-
## Work Items
* Meta-device initialization / `_apply()` methods
- [x] Support initial meta-device initialization using `swap_tensors` path
- [ ] Remove manual padding logic after https://github…
awgu updated
2 months ago
-
### 🚀 The feature, motivation and pitch
## Motivation
By now, PyTorch has defined > 2000 operations. Meanwhile, PyTorch users start from eager mode. If a new backend intends to support PyTorch e…
-
```python
import itertools
import torch
def filt(x):
return x < 10
@torch.compile(backend="eager", fullgraph=True)
def f(x):
x = x + 1
return zip(range(3), filter(filt, itertoo…
-
I am trying pytorch tutorial for `torch.compile()`: https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html#demonstrating-speedups adopting it for `xpu` backend by `s/cuda/xpu`. Using h…
-
[Scala Dependency Injection / Eager bindings](https://www.playframework.com/documentation/2.8.x/ScalaDependencyInjection#Eager-bindings):
> If you want to create them once and also create them eage…
-
### Description
Follows up on #41929.
https://github.com/quarkusio/quarkus/pull/41810 introduces a way for extensions to declare beans as "inactive", so that:
1. Any attempt to access those b…
-
This is a very good piece of work! As a beginner in this field, I am eager to reproduce your code for learning purposes. Could you please provide the original data so that we can effectively reproduce…
-
## Expected Behavior
Eager workflow start task should have the same search attributes as polling workflow task when starting a workflow with a custom search attribute.
## Actual Behavior
Eage…
-
I took a benchmark of `torch.compile`'d `Adam.step` with Thunder backend. Surprisingly, the compiled `Adam.step` was even slower than the eager mode.
||compilation (s)|execution (ms)|
|:-|-:|-:|
…
-
Before Laravel 11 to use compoships with [Eloquent Eager Limit](https://github.com/staudenmeir/eloquent-eager-limit) it was necessary to use this [Compoships Eager Limit](https://github.com/mpyw/compo…