minhash.py

from datatrove.executor import LocalPipelineExecutor
from datatrove.pipeline.readers import ParquetReader
from datatrove.utils.hashing import HashConfig
from custom_minhash import (
    CustomMinhashConfig,
    CustomMinhashDedupSignature,
    CustomMinhashDedupBuckets,
    CustomMinhashDedupCluster,
    CustomMinhashDedupFilter,
)
from datatrove.pipeline.readers import JsonlReader
from datatrove.pipeline.tokens import TokensCounter
from datatrove.pipeline.writers.jsonl import JsonlWriter
from datatrove.utils.hashing import HashConfig
 
import argparse
from glob import glob

custom_minhash_config = CustomMinhashConfig(
    hash_config=HashConfig(precision=32, hash_fc='sha1'),
)
 
if __name__ == '__main__':
    parser = argparse.ArgumentParser(description="Demo script with argparse")
    parser.add_argument("--sub-folder", type=str, required=True, help="Subfolder path")
    parser.add_argument("--offset", type=int, default=0, required=False, help="task offset")
    # parser.add_argument("--local-task", type=int, default=2, required=False, help="local task number")

    args = parser.parse_args()
    print(args)

    MINHASH_BASE_PATH = "minhash"
     
    LOGS_FOLDER = "minhash_logs"
    
    n_file = len(glob(f"/gpfs/public/research/liyizhi/huggingface/datasets/fineweb-edu-score-2/data/{args.sub_folder}/*.parquet"))
    TOTAL_TASKS = n_file
    print(f"Total files in {args.sub_folder}: {n_file}")

    INPUT_READER = ParquetReader(
        "/gpfs/public/research/liyizhi/huggingface/datasets/fineweb-edu-score-2", 
        glob_pattern=f"data/{args.sub_folder}/*.parquet",
        batch_size=100_000)
    # INPUT_READER = ParquetReader("data/CC-MAIN-2013-20/", limit=3)
    # INPUT_READER = JsonlReader('jsonl')
    
    # stage 1 computes minhash signatures for each task (each task gets a set of files)
    stage1 = LocalPipelineExecutor(
        # job_name="mh1",
        pipeline=[
            INPUT_READER,
            CustomMinhashDedupSignature(
                output_folder=f"{MINHASH_BASE_PATH}/signatures",
                config=custom_minhash_config,
                naming_prefix=args.sub_folder,
            ),
        ],
        tasks=TOTAL_TASKS,
        # local_tasks=LOCAL_TASKS,
        # local_rank_offset=TASK_OFFSET,
        logging_dir=f"{LOGS_FOLDER}/signatures",
        # slurm_logs_folder=f"{LOCAL_LOGS_FOLDER}/signatures/slurm_logs",
    )
     
    # stage 2 finds matches between signatures in each bucket
    stage2 = LocalPipelineExecutor(
        # job_name="mh2",
        pipeline=[
            CustomMinhashDedupBuckets(
                input_folder=f"{MINHASH_BASE_PATH}/signatures",
                output_folder=f"{MINHASH_BASE_PATH}/buckets",
                config=custom_minhash_config,
            ),
        ],
        tasks=custom_minhash_config.num_buckets,
        local_tasks=1,
        local_rank_offset=args.offset,
        # time="90:00:00",
        # partition="hopper-prod",
        logging_dir=f"{LOGS_FOLDER}/buckets",
        # depends=stage1,
    )
    
    # stage 3 creates clusters of duplicates using the results from all buckets
    stage3 = LocalPipelineExecutor(
        # job_name="mh3",
        pipeline=[
            CustomMinhashDedupCluster(
                input_folder=f"{MINHASH_BASE_PATH}/buckets",
                output_folder=f"{MINHASH_BASE_PATH}/remove_ids",
                config=custom_minhash_config,
            ),
        ],
        # tasks=1,
        # time="90:00:00",
        # partition="hopper-prod",
        logging_dir=f"{LOGS_FOLDER}/clusters",
        # mem_per_cpu_gb=70,
        # cpus_per_task=2,
        # depends=stage2,
        # slurm_logs_folder=f"{LOCAL_LOGS_FOLDER}/clusters/slurm_logs",
    )
    
    # stage 4 reads the original input data and removes all but 1 sample per duplicate cluster
    # the data must match exactly stage 1, so number of tasks and the input source must be the same
    stage4 = LocalPipelineExecutor(
        # job_name="mh4",
        pipeline=[
            INPUT_READER,
            TokensCounter(),  # nice way to see how many tokens we had before and after deduplication
            CustomMinhashDedupFilter(
                remove_id_input_folder=f"{MINHASH_BASE_PATH}/remove_ids",
                sig_input_folder=f"{MINHASH_BASE_PATH}/signatures",
                exclusion_writer=JsonlWriter(f"{MINHASH_BASE_PATH}/removed"),
                config=custom_minhash_config,
                naming_prefix=args.sub_folder,
            ),
        ],
        tasks=TOTAL_TASKS,
        # time="50:00:00",
        # partition="hopper-cpu",
        logging_dir=f"{LOGS_FOLDER}/filter",
        depends=stage3,
        # slurm_logs_folder=f"{LOCAL_LOGS_FOLDER}/filter/slurm_logs",
    )
    
    stage3.run()