tiga-lang Alpha
A differentiable JIT compiler for graph message passing, with relation-aware IR, fused message aggregation, and CPU / NVIDIA GPU execution.面向图消息传递的可微 JIT 编译器,保留图关系语义,实现消息计算与聚合融合,并支持 CPU / NVIDIA GPU 执行。
Code, methods & papers代码、方法与论文
Large language models, differentiable physics simulation, scientific computing, and compilers. Project figures link to their full-size originals; papers link to the published source.大语言模型、可微物理仿真、科学计算与编译器。点击项目配图可查看完整原图,论文链接指向原始发表页面。
A differentiable JIT compiler for graph message passing, with relation-aware IR, fused message aggregation, and CPU / NVIDIA GPU execution.面向图消息传递的可微 JIT 编译器,保留图关系语义,实现消息计算与聚合融合,并支持 CPU / NVIDIA GPU 执行。
Sparse matrix structure ↗稀疏矩阵结构 ↗Differentiable sparse linear algebra, adjoint solvers, and multi-GPU sparse solving in PyTorch.基于 PyTorch 的可微稀疏线性代数、伴随求解器与多 GPU 稀疏求解。
3D elasticity simulation ↗三维弹性仿真 ↗PyTorch-native finite elements for differentiable simulation, inverse problems, and optimization.PyTorch 原生有限元工具,用于可微仿真、逆问题与优化。
Assembly overview · Wen, Chi et al. ↗装配方法概览 · Wen, Chi 等 ↗Efficient Galerkin assembly for learning, solving, and optimizing partial differential equations.高效的 Galerkin 装配方法,用于偏微分方程的学习、求解与优化。
arXiv:2406.05085
ICCV Workshops · pp. 1358–1363
ACM/IMS Transactions on Data Science · DOI: 10.1145/3488059
Mingyuan Chi · Technical report / manuscript · Compiler design, automatic differentiation, memory hierarchy and evaluation.迟明远 · 技术报告 / 手稿 · 编译器设计、自动微分、内存层次与实验评估。
Semantic Chunk Sparse Attention for Large Language Models ↗
Mingyuan Chi · ETH Zürich · 2024迟明远 · 苏黎世联邦理工学院 · 2024