-
Thanks for sharing the code.
I tried to train kl-reg autoencoder with custom datasets.
Reconstruction is possible with further learning, but image generation from noise is not possible no matter h…
-
### 🐛 Describe the bug
Re-triage it by https://github.com/intel/torch-xpu-ops/commit/cbb4ab17a781c77108443f12f7ce254a345f1a14. Old issue is https://github.com/intel/torch-xpu-ops/issues/325
# Runt…
-
### 🐛 Describe the bug
I'm experimenting in exporting various MSS models to Executorch. Following the example export scenario in Executorch Documentation, the export terminates in error: 'Operator …
-
We use this Library in JDK 8 on one dedicated server whose CPU is Intel i5-13500 and this CPU uses big.LITTLE architecture. It has 6 Performance cores(which implements HyperThread Technology and can b…
-
### 🐛 Describe the bug
Using `torch.distributed.barrier()` doesn't work with NCCL. I use the code
```python
import torch
def train() -> None:
torch.distributed.init_process_group('nccl'…
-
### Your current environment
The output of `python collect_env.py`
```text
PyTorch version: 2.6.0.dev20240918+rocm6.2
Is debug build: False
CUDA used to build PyTorch: N/A
ROCM used to bui…
-
### 🐛 Describe the bug
Hi @atalman ,
There is an ImportError in inductor with Triton 2.3.x. `triton_key` is missing in python/triton/compiler/compiler.py, which is added in https://github.com/t…
-
### 🐛 Describe the bug
512M parameters
Mostly vanilla LM transformer. FlashAttention 2.4.2, PyTorch 2.2.0. Uses both FA and FlashRotary.
Dtype: bf16
Nvidia A40. single-GPU
Unfused: 85 TFLOPS
F…
ad8e updated
5 months ago
-
### 🐛 Describe the bug
This is only a minor issue, but Tensors created with `torch.normal` or `torch.from_numpy` creates a tensor on CPU and not on the `default_device`. At least if only float inpu…
-
### 🐛 Describe the bug
```
import torch
import time
import torch.nn as nn
import torch.utils.data as Data
torch.manual_seed(10)
class ATT(nn.Module):
def __init__(self, ):
super…