lavello_mlips

A Python package for processing omol-25 data using MPI.

Installation

You can install this package locally:

uv pip install -e .

Usage

This package provides three primary command-line interfaces:

1. Processing Data

Extract, process, and combine molecular data from an S3 bucket (or local directory):

lavello_mlips --help

• MPI Support: Add --mpi and run via mpirun to distribute tasks across multiple workers natively via hybrid RMA.
• Smart Restart: Add --restart to automatically sweep the output directory, recover orphaned Parquet/XYZ pairs, and pick up right where you left off.
• Logging: Specify --log-file my_log.log to write text streams to disk (existing logs are automatically appended to, not overwritten).
• Batch Flushing: Use --batch-size N to control disk writes. If not specified, workers dynamically flush at 1% increments (with a strict minimum of 100 output structures).

2. Downloading Raw Data

Download original raw orca.out datasets from S3 without running processing logic natively on them:

download_omol25 --help

3. Verification Utility

Cross-reference a generated Parquet dataset with its respective ExtXYZ file to guarantee absolutely zero data corruption or structural mismatching:

verify_processed_omol25 --parquet props_group.parquet --extxyz structs_group.xyz

• This rigorously structurally aligns both tables via geom_sha1 and flags any mathematically misassigned properties.
• Embedded timing metadata such as process_time_s are strictly and unconditionally excluded to prevent false-positive errors.

License

This project is licensed under the BSD 3-Clause License - see the LICENSE file for details.