-
### 🐛 Describe the bug
Bug as shown in the title.
Solution: export CUDA_PATH=\~/anaconda3/envs/\:\~/anaconda3/envs/\/targets/x86_64-linux
### Versions
Collecting environment information...
PyTor…
-
### 🐛 Describe the bug
```torch.onnx.errors.SymbolicValueError: ONNX symbolic expected the output of `%2212 : Tensor = onnx::Squeeze(%2186, %2211), scope: SimpleLSTMNet::/torch.ao.nn.quantized.modu…
-
co-authored with @iffsid
Statistically, batching allows us to trade off the speed of taking gradient steps for lower variance gradient estimators. This can be done in a for loop. However, batching…
-
### 🐛 Describe the bug
distributed scatter doesn't free memory on the source rank. The allocated and reserved memory values are correct but leaves less free memory on the source rank device.
```py…
-
### 🐛 Describe the bug
When profiling using `with_stacks=True`, the chrome trace export can be corrupted due to corrupted function names. However, there seems to be a discrepancy between Python 3.1…
-
### Your current environment information
```
libibverbs not available, ibv_fork_init skipped
Collecting environment information...
PyTorch version: 2.1.2+cu121
Is debug build: False
CUDA used …
-
### Describe the bug
I found in jupyter notebook, `to('xpu')` makes the Jupyter kernel die.
### Notebook to reproduce
![image](https://github.com/intel/intel-extension-for-pytorch/assets/105281…
-
### 🐛 Describe the bug
In the following code, I am encountering a `HIP error` during backpropagation that is thrown when executing `bond_losses[atompair_mask].mean()`. I initially suspected (based …
-
### 🐛 Describe the bug
Dtensor shard uses more gpu memory than raw tensor.
With test, Shard gpu mem: 21890MiB > Replicate gpu mem: 17448MiB > Raw tensor gpu mem: 16804MiB.
Confused for a long time…
v4if updated
2 months ago
-
### 🐛 Describe the bug
`to(torch.int8)` will get different result on XPU.
```python
import torch
torch.tensor([[ 57.7637, 215.2612, 212.4291],[193.8332, 227.0923, 158.8016]], device='cpu').t…