Skip to content

[BENCHMARK] Add HPS (Human Preference Score) benchmark #512

@davidberenstein1957

Description

@davidberenstein1957

Description

Add HPS benchmark for text-to-image evaluation. Uses zhwang/HPDv2 dataset with 4 categories.

Details

  • Dataset: zhwang/HPDv2 (benchmark subfolder contains JSON files)
  • Categories: anime, concept-art, paintings, photo
  • Collate: prompt_with_auxiliaries_collate

Implementation

  • Add setup_hps_dataset in src/pruna/data/datasets/prompt.py
  • Register in base_datasets in src/pruna/data/__init__.py
  • Add BenchmarkInfo entry with metrics: ["hps"]
  • Add test in tests/data/test_datamodule.py

Acceptance Criteria

  • PrunaDataModule.from_string("HPS") works
  • Category filter: PrunaDataModule.from_string("HPS", category="anime")
  • Auxiliaries include category field
  • Test passes

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions