-
### Suggestion Description
Dear ROCm developers,
according to some tests I performed, Managed Memory was not really working in ROCm 5.x but it does work at least in ROCm 6.1.2. Is the XLA implemen…
-
## ❓ Questions and Help
pip install torch_xla-2.2.0-cp310-cp310-manylinux_2_28_x86_64.whl
But got the error:
ERROR: torch_xla-2.2.0-cp310-cp310-manylinux_2_28_x86_64.whl is not a supported wheel on…
-
Originally reported as google/jax#20184 and google/jax#16008:
> ### Description
> When inspecting the estimated flop count of a compiled function, dot_general, einsum, '@', jnp.dot, etc show "-1.0…
-
## 🐛 Bug
Running into the following error when using `torch.compile(backend="openxla")`:
```
File "torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
return self._call_impl(*a…
-
## 🐛 Bug
Running the upstreamed benchmarking scripts with the following command results in an unexpected error. It does work when using CPU OpenXLA fallback, though.
```bash
python xla/benchmar…
-
Hi, I have following setup:
- Transformer model with N layers scanned over input
- fully sharded data parallel sharding
- asynchronous communications (latency-hiding scheduler, pipelined all-gather…
-
## 🐛 Bug Report
When using [dynamo sharding](https://github.com/pytorch/xla/blob/88bcb45fda546e5c1fb4f12de75251bfa5fd332e/torch_xla/core/custom_kernel.py#L17) inside `torch.compile`, I encounter th…
-
Hi,
The XLA:GPU profiler has a segfault bug when CUPTI initialization failed:
```
Thread 1 "python" received signal SIGSEGV, Segmentation fault.
0x00007fff0401cc7e in nsync::nsync_mu_lock(nsyn…
-
In the following MWE
```cpp
xla::XlaBuilder root("root");
auto zero = xla::ConstantR0(&root, 0);
absl::Span xs = {0};
auto zeros = xla::ConstantR1(&root, xs);
…
-
I'm trying to build xla from source for CPU following the instructions [here](https://github.com/openxla/xla/blob/main/docs/developer_guide.md) and it's failing with:
```
xla/service/gpu/runtime/…