UMLoad: Fast On-Demand Model Loading on Unified Memory Architecture
cs.OS, cs.AR
Submitted: 2026-11-03
Updated: 2026-11-04
Code: https://github.com/ggml-org/llama.cpp
Terminology
Sources
- Small Language Models are the Future of Agentic AI
- GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers
- Scaling Up On-Device LLMs via Active-Weight Swapping Between DRAM and Flash
- A Comprehensive Study on Quantization Techniques for Large Language Models
- CSGO: Generalized Optimization for Cold Start in Wireless Collaborative Edge LLM Systems
- Exascale Implicit Kinetic Plasma Simulations on El~Capitan for Solving the Micro-Macro Coupling in Magnetospheric Physics
- Minions: Cost-efficient Collaboration Between On-device and Cloud Language Models
- RouteLLM: Learning to Route LLMs with Preference Data
- LLMartini: Seamless and Interactive Leveraging of Multiple LLMs through Comparison and Composition
- MLLM-Tool: A Multimodal Large Language Model For Tool Agent Learning
- Mixture-of-Agents Enhances Large Language Model Capabilities
- Ghidorah: Fast LLM Inference on Edge with Speculative Decoding and Hetero-Core Parallelism
- Agent.xpu: Efficient Scheduling of Agentic LLM Workloads on Heterogeneous SoC
- Understanding Large Language Models in Your Pockets: Performance Study on COTS Mobile Devices
- EdgeFlow: Fast Cold Starts for LLMs on Mobile Devices
- LLM as a System Service on Mobile Devices
- mzCache: On-Device LLM Memory Management under Multitasking
- LMSYS-Chat-1M: A Large-Scale Real-World LLM Conversation Dataset
- Tangram: Accelerating Serverless LLM Loading through GPU Memory Reuse and Affinity