TorchRec (Beta Release)


TorchRec is a PyTorch domain library built to provide common sparsity & parallelism primitives needed for large-scale recommender systems (RecSys). It allows authors to train models with large embedding tables sharded across many GPUs.

TorchRec contains:

  • Parallelism primitives that enable easy authoring of large, performant multi-device/multi-node models using hybrid data-parallelism/model-parallelism.
  • The TorchRec sharder can shard embedding tables with different sharding strategies including data-parallel, table-wise, row-wise, table-wise-row-wise, and column-wise sharding.
  • The TorchRec planner can automatically generate optimized sharding plans for models.
  • Pipelined training overlaps dataloading device transfer (copy to GPU), inter-device communications (input_dist), and computation (forward, backward) for increased performance.
  • Optimized kernels for RecSys powered by FBGEMM.
  • Quantization support for reduced precision training and inference.
  • Common modules for RecSys.
  • Production-proven model architectures for RecSys.
  • RecSys datasets (criteo click logs and movielens)
  • Examples of end-to-end training such the dlrm event prediction model trained on criteo click logs dataset.


Torchrec requires Python >= 3.7 and CUDA >= 11.0 (CUDA is highly recommended for performance but not required). The example below shows how to install with CUDA 11.3. This setup assumes you have conda installed.


Experimental binary on Linux for Python 3.7, 3.8 and 3.9 can be installed via pip wheels


conda install pytorch cudatoolkit=11.3 -c pytorch-nightly
pip install torchrec-nightly

CPU Only

conda install pytorch cpuonly -c pytorch-nightly
pip install torchrec-nightly-cpu

Colab example: introduction + install

See our colab notebook for an introduction to torchrec which includes runnable installation. – Tutorial Source – Open in Google Colab

From Source

We are currently iterating on the setup experience. For now, we provide manual instructions on how to build from source. The example below shows how to install with CUDA 11.3. This setup assumes you have conda installed.

  1. Install pytorch. See pytorch documentation

    conda install pytorch cudatoolkit=11.3 -c pytorch-nightly
  2. Install Requirements

    pip install -r requirements.txt
  3. Next, install FBGEMM_GPU from source (included in third_party folder of torchrec) by following the directions here. Installing fbgemm GPU is optional, but using FBGEMM w/ CUDA will be much faster. For CUDA 11.3 and SM80 (Ampere) architecture, the following instructions can be used:

    export CUB_DIR=/usr/local/cuda-11.3/include/cub
    export CUDA_BIN_PATH=/usr/local/cuda-11.3/
    export CUDACXX=/usr/local/cuda-11.3/bin/nvcc
    python install -DTORCH_CUDA_ARCH_LIST="7.0;8.0"

    The last line of the above code block (python install…) which manually installs fbgemm_gpu can be skipped if you do not need to build fbgemm_gpu with custom build-related flags. Skip to the next step if that is the case.

  4. Download and install TorchRec.

    git clone --recursive
    # cd to the directory where torchrec's is located. Then run one of the below:
    cd torchrec
    python install develop --skip_fbgemm  # If you manually installed fbgemm_gpu in the previous step.
    python install develop                # Otherwise. This will run the fbgemm_gpu install step for you behind the scenes.
    python install develop --cpu_only     # For a CPU only installation of FBGEMM
  5. Test the installation.

    torchx run -s local_cwd --script

    See TorchX for more information on launching distributed and remote jobs.

  6. If you want to run a more complex example, please take a look at the torchrec DLRM example.


TorchRec is BSD licensed, as found in the LICENSE file.


View Github