peterbell10

Declare public dependencies on libshm

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #76104 * #82695 * __->__ #82694 * #82572

Use public/cuda.cmake to declare cusparse and cusolver

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #76104 * #82695 * #82694 * __->__ #82572 This uses the `caffe2::{lib}` pattern to link in ATen's cuda dependencies, and adds targets for...

WIP: Use CUDAToolkit to find cupti

3

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #76104 * #82695 * #82694 * #82572

WIP: Use FindCUDAToolkit to find cuda dependencies

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #76104 * __->__ #82695 * #82694 * #82572

Add an NBT tree structure

12

This is the first step in fixing #3682 and adding support for storing arbitrary NBT in `cItem`.

GEMM much slower than GEMV for multiplying column or row vectors

2

### What is the expected behavior GEMM where one of m or n is 1 should perform similarly to the equivalent GEMV call, because it could simply call the GEMV...

OpInfo: Add test that sample_inputs_func returns a generator

2

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #84631 * #84455 * __->__ #84567 * #84454 * #84554

test_decomp.py: Use freeze_rng_state for sample reproducibility

2

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #84631 * #84455 * #84567 * #84454 * __->__ #84554

OpInfo: Extend generator testing to reference and error input funcs

2

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #84631 * #84455 * #84567 * #84454 * #84554 This tests that `error_inputs_func` and `reference_inputs_func` both return generators and also tightens the...

OpInfo: Use yield consistently instead of appending to lists

3

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #84631 * #84455 * #84567 * __->__ #84454 * #84554 The yield syntax reduces the numbers of tensors held in-memory when running opinfo...