Building and Benchmarking DLRM on Arm Neoverse V2 with MLPerf
Introduction
Overview and setup
Download model weights and data
Run the benchmark
Next Steps

  Log an issue

  Fork and edit

  Discuss on Discord

Building and Benchmarking DLRM on Arm Neoverse V2 with MLPerf

Before building the model, you’ll need to obtain the data and model weights.

Start by creating directories for the data and model weights on your cloud instance.

    

        
        
cd $HOME
mkdir data
mkdir model

Install rclone

You’ll use rclone to download the data and model weights .

Install rclone using the bash script:

    

        
        
curl https://rclone.org/install.sh | sudo bash

You should see a similar output if the tools installed successfully:

    

        
        rclone v1.69.1 has successfully installed.
Now run "rclone config" for setup. Check https://rclone.org/docs/ for more details.

Configure the following credentials for rclone:

    

        
        
rclone config create mlc-inference s3 provider=Cloudflare \
    access_key_id=f65ba5eef400db161ea49967de89f47b \
    secret_access_key=fbea333914c292b854f14d3fe232bad6c5407bf0ab1bebf78833c2b359bdfd2b \
    endpoint=https://c2686074cb2caf5cbaf6d134bdba8b47.r2.cloudflarestorage.com

Run the commands below to download the data and model weights. This process can take 30 minutes or more depending on the internet connection in your cloud instance.

    

        
        
rclone copy mlc-inference:mlcommons-inference-wg-public/dlrm_preprocessed $HOME/data  -P
rclone copy mlc-inference:mlcommons-inference-wg-public/model_weights $HOME/model/model_weights -P

Once it finishes, you should see that the model and data directories are populated. With the data in place, you can proceed to run the benchmark in order to measure the performance of the downloaded DLRM model.

Back
Next