TCO-efficient Inference of Transformer-based Large Language Models
Communication Optimization & Network Infrastructure for Distributed ML
- Accelerating Model Training in Multi-cluster Environments with Consumer-grade GPUs
- Accelerating MoE with Dynamic In-Switch Computing on Multi-GPUs
- An in-network architecture for accelerating shared-memory multiprocessor collectives
- ARK: GPU-driven Code Execution for Distributed Deep Learning
- Astral: A Datacenter Infrastructure for Large Language Model Training at Scale
- A Unified Architecture for Accelerating Distributed DNN Training in Heterogeneous GPU/CPU Clusters
Fault tolerance & Straggler mitigation
- AdaCheck: An Adaptive Checkpointing System for Efficient LLM Training with Redundancy Utilization
- An Empirical Study on Program Failures of Deep Learning Jobs
- Attack of the Bubbles: Straggler-Resilient Pipeline Parallelism for Large Model Training
- Bamboo: Making Preemptible Instances Resilient for Affordable Training of Large DNNs
- BCP: A Unified Checkpointing System for Large Foundation Model Development
- Capricorn: Efficient In-Memory Checkpointing for MoE Model Training with Dynamicity Awareness
Data Processing
- A Deep Learning Dataloader with Shared Data Preparation
- A Selective Preprocessing Offloading Framework for Reducing Data Traffic in DL Training
- Bagpipe: Accelerating Deep Recommendation Model Training
- cedar: Composable and Optimized Machine Learning Input Data Pipelines
- Demystifying Synthetic Data in LLM Pre-training: A Systematic Study of Scaling Laws, Benefits, and Pitfalls
- DIESEL: A Dataset-Based Distributed Storage and Caching System for Large-Scale Deep Learning Training
Showing a sample of 800 resources. View the full list on GitHub →