qdrant's bencnmark is reporting an extremely high latencies for on-disk index qith 140M vectors

Testing qdrant on 8 cores, 64GB of memory r6i.2xlalrge instance

Here is the collection's configuration:

{
  "params":{
    "vectors":{
      "size":96
      "distance":"Euclid"
     }
   "shard_number":1
   "replication_factor":1
   "write_consistency_factor":1
   "on_disk_payload":true
  }
"hnsw_config":{
    "m":16
    "ef_construct":128
    "full_scan_threshold":10000
    "max_indexing_threads":0
    "on_disk":true
}
"optimizer_config":{
    "deleted_threshold":0.2
    "vacuum_min_vector_number":1000
    "default_segment_number":0
    "max_segment_size": NULL
    "memmap_threshold": NULL
    "indexing_threshold":20000
    "flush_interval_sec":5
    "max_optimization_threads":0
}
"wal_config":{
    "wal_capacity_mb":32
    "wal_segments_ahead":0
}
"quantization_config": NULL
}

Current Behavior

I downloaded and inserted about 140M vectors from Yandex https://research.yandex.com/blog/benchmarks-for-billion-scale-similarity-search. Upon testing I'm getting about 1.6 - 1.7 vector lookups per second, so overall 580-625ms search latency per query lookup. We are specifically trying to observe mem-mapped file performance in this case. Would you advise on anything in a configuration that would help us to optimize index performance and get better results?

Steps to Reproduce

See above

Expected Behavior

Expecting to see decent latencies for vector lookups.

qdrant / vector-db-benchmark