Added correct handling for tensor allocation for large tensors when using torch.resize on CUDA (#52672). Fixed an illegal memory access that could happen when computing the inverse of a batch of matrices on CUDA (#53064). Fixed a bug where torch.sparse.addmm would compute the wrong results...
Introducing dim_order: A New API for Understanding Dense Tensor Memory Layout #146086 opened Jan 30, 2025 DISABLED test_device_type (__main__.TestScript) #146083 opened Jan 30, 2025 DISABLED test_module_parameters_and_buffers (__main__.TestScript) #146082 opened Jan 30, 2025 ...
Compute Capability 5.2 malloc(): invalid size (unsorted) Fatal Python error: Aborted Current thread 0x00007fd58c704740 (most recent call first): File "/home/stas/anaconda3/envs/main-38/lib/python3.8/site-packages/tensorflow/python/ops/gen_...