Report
TorchLean expands its shape-checked tensor API with einsum and verified lowering
TorchLean developer Robert Joseph George says supported Lean functions can now compile to CUDA, while LibTorch handles supported model operations.
TLDR
In an October 15 update, TorchLean developer Robert Joseph George says the project is becoming a broader tensor programming and execution framework inside Lean. Its expanded API includes shape-safe operations, einsum contractions, and tools to rearrange, expand and reduce tensor dimensions. He says einsum’s verified lowering preserves the specified arithmetic order. His linked write-up distinguishes those proofs from verification of the external CUDA compiler or GPU.
Combined views
52
1 Source, first seen ago
