-
### 🐛 Describe the bug
`get_symm_mem_workspace` throws an OOM error when it is passed to 2GB or a larger value as the size of the workspace. Here is the repro:
```
import torch
import torch.dist…
-
### 🐛 Describe the bug
I can't narrow it down further, but torch.arctan2 seemingly calculates the correct gradients and optimises correctly for fp32,fp64 and bfloat16, but for some reason, the versio…
-
### 🐛 Describe the bug
Consider the following piece of code:
```python
# pylint: disable=missing-docstring
from torch import autograd
class Exp(autograd.Function):
@staticmethod
…
RuRo updated
6 months ago
-
### 🐛 Describe the bug
Executing a pytorch-lightning models results in the following error:
```
" File "/local_data/user1/miniforge3/envs/lightning_py310/lib/python3.10/site-packages/lightning/…
-
### 🐛 Describe the bug
Calling `.generate` on a HuggingFace model that has been FSDP wrapped results in an error. I was able to work around this error by summoning full params without recurse, which …
-
```bash
Assertion failed: mmu::phys_bits
-
| | |
|--------------------|----|
| Bugzilla Link | [PR35785](https://bugs.llvm.org/show_bug.cgi?id=35785) |
| Status | NEW |
| Importance | P normal |
|…
-
I've been trying to setup Sacrab on a Windows Linux Subsystem (Ubuntu). I've installed g++, gcc, Clang, and Python 3 using
`sudo apt install clang g++ gcc python3`.
The versions that apt install…
-
**I confirm this bug has not already been reported**
- [ x] I have searched the issues and this bug has not been reported previously
**Describe the bug**
When starting Ubuntu 22.04 on an Intel Ma…
-
### 🐛 Describe the bug
I am trying to wrap a sequence of Llama decoder layers (as implemented by huggingface) with a cuda graph to speed up training. Without cuda graphs, the loss decreases normally.…