Segmentation Fault for retrieval using index | Apple M3 chip

[Alireza-Zwolf]

Hi all,

I'm using the following system for reproducing the Pyserini: BGE-base Baseline for NFCorpus section results:

Macbook Air M3 (macOS 14.6)
Python 3.10
java 21.0.5
Maven 3.9.9

The index building phase works for both BAAI/bge-base-en-v1.5 and facebook/contriever-msmarco embedding models. However, I got segmentation fault in the retrieving phase for both indexes:

Running collections/nfcorpus/queries.tsv topics, saving to runs/run.beir.bge-base-en-v1.5.nfcorpus.txt...
  0%|                                                                                                 | 0/3237 [00:00<?, ?it/s][thread 27907 also had an error]
#
# A fatal error has been detected by the Java Runtime Environment:
#
#  SIGSEGV (0xb) at pc=0x000000011f48d828, pid=11716, tid=26883
#
# JRE version: Java(TM) SE Runtime Environment (21.0.5+9) (build 21.0.5+9-LTS-239)
# Java VM: Java HotSpot(TM) 64-Bit Server VM (21.0.5+9-LTS-239, mixed mode, sharing, tiered, compressed oops, compressed class ptrs, g1 gc, bsd-aarch64)
# Problematic frame:
# C  [libomp.dylib+0x69828]  void __kmp_suspend_64<false, true>(int, kmp_flag_64<false, true>*)+0x2c
#
# Core dump will be written. Default location: /cores/core.11716
#

Seems like there have been some similar issues like #938 and #1874 on M1 processors. However, I use the development installation (not the regular one which is recommended for Mac: link), and all relevant pyserini tests passed after the installation. Any idea on the reason behind of the mentioned issue?

[Alireza-Zwolf]

By using the Pyserini on Conda virtual environment as stated in #938, retrieval works without the mentioned SIGSEGV (signal segmentation violation) error.