torchrec
torchrec copied to clipboard
pytorch
Pytorch domain library for recommendation systems
It's much more efficient and is also idiomatic.
when changing the number of embeddings to 4091, and mch_size to 1021 of the code below, it will throw the following exception ``` ValueError: ShardedTensor global_size property does not match...
Summary: fbgemm/sparse_ops.py meta fn, quant_embedding.py changes that should allow to run dynamo export Original diff D51354832 from IvanKobzarev Differential Revision: D52207064
In `/examples/inference/dlrm_client.py`, the sparse features values are parsed as `Int64` type. ``` id_list_features = predictor_pb2.SparseFeatures( num_features=args.num_id_list_features, values=to_bytes(batch.sparse_features.values()), lengths=to_bytes(batch.sparse_features.lengths()), ) ``` However, in `torchrec/inference/src/Batching.cpp` line 171 and line 208, the combineSparse...
Summary: 1. Some general MCH changes and fixes such as ```is self.training``` control flow. 2. Added ```threshold_filtering_func``` to allow users to 'gate' entry into ZCH based on a user-provided frequency-based...
Summary: Micro hash Lookup, a LUT learnt from both sparse ID and optionally embeddings. 1. reduced hash collision: id is mapped to `num_shards` hash slots and embedding is concat from...
Currently attempting to export a model that uses KeyedJaggedTensor to ONNX format results in the following error: ```python gm = symbolic_trace(model) scripted_gm: torch.jit.ScriptModule | torch.jit.ScriptFunction = torch.jit.script(gm) # type: ignore...
Summary: This diff is reverting D51718962 D51718962: [torchrec][vbe] fix publish error from fx tracing check in unsharded ebc by joshuadeng has been identified to be causing the following test failure:...
Summary: Created from CodeHub with https://fburl.com/edit-in-codehub Differential Revision: D51922412