---
license: apache-2.0
task_categories:
- visual-question-answering
- question-answering
- image-retrieval
language:
- en
pretty_name: MM-MemT processed datasets
---

# MM-MemT processed datasets

This repository contains processed datasets used by MM-MemT experiments and OOD evaluations. The folders are arranged to match the local MM-MemT project structure so the evaluation scripts can be run with the same `--dataset-dir` paths after download.

Included folders:

- `processed/dataset/dataset_curated_750`
- `processed/dataset/dataset_expanded_retrieval_wma_mg_memeye`
- `processed/dataset/dataset_worldmemarena_train_valid_300`
- `processed/dataset/dataset_memgallery`
- `processed/dataset_memeye_train_valid`
- `processed/dataset_memlens_32k_agent195`
- `processed/dataset_mixed_card`
- `processed/dataset_disbench_metadata_only`

Large raw upstream datasets are not duplicated here unless needed by the processed MM-MemT format.