-
### 🐛 Describe the bug
When using dcp.save or dcp.async_save as torch.save alternative and then loading state_dicts back, non-tensor values is not updated as expected, for example(edited from [async_…
-
### 🐛 Describe the bug
We recently try to generate high resolution contents using diffusion models but notice unusual memory usage when resolution becoms large
Shown in above, the memory usa…
-
### 🐛 Describe the bug
As in https://github.com/pytorch/pytorch/pull/112140, we plan to remove the Conv/GEMM's output annotation in `X86InductorQuantizer`, it works for PTQ Quantization, but failed…
-
### 🐛 Describe the bug
When I use the CUDA graphs API to capture execution of a model that performs NCCL collectives as part of that execution, it puts the NCCL process group into a state where it ca…
-
### 🐛 Describe the bug
I have noticed that `torch.onnx.export` produces wrong models if a module contains `Tensor.index_add_` function and the `index` parameter contains duplicate values. See the m…
-
### 🐛 Describe the bug
**Summary**
When attempting to compile a math sdp operation using torch.compile in AMP mode, the script encounters a crash. This issue does not occur with a single Flash Atten…
-
### Your current environment
```text
Collecting environment information...
PyTorch version: 2.3.0+cu121
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A
…
-
I only copied the code from the ReadMe, I installed the LLama NuGet package with the CPU-Only backend, and it always returns
System.AccessViolationException: "Attempted to read or write protected …
-
I have noticed that Alpaca uses my CPU instead of my GPU. Here's a screenshot showing how it's using almost 40% of my CPU, and only 1% of my GPU.
![Captura desde 2024-07-10 06-51-39](https://github…
-
### 🐛 Describe the bug
Debian 13
python 3.10.12 venv
ROCm 6.1 / HIP
Using PyTorch2.4.1_rocm & compile xformers from source
When using txt2img and having xformers utilized in the script the runn…