-
### 🐛 Describe the bug
If I torch.compile `torch.amp.GradScaler`, it works. But if I copy paste grad_scaler.py and import GradScaler from there, I receive an error.
To reproduce (testcase taken …
-
For some reason, `torch.argsort` is crashing when the tensor is on a Neuron device. For example, the code snippet bellow works as expected (on CPU):
```
>>> import torch
>>> import torch_neuronx
>…
-
torch-npu 2.2.0
`import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "npu"
# THUDM/glm49bchat
tokenizer = AutoTokenizer.from_pretrained("/home…
-
Naively, observable sharing means that we either need to put sharing in the hands of users (by giving them the choice to share or not), or give them a verbose API that doesn't really resemble maths. W…
-
Hello! I have recently been trying to train my mesh to ground truth by using the normals of some images in the COLMAP dataset, specifically, the camera information that provides each view in the datas…
-
have a try to reproduce Nvidia's results on using slurm + enroot + pyxis
1. downgrade the transformers and huggingface_hub libs (huggingface_hub==0.23.2 transformers==4.40.2) because the versions…
-
-
### RuntimeError: The expanded size of the tensor (628) must match the existing size (129) at non-singleton dimension 3. Target sizes: [1, 32, 1, 628]. Tensor sizes: [1, 1, 1, 129]
Issue: Upon ru…
-
Hi, we have the compiled SDXL model consistantly running inference with different dimensions and from time to time some of the jobs will have this following erros that the tensor dimension doesn't mat…
-
Pytorch 2.3 is introducing unsigned integer dtypes like `uint16`, `uint32` and `uint64` in https://github.com/pytorch/pytorch/pull/116594.
Quoting Ed:
> The dtypes are very useless right now (not…