README.md

Prize Qualification Baselines

This directory contains the baseline(s) that submissions must beat to qualify for prizes, see the Scoring Section of the competition rules. For each ruleset there are 2 baselines (*_target_setting.py and *_full_budget.py). A submission must beat both baselines to be eligible for prizes.

The experiment logs with training metrics are in prize_qualification_baselines/logs

Externally Tuned Ruleset

JAX

The prize qualification baseline submissions for JAX are:

• prize_qualification_baselines/external_tuning/jax_nadamw_target_setting.py
• prize_qualification_baselines/external_tuning/jax_nadamw_full_budget.py

Example command:

python3 submission_runner.py \
    --framework=jax \
    --data_dir=<data_dir> \
    --experiment_dir=<experiment_dir> \
    --experiment_name=<experiment_name> \
    --workload=<workload> \
    --submission_path=prize_qualification_baselines/external_tuning/jax_nadamw_target_setting.py \
    --tuning_search_space=prize_qualification_baselines/external_tuning/tuning_search_space.json

PyTorch

The prize qualification baseline submissionss for PyTorch are:

• prize_qualification_baselines/external_tuning/pytorch_nadamw_target_setting.py
• prize_qualification_baselines/external_tuning/pytorch_nadamw_full_budget.py

Example command:

torchrun --redirects 1:0,2:0,3:0,4:0,5:0,6:0,7:0 --standalone --nnodes=1 --nproc_per_node=8 submission_runner.py \
    --framework=pytorch \
    --data_dir=<data_dir> \
    --experiment_dir=<experiment_dir> \
    --experiment_name=t<experiment_name> \
    --workload=<workload>\
    --submission_path=prize_qualification_baselines/external_tuning/pytorch_nadamw_target_setting.py \
    --tuning_search_space=prize_qualification_baselines/external_tuning/tuning_search_space.json

Self-tuning Ruleset

JAX

The prize qualification baseline submissionss for jax are:

• prize_qualification_baselines/self_tuning/jax_nadamw_target_setting.py
• prize_qualification_baselines/self_tuning/jax_nadamw_full_budget.py

Example command:

python3 submission_runner.py \
    --framework=jax \
    --data_dir=<data_dir> \
    --experiment_dir=<experiment_dir> \
    --experiment_name=<experiment_name> \
    --workload=<workload> \
    --submission_path=prize_qualification_baselines/self_tuning/jax_nadamw_target_setting.py \
    --tuning_ruleset=self

PyTorch

The prize qualification baseline submissionss for PyTorch are:

• prize_qualification_baselines/self_tuning/pytorch_nadamw_target_setting.py
• prize_qualification_baselines/self_tuning/pytorch_nadamw_full_budget.py

Example command:

torchrun --redirects 1:0,2:0,3:0,4:0,5:0,6:0,7:0 --standalone --nnodes=1 --nproc_per_node=8 submission_runner.py \
    --framework=pytorch \
    --data_dir=<data_dir> \
    --experiment_dir=<experiment_dir> \
    --experiment_name=t<experiment_name> \
    --workload=<workload>\
    --submission_path=prize_qualification_baselines/self_tuning/pytorch_nadamw_target_setting.py \
    --tuning_ruleset=self