torchrec
torchrec copied to clipboard
pytorch
Pytorch domain library for recommendation systems
Summary: Adding numerical equivalence tests for different kernel types. Note that when stochastic rounding = True + is_training = True is always a bit tricky. However, from what FBGEMM team...
Summary: the d2h sync in `keyed_jagged_index_select_dim1` can be prevented if we provide the output tensor size ahead of time. when we know the recat tensor is same size as the...
Hello, I compressed the values, and lengths to reduce the PCIe cost. then I want to move the keys, compressed values and lengths to GPU, and then decompress and generate...
Running the Torchrec_Introduction on pytorch.org at https://pytorch.org/tutorials/intermediate/torchrec_tutorial.html with the provided Google Colab https://colab.research.google.com/github/pytorch/torchrec/blob/main/Torchrec_Introduction.ipynb#scrollTo=fgB-KSOTbPf7 gives an error when trying to run the first bit of python code ImportError: Failed to load...
Summary: dynamo does not support dataclass.field default_factory with Lists, avoiding them for now with specifying all arguments. Differential Revision: D53854370
I am attempting to convert an original PyTorch model into the torchrec version. During the experiment, I noticed that there is a clear inconsistency between results of the torchrec optimizer...
Summary: Copying from internal design doc: prototype generic training pipeline that can support arbitrary number of stages. The design is illustrated in the figure below. A pipeline is composed of...
Differential Revision: D53786021
Summary: `constraints` argument for `torch.export` has been deprecated in favor of the `dynamic_shapes` argument. This PR updates the use of the deprecated API in `apf`. In addition, this diff registers...
