Skip to content

feat(benchmark): add gist and text2image cases and dataset preparation tools - #820

Open
richyreachy wants to merge 2 commits into
alibaba:mainfrom
richyreachy:feat/more-metrics-to-cont-bench
Open

richyreachy wants to merge 2 commits into
alibaba:mainfrom
richyreachy:feat/more-metrics-to-cont-bench

Conversation

@richyreachy

Copy link
Copy Markdown
Collaborator

Extend continuous benchmark coverage with GIST-1M (squared L2) and
Text2Image-1M (inner product), while preserving the existing Cohere and
OpenAI HNSW cases and their parameters.

Changes

  • Add dataset download, Parquet conversion, and exact ground-truth generation
    scripts, with cache validation and source metadata.
  • Expand the HNSW matrix from 12 to 20 runs across int8, int4, fp16, and fp32.
  • Isolate index paths, logs, results, and monitoring identities for each case.
  • Add a side-effect-free --dry-run mode and local configuration overrides.
  • Add resumable downloads and preparation tools for:
    • GIST-1M from a local TexMex archive, using its queries and published GT.
    • Text2Image-1M/10M with all 100K public queries, using exact generated
      1M GT and official 10M GT.
    • Native Cohere-1M/10M evaluation files.
  • Document dataset layouts, preparation commands, and benchmark usage.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant