Fix Milvus HNSW-PQ m parameter conflicts with M.

[bashimao]

The m parameter in Milvus HNSW-PQ is not adjustable when using the command-line. Even worse, setting M and m leads to a situation where the order in which you define the parameters in your YAML-file determines what value ends up in M.

Proposed solution

Add a separate command-line switch --pq-m to specify the m for product quantization.

However, while this fixes the frontend, which is sufficient for my current needs, I strongly suggest you consider the renaming the parameter in the backend as well. This idea to have parameters that only differ in upper/lower-case (i.e., M vs. m) is an accident bound to happen.

To replicate the problem

git checkout main

echo "milvushnswpq:
  dry_run: True
  drop_old: False
  load: False
  load_concurrency: 48
  search_serial: True
  search_concurrent: True
  case_type: Performance768D1M
  db_label: perf_768d_1m
  k: 200
  concurrency_duration: 60
  num_concurrency: 128
  concurrency_timeout: 1800
  serial_cooldown: 5
  task_label: search_perf_768d_1m
  uri: http://milvus-standalone:19530
  db_name: perf_768d_1m
  num_shards: 2
  replica_number: 2
  use_partition_key: True
  M: 8
  ef_construction: 384
  ef_search: 128
  m: 64
  refine: True
  refine_type: fp16
  refine_k: 2.0
  nbits: 16" > test.yaml

python3 -m vectordb_bench.cli.vectordbbench milvushnswpq --config-file test.yaml

Output:

2026-07-17 14:49:19,728 | INFO: Task:
TaskConfig(
  db=<DB.Milvus: 'Milvus'>,
  ...
  db_case_config=HNSWPQConfig(index=<IndexType.HNSW_PQ: 'HNSW_PQ'>,
    metric_type=None, 
    use_partition_key=True, M=64, efConstruction=384, ef=128,
    m=32, nbits=16, refine=True, refine_type=<SQType.FP16: 'FP16'>, refine_k=2.0
  ), ...

Note how M=64 now. Because m is the later value. m=32 because it is literally not hooked up in the backend.

To replicate the fix

git checkout hnswpq_pram_passing  # assuming you got bashimao/VectorDBBench cloned

python3 -m vectordb_bench.cli.vectordbbench milvushnswpq --config-file test.yaml

Output:

Usage: python -m vectordb_bench.cli.vectordbbench milvushnswpq
           [OPTIONS]
Try 'python -m vectordb_bench.cli.vectordbbench milvushnswpq --help' for help.

Error: Missing option '--pq-m'.

git checkout hnswpq_pram_passing  # assuming you got bashimao/VectorDBBench cloned

echo "milvushnswpq:
  dry_run: True
  drop_old: False
  load: False
  load_concurrency: 48
  search_serial: True
  search_concurrent: True
  case_type: Performance768D1M
  db_label: perf_768d_1m
  k: 200
  concurrency_duration: 60
  num_concurrency: 128
  concurrency_timeout: 1800
  serial_cooldown: 5
  task_label: search_perf_768d_1m
  uri: http://milvus-standalone:19530
  db_name: perf_768d_1m
  num_shards: 2
  replica_number: 2
  use_partition_key: True
  m: 8
  ef_construction: 384
  ef_search: 128
  pq_m: 64
  refine: True
  refine_type: fp16
  refine_k: 2.0
  nbits: 16" > test.yaml

Output:

2026-07-17 15:11:49,018 | INFO: Task:
TaskConfig(
  db=<DB.Milvus: 'Milvus'>,
  ...
  db_case_config=HNSWPQConfig(index=<IndexType.HNSW_PQ: 'HNSW_PQ'>,
    metric_type=None, use_partition_key=True, M=8, efConstruction=384, ef=128,
    m=64, nbits=16, refine=True, refine_type=<SQType.FP16: 'FP16'>, refine_k=2.0
  ), ...

Note how M=8 regardless of whether it is spelled upper- or lowercase in the YAML, and m=64 because in the YAML pq_m: 64.

[sre-ci-robot]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by: bashimao
To complete the pull request process, please assign xuanyang-cn after the PR has been reviewed.
You can assign the PR to them by writing /assign @xuanyang-cn in a comment when ready.

The full list of commands accepted by this bot can be found here.

Details

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment