Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[WIP][Experiment] Use multi-threading for multi-file json reads #94

Draft
wants to merge 4 commits into
base: main
Choose a base branch
from
Draft

[WIP][Experiment] Use multi-threading for multi-file json reads #94

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
c7a2e0a
add experimental code path for multi-file json reads
rjzamora May 31, 2024
0f80085
adjust name
rjzamora May 31, 2024
51fa0e2
only use with cudf for now
rjzamora May 31, 2024
139bd51
support pandas
rjzamora May 31, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 19 additions & 1 deletion nemo_curator/utils/distributed_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,15 +15,20 @@

import ast
import os
from io import BytesIO

import fsspec

os.environ["RAPIDS_NO_INITIALIZE"] = "1"
import warnings
from contextlib import nullcontext
from pathlib import Path
from typing import Union

import dask
import dask.dataframe as dd
import pandas as pd
from dask.base import tokenize
from dask.distributed import Client, LocalCluster, get_worker, performance_report

from nemo_curator.utils.gpu_utils import GPU_INSTALL_STRING, is_cudf_type
Expand Down Expand Up @@ -240,7 +245,7 @@ def read_single_partition(
if add_filename:
read_files_one_at_a_time = True
else:
if backend == "cudf":
if backend == "cudf" or filetype == "jsonl":
# cuDF supports reading multiple files at once
read_files_one_at_a_time = False
else:
Expand All @@ -259,6 +264,19 @@ def read_single_partition(
df["filename"] = os.path.basename(file)
df_ls.append(df)
df = concat_f(df_ls, ignore_index=True)
elif filetype == "jsonl":
fs = fsspec.core.get_fs_token_paths(files[0])[0]
token = tokenize(files)
name = f"get_bytes-{token}"
dsk = {(name, i): (fs.cat_file, path) for i, path in enumerate(files)}
if backend == "cudf":
# Cudf can read from list of bytes
dsk[name] = (lambda x: x, list(dsk.keys()))
df = read_f(dask.threaded.get(dsk, name), **read_kwargs)
else:
# Pandas requires single BytesIO object
dsk[name] = (b"".join, list(dsk.keys()))
df = read_f(BytesIO(dask.threaded.get(dsk, name)), **read_kwargs)
else:
df = read_f(files, **read_kwargs)
df = df[sorted(df.columns)]
Expand Down