Plonk / data /to_webdataset /launch_osv_5m_embeddings.py
nicolas-dufour's picture
squash: merge all unpushed commits
c4c7cee
import sys
from pathlib import Path
sys.path.append(str(Path(__file__).resolve().parent.parent.parent))
import argparse
import os
from jean_zay.launch import JeanZayExperiment
def parse_mode():
parser = argparse.ArgumentParser(description="Process some integers.")
parser.add_argument("--launch", action="store_true")
parser.add_argument("--src", help="path to source files")
parser.add_argument("--dest", help="path to destination files")
args = parser.parse_args()
return args
args = parse_mode()
dataset_path = Path(args.src)
list_of_shards = list(dataset_path.glob("*.tar"))
list_of_shards.sort()
cmd_modifiers = []
exps = []
exp_name = f"preprocess_data"
job_name = f"preprocess_data"
jz_exp = JeanZayExperiment(
exp_name,
job_name,
slurm_array_nb_jobs=len(list_of_shards),
cmd_path="data/to_webdataset/osv_to_wds.py",
num_nodes=1,
qos="t3",
account="syq",
gpu_type="a100",
time="01:00:00",
)
exps.append(jz_exp)
trainer_modifiers = {}
exp_modifier = {
"--src": dataset_path,
"--dest": Path(args.dest),
"--shard_id": "${SLURM_ARRAY_TASK_ID}",
}
cmd_modifiers.append(dict(trainer_modifiers, **exp_modifier))
if __name__ == "__main__":
for exp, cmd_modifier in zip(exps, cmd_modifiers):
exp.build_cmd(cmd_modifier)
if args.launch == True:
exp.launch()