main

GGitHubFix torchcsprng build with conda and PyTorch CUDA wheel layouts (#12)

ef2c3e06创建于 23 天前历史提交

文件	最后提交记录	最后更新时间
application	refactor(NssMPClib)!: v1.0.0 major architectural overhaul BREAKING CHANGES: - Package renamed from NssMPC to nssmpc - High-level APIs completely redesigned with standardized exports - All communication backends except tensorpipe_rpc temporarily removed - Protocol layer function naming conventions normalized Major Improvements: • Redesigned high-level interfaces with cleaner export structure • Normalized protocol layer function naming • Updated neural network inference toolkit components • Reorganized utils module with new context manager-based performance profiler • Added offline generation for neural network matrix beavers with online lazy loading • Unified all auxiliary parameter providers into single interface • Synchronized all example code with new API patterns Migration Guide: 1. Update imports: from nssmpc import * instead of from NssMPC import * 2. Review protocol function calls - naming now follows snake_case 3. Communication backend configuration now defaults to tensorpipe_rpc only 4. Check utility imports as module structure has been reorganized 5. Regenerate auxiliary parameters using new offline generation toolchain Note: This release focuses on API stability and performance optimization. Future releases will reintroduce additional communication backends.	5 个月前
config	Fix torchcsprng build with conda and PyTorch CUDA wheel layouts (#12) * change the method of install * add install_advice.py * simplify the install process * fix readme * fix again * complate install advice * change the api of restore() in tutorials	23 天前
infra	feat: add CUTLASS support for high-performance matrix multiplication and update setup for CUDA extensions (#10) Key changes: Add CUTLASS as a git submodule (.gitmodules) to ensure precise version locking and avoid repo bloat. Implement highly optimized custom CUDA kernels in nss_cutlass_kernels.cu. Refactor ring_tensor.py and arithmetic.py to dispatch computations to the new CUTLASS-backed backend. Update setup.py with robust CUDA architecture auto-detection, submodule validation checks, and preserve essential package metadata. Add MANIFEST.in to guarantee C++/CUDA headers are properly included in source distributions (sdist). Update README.md with --recursive cloning instructions for new developers.	3 个月前
primitives	feat: add CUTLASS support for high-performance matrix multiplication and update setup for CUDA extensions (#10) Key changes: Add CUTLASS as a git submodule (.gitmodules) to ensure precise version locking and avoid repo bloat. Implement highly optimized custom CUDA kernels in nss_cutlass_kernels.cu. Refactor ring_tensor.py and arithmetic.py to dispatch computations to the new CUTLASS-backed backend. Update setup.py with robust CUDA architecture auto-detection, submodule validation checks, and preserve essential package metadata. Add MANIFEST.in to guarantee C++/CUDA headers are properly included in source distributions (sdist). Update README.md with --recursive cloning instructions for new developers.	3 个月前
protocols	chore: clean up imports in multiplication.py Removed unused imports from multiplication.py.	4 个月前
runtime	refactor(NssMPClib)!: v1.0.0 major architectural overhaul BREAKING CHANGES: - Package renamed from NssMPC to nssmpc - High-level APIs completely redesigned with standardized exports - All communication backends except tensorpipe_rpc temporarily removed - Protocol layer function naming conventions normalized Major Improvements: • Redesigned high-level interfaces with cleaner export structure • Normalized protocol layer function naming • Updated neural network inference toolkit components • Reorganized utils module with new context manager-based performance profiler • Added offline generation for neural network matrix beavers with online lazy loading • Unified all auxiliary parameter providers into single interface • Synchronized all example code with new API patterns Migration Guide: 1. Update imports: from nssmpc import * instead of from NssMPC import * 2. Review protocol function calls - naming now follows snake_case 3. Communication backend configuration now defaults to tensorpipe_rpc only 4. Check utility imports as module structure has been reorganized 5. Regenerate auxiliary parameters using new offline generation toolchain Note: This release focuses on API stability and performance optimization. Future releases will reintroduce additional communication backends.	5 个月前
__init__.py	refactor(NssMPClib)!: v1.0.0 major architectural overhaul BREAKING CHANGES: - Package renamed from NssMPC to nssmpc - High-level APIs completely redesigned with standardized exports - All communication backends except tensorpipe_rpc temporarily removed - Protocol layer function naming conventions normalized Major Improvements: • Redesigned high-level interfaces with cleaner export structure • Normalized protocol layer function naming • Updated neural network inference toolkit components • Reorganized utils module with new context manager-based performance profiler • Added offline generation for neural network matrix beavers with online lazy loading • Unified all auxiliary parameter providers into single interface • Synchronized all example code with new API patterns Migration Guide: 1. Update imports: from nssmpc import * instead of from NssMPC import * 2. Review protocol function calls - naming now follows snake_case 3. Communication backend configuration now defaults to tensorpipe_rpc only 4. Check utility imports as module structure has been reorganized 5. Regenerate auxiliary parameters using new offline generation toolchain Note: This release focuses on API stability and performance optimization. Future releases will reintroduce additional communication backends.	5 个月前