arXiv · ML· Shengjun Kris Dong, Dima Nikiforov, Sahil Modi, Sana Damani, Edward C Lin, Siva Kumar Sastry Hari, Christos Kozyrakis·· 17 小时前AI 评分28
ModelBlaster 通过跨模型、内核和硬件后端的增强学习 Agent 缩小性能差距
ModelBlaster: Closing the Performance Gap through Learning-Augmented Agents Across Models, Kernels and Hardware Backends
AI 导读
ModelBlaster 提出 Memory-Augmented In-Context Reinforcement Learning (MAIC-RL) 框架，利用持久知识库跨 CUDA、OpenCL 及 RISC-V 硬件复用优化经验。
来源：arXiv · ML · arxiv.org