Yukio Siraichi

At the current state, `Enfield` is annotated with `doxygen` style annotations. I haven't tested it, so it would be great to have a documentation working and buildable.

Keep track of `ViewMeta` with symbolic inputs.

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #125876 Fix: #125387 This PR helps keep track of whether an instantiated `ViewMeta` has symbolic values as input or not. This is...

Make `_embedding_bag_backward` explicitly dispatch to CPU and CUDA.

3

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #129691 This PR modifies `_embedding_bag_backward` item inside _native_functions.yaml_, so that it dispatches to CPU and CUDA directly, instead of `CompositeImplicitAutograd`. *Context:* PyTorch...

Introduce CUDA OpenXLA fallback.

13

This PR introduces OpenXLA fallback on PyTorch GPU eager. Instead of running fallback operations (i.e. whenever a operation has no lowering implemented) on CPU, we now make it possible to...

Fix: #6008 This PR, in addition to https://github.com/pytorch/pytorch/pull/129691, add support for embedding bag calls when its `sparse` parameter is true. **Problem:** when `sparse=true`, `_embedding_bag_backward` called `at::_sparse_coo_tensor_unsafe_symint` function which returns a...

`F.embedding_bag(..., mode='max')` yields different results than PyTorch eager.

2

## 🐛 Bug Running the script bellow yields in an `AssertionError`. This issue occurs only when `requires_grad=False`, which triggers the execution of `_embedding_bag_forward_only`. ```python EMB = 10 DIM = 5...

`DALLE2_pytorch` moved to `canary_models` due to NumPy 2.0 upgrade.

1

In #2311, `DALLE2_pytorch` benchmark was moved into `canary_models` due to the NumPy 2.0 upgrade. I'm opening this issue for tracking the progress towards moving it back to the `models` folder.

### 🐛 Describe the bug List of operations, whose `out=` functions on meta inputs are not consistent when ran with real tensors (e.g. CPU or CUDA). Specifically, I changed the...

`aot_eager` does not error on mismatching data-types in out tensors.

8

### 🐛 Describe the bug There is an inconsistency when running `torch.aminmax` out variant on eager and inductor. This behavior is also found when using `aot_eager` backend. ```python def run(op):...

List of operations, whose `out=` variants are not consistent with eager (i.e. run on CPU/CUDA, but fail when using meta devices). I have grouped them according to the error each...