davanstrien's picture
download
raw
1.19 kB
# /// script
# requires-python = ">=3.11"
# dependencies = ["saturate[hf]>=0.1.2"]
# ///
"""Cheap laptop check of the source half of the driver: does ids='id' key rows by
the dataset's own id, and does batched() carry them through intact?"""
import importlib.util, sys
spec = importlib.util.spec_from_file_location("ec", "embed_column.py")
ec = importlib.util.module_from_spec(spec); spec.loader.exec_module(ec)
from saturate import dataset_rows
for col in ["caption", "scene", "events_text"]:
rows = dataset_rows(ec.SOURCE, config="default", split="train",
columns=[col], ids="id")
n_items = 0
for k, (bid, payload) in enumerate(ec.batched(rows, col)):
assert bid.startswith("b-"), bid
assert len(payload["ids"]) == len(payload["texts"]), payload["ids"]
assert all(isinstance(i, str) and i for i in payload["ids"])
n_items += len(payload["texts"])
if k == 0:
print(f"{col}: batch0 id={bid!r} ids[:2]={payload['ids'][:2]} "
f"len={len(payload['texts'])} text0={payload['texts'][0][:60]!r}")
if k == 4:
break
print(f"{col}: 5 batches -> {n_items} items\n")

Xet Storage Details

Size:
1.19 kB
·
Xet hash:
e6759014b11434f1acf6aa48ab133b57249db5d828cf21540f326fa4709291c5

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.