pytorch
pytorch copied to clipboard
pytorch
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Today in PyTorch there are a variety of logging and warning types, facilities and feature requests. A partial list: - determinism warnings, controlled with [set_deterministic_debug_mode](https://pytorch.org/docs/master/generated/torch.set_deterministic_debug_mode.html#torch.set_deterministic_debug_mode) - deprecation warnings, which rely...
`derivatives.yaml` can now take a `dispatch` entry which registers per-autograd dispatch key derivatives such as ``` name: foo(Tensor self, Tensor y) -> Tensor dispatch: default: x: grad y: grad.expand(y.sizes()) AutogradNestedTensor:...
### Description When someone cancels a PR merge, it's not apparent that it's canceled unless the user clicks into that job. In this PR, we add a message if the...
### 🐛 Describe the bug [File](https://gist.github.com/DavidSlayback/dd327d17d41e67267f0026efb0686fc5) In an RL workflow, when I use CUDA, JIT, and a LayerNorm together in this script, I get the following error ```python /home/david/Documents/GitHub/cleanoc/cleanrl/ppo_lngru_jit.py:281: UserWarning:...
Today, when doing `t.grad_fn.register_hook(foo)`, this registers a post hook on the Node corresponding to `t.grad_fn`. There is no API to add a pre-hook and we should add one. Adding a...
### 🐛 Describe the bug Over the past day, this job has been failing randomly after 20 minutes for memory issues. We are now disabling it in #82862 until this...
In the context of AOTAutograd, we want to be able to use anomaly mode to enhance the backward graph with information about the forward's stack trace. Such information can be...
### 🐛 Describe the bug In exporter, there is a onnx.shape_inference_infer_shapes checker before exporting, but it is not using strict_mode as onnxruntime does before executing the model, so the discrepancy...
Following https://github.com/pytorch/pytorch/pull/74128 and https://github.com/pytorch/pytorch/pull/74362, this would be part 3 of the effort to reduce code duplication in the code that lowers reference quantized patterns to native quantized ops in fbgemm/qnnpack....
Fixes: https://github.com/pytorch/data/issues/718 Stack from [ghstack](https://github.com/ezyang/ghstack): * **#82975 [DataLoader] BC shuffle for MapDataPipe** * #82974 [DataPipe] Align shuffling behavior for IterDataPipe and MapDataPipe Add shuffling logic for `MapDataPipe` when using `DataLoader`
