Skip to content

fix: isolate embedding staging dir to prevent cross-model contamination #18

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

Open
vojtech-cifka wants to merge 2 commits into
base: master
Choose a base branch
from
Open

fix: isolate embedding staging dir to prevent cross-model contamination #18

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
fa85342
fix: isolate embedding staging dir to prevent cross-model contamination
vojtech-cifka Jun 19, 2026
4167bb4
fix: format
vojtech-cifka Jul 12, 2026
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
30 changes: 18 additions & 12 deletions preprocessing/embeddings.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
import asyncio
import shutil
import tempfile
import time
from pathlib import Path
from typing import Any
Expand Down Expand Up @@ -221,20 +221,26 @@ def main(config: DictConfig, logger: MLFlowLogger) -> None:
max_concurrency=config.concurrency,
)

split_dir = Path(config.output_dir) / str(name)
split_dir.mkdir(parents=True, exist_ok=True)
tiles_parquet_dir = split_dir / "tiles"
if tiles_parquet_dir.exists():
shutil.rmtree(tiles_parquet_dir)
# Isolated per-run staging dir: avoids cross-model contamination and
# stale-file carryover from concurrent or previous runs sharing
# output_dir. Auto-removed after upload; nothing reads it afterward
# (training reads the mlflow run artifacts via runs:/ URIs).
Path(config.output_dir).mkdir(parents=True, exist_ok=True)
with tempfile.TemporaryDirectory(dir=config.output_dir) as tmp_root:
split_dir = Path(tmp_root) / str(name)
tiles_parquet_dir = split_dir / "tiles"
split_dir.mkdir(parents=True, exist_ok=True)

slides.to_parquet(split_dir / "slides.parquet", index=False)
slides.to_parquet(split_dir / "slides.parquet", index=False)

t = time.monotonic()
print(f"[main] starting write_parquet for split={name}")
ds.write_parquet(str(tiles_parquet_dir), min_rows_per_file=config.rows_per_file)
print(f"[main] write_parquet finished in {time.monotonic() - t:.1f}s")
t = time.monotonic()
print(f"[main] starting write_parquet for split={name}")
ds.write_parquet(
str(tiles_parquet_dir), min_rows_per_file=config.rows_per_file
)
print(f"[main] write_parquet finished in {time.monotonic() - t:.1f}s")

logger.log_artifacts(str(split_dir), str(name))
logger.log_artifacts(str(split_dir), str(name))


if __name__ == "__main__":
Expand Down