torchrec
torchrec copied to clipboard
pytorch
Pytorch domain library for recommendation systems
Summary: We got positive result from AI Labs for the cache but would like to quantify the gains in produciton. As torchrec is an open source lib, we cannot use...
Hi team, I'm wondering whether there is any way to apply AdamW for sparse features. I wrapped torch.optim.AdamW in KeyedOptimizerWrapper for dense features and it works. However, it seems that...
Summary: Symbolic trace passes in Proxy object which is incompatible with existing logic for caching kjt feature order in D68991644. Skipping this caching in these cases to avoid errors. Differential...
Summary: This diff adds support for compiling the optimizer step in the train pipeline of TorchRec. The changes include: 1. Add a new argument to the TrainPipeline constructor to enable...
Summary: Current failure on ``` buck run fbcode//mode/dev-nosan fbcode//caffe2/torch/sparse/tests:faster_hash_test ``` ``` Traceback (most recent call last): File "/data/users/kaus/fbsource/buck-out/v2/gen/fbcode/90ea5faddcf375ad/caffe2/torch/sparse/tests/__faster_hash_test__/faster_hash_test#link-tree/__run_lpar_main__.py", line 38, in __invoke_main() File "/data/users/kaus/fbsource/buck-out/v2/gen/fbcode/90ea5faddcf375ad/caffe2/torch/sparse/tests/__faster_hash_test__/faster_hash_test#link-tree/__run_lpar_main__.py", line 35, in __invoke_main run_as_main(module, main_function)...
Differential Revision: D71001957
Summary: Add ability to specify separate device for a CW sharded table during inference. NOTE: This diff only adds support for case where number of CW shards equal the world...
Differential Revision: D71000199
Differential Revision: D71000056
Differential Revision: D70999813