LongMemEval Dataset Download Guide

The LongMemEval datasets are large files (several GB) hosted on HuggingFace with Git LFS. Here are all the ways to download them:

Option 1: JavaScript/Node.js Download

bash

export HF_TOKEN=your_token_here
# or
export HUGGINGFACE_TOKEN=your_token_here

bash

pnpm install
pnpm download:hf

bash

# macOS
brew install git-lfs

# Ubuntu/Debian
sudo apt-get install git-lfs

# Initialize Git LFS
git lfs install

bash

git clone https://huggingface.co/datasets/xiaowu0162/longmemeval
cd longmemeval
cp *.json ../data/

bash

cd packages/longmemeval/data
tar -xzvf ~/Downloads/longmemeval_data.tar.gz

bash

ls -lh *.json
# You should see:
# - longmemeval_s.json (~40MB)
# - longmemeval_m.json (~200MB)
# - longmemeval_oracle.json (~2MB)

If you have a HuggingFace account:

This means the download failed due to authentication. Use one of the authenticated methods above.

HuggingFace has bandwidth limits. Try:

Make sure you're logged in to HuggingFace and have accepted any dataset terms of use.

After downloading, verify the files:

bash

# Check file sizes
ls -lh data/*.json

# Check file content (should be valid JSON)
head -n 5 data/longmemeval_s.json

Expected sizes: