CatXG/carememx-benchmark-v0

CareMem-X Benchmark v0

Multilingual Multimodal Memory Recall Benchmark for evaluating Gemma 4 as a memory assistant.

Overview

• Episodes: 200 synthetic memory episodes
• Total memories: 1745 items
• Total questions: 1187 recall questions
• Languages: English, Hindi, Hinglish
• Task families: recall, update, disambiguation, cross-lingual, repeated-question, noisy-capture, caregiver-handoff, uncertainty

Schema

Each episode contains:

• memories: list of stored memory items
• questions: list of recall questions with expected answers
• task_families: which benchmark tasks are covered
• languages: which languages appear

Usage

from datasets import load_dataset
ds = load_dataset("CatXG/carememx-benchmark-v0", split="train")