摘要:
Abstract good words: subjectivity, variability, scale Task: Survey of LLM-as-a-Judge, benchmark & evaluation of LLM-as-a-Judge systems Core question: 阅读全文
摘要:
Abstract Task: Defense LLM from prompt injection attacks Tool: TaskTracker Methods: use activation deltas( the difference in activations before and af 阅读全文
摘要:
Abstract Github: https://github.com/JailbreakBench/jailbreakbench https://jailbreakbench.github.io/ Task: Opensource benchmark an evolving repository 阅读全文
摘要:
3. TensorGuard: A RAG-Based Multi-agent framework to detect and fix DL Checker Bugs RAG Design relevant contextual information from a large corpus of 阅读全文
摘要:
Abstract Tool: Tensorflow Eager Task: domain specific language for hardware-accelerated machine learning 特点:multi-stage, Python embedded 背景:为何要extend 阅读全文
摘要:
Abstract 工具:TorchDynamo, TorchInductor Task: 实现了torch.compile的解释器和编译器 TorchDynamo: Task: Python-level just-in-time(JIT) compiler Method: 允许graph compi 阅读全文
摘要:
Abstract 特点:能在大量异构系统上运行 1. Intro 前身: DistBelief tensorflow: 效果: 从外围设备到几千个卡的集群,可以简单完成不同规模的模型训练与推理(multi-scale training) 节约maintainance burden,避免leaky a 阅读全文
摘要:
https://github.com/mukobi/Minitorch-Self-Study-Guide-SAIA/blob/main/README.md#what-is-this https://minitorch.github.io/ srush/GPU-Puzzles: Solve puzzl 阅读全文
摘要:
Abstract 去中⼼化⾝份 去中⼼化访问控制 符合政策的去中⼼化计算((i) decentralized identity, (ii) decentralized access control (iii) policy-compliant decentralized computation) 1 阅读全文
摘要:
Paper https://arxiv.org/abs/2309.06615 Abstract 背景: 自动机A被称作查分隐私自动机:当对某些D,对任何隐私预算ε>0,该自动机是Dε-differentially private( A DiP automaton is a parametric au 阅读全文