trunk/1c3d446cda0cb4231a810427809d2b3382395e0b: [CUDA] Explain custom MemPool allocator failures in OOM errors (#199363)
- PyTorch: 1780 events in the last 90 days
- PyTorch: 1758th Release in the last 90 days
- Previous: earlier the same day · trunk/dcd7ed975a6b090ec2bcf2360c28c9a263be8fe5: [Profiler] Update Kineto readme (#199364)
What happened
Human note: make a better error message when custom mempool fails allocation Custom allocators signal allocation failure by returning nullptr. The caching allocator maps this to an out-of-memory error, so failures such as invalid allocator configuration can appear to be device-wide memory exhaustion even when the custom allocator logged the underlying cause. Track whether the custom allocator returned nullptr for th…
Summary assembled by rule from the sources below