Skip to content

Latest commit

 

History

History
96 lines (77 loc) · 3.75 KB

README.md

File metadata and controls

96 lines (77 loc) · 3.75 KB

PyTorch MEMREC-DLRM inference

Description

This document has instructions for running MEMREC-DLRM inference using Intel-optimized PyTorch for bare metal.

Bare Metal

Setting up MEMREC-DLRM:

  • Create conda environment: conda create -n memrec python=3.11
  • Activate conda environment: conda activate memrec
  • Install Intel® Extension for PyTorch: pip install intel-extension-for-pytorch
  • Install torch:
    • CPU: conda install pytorch torchvision torchaudio cpuonly -c pytorch
  • Install mlperf logger:
    • git clone https://github.com/mlperf/logging.git mlperf-logging
    • pip install -e mlperf-logging
  • Install dependencies: pip install -r quickstart/recommendation/pytorch/memrec_dlrm/requirements.txt
  • Install memrec encoder:
    • cd models/recommendation/pytorch/memrec_dlrm/memrec_encoder/code/
    • python setup.py install --user

Datasets

Criteo Terabyte Dataset

The Criteo Terabyte Dataset is used to run MEMREC. To download the dataset, you will need to visit the Criteo website and accept their terms of use: https://labs.criteo.com/2013/12/download-terabyte-click-logs/. Copy the download URL into the command below as the <download url> and replace the <dir/to/save/MEMREC_data> to any path where you want to download and save the dataset.

export DATASET_DIR=<dir/to/save/MEMREC_data>

mkdir ${DATASET_DIR} && cd ${DATASET_DIR}
curl -O <download url>/day_{$(seq -s , 0 23)}.gz
gunzip day_*.gz

The raw data will be automatically preprocessed and saved as day_*.npz to the DATASET_DIR when MEMREC is run for the first time. On subsequent runs, the scripts will automatically use the preprocessed data.

Quick Start Scripts

Script name Description
inference_performance.sh Run inference to verify performance for the specified precision (fp32).
accuracy.sh Measures the inference accuracy for the specified precision (fp32).

Run the model

Download the MEMREC-DLRM PyTorch weights and set the WEIGHT_PATH to point to the weights file.

export WEIGHT_PATH=<path to the memrec_CriteoTB_D75000dw75000_K1kw1.pt file>

Follow the instructions above to setup your bare metal environment, do the model-specific setup and download and prepropcess the dataset. Once all the setup is done, the Model Zoo can be used to run a quickstart script. Ensure that you have enviornment variables set to point to the dataset directory, precision, weights file, and an output directory.

# Clone the model zoo repo and set the MODEL_DIR
git clone https://github.com/IntelAI/models.git
cd models
export MODEL_DIR=$(pwd)

# Env vars
export PRECISION=<specify the precision to run>
export WEIGHT_PATH=<path to the memrec pt file> # only needed for testing accuracy
export DATASET_DIR=<path to the dataset>
export OUTPUT_DIR=<directory where log files will be written>

# Run a quickstart script (for example, bare metal performance)
cd ${MODEL_DIR}/quickstart/recommendation/pytorch/memrec_dlrm/inference/cpu
bash inference_performance.sh

# Evaluate Inference Accuracy (for example, bare metal performance)
cd ${MODEL_DIR}/quickstart/recommendation/pytorch/memrec_dlrm/inference/cpu
bash accuracy.sh

License

LICENSE

THIRD PARTY LICENSES