Large Relational Model (LRM)

This project implements LRM, a hybrid graph–embedding model that predicts the next relation in a sequence of knowledge graph triples.
It supports training on standard benchmarks such as FB15k-237 and WN18RR.

---

Setup

1. Create the data folder

mkdir -p data && cd data

2. Download FB15k-237

Clone the dataset from Hugging Face:

# Make sure git-lfs is installed (https://git-lfs.com)
git lfs install
git clone https://huggingface.co/datasets/KGraph/FB15k-237
# Optional: clone only pointer files (no large data)
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/KGraph/FB15k-237

After cloning, the train/valid/test files will be in:

data/FB15k-237/data/

3. Download WN18RR

# Make sure git-lfs is installed (https://git-lfs.com)
git lfs install
git clone https://huggingface.co/datasets/VLyb/WN18RR
# Optional: clone only pointer files (no large data)
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/VLyb/WN18RR

---

Running

From the project root:

python3 main.py

By default, it trains on data/FB15k-237/data. To switch to WN18RR, edit the last line of main.py:

model, ent2id, rel2id = train_lrm("data/WN18RR", epochs=20)

---

Requirements

• Python 3.10+
• PyTorch
• NumPy
• Git LFS

pip install torch numpy

---

Folder Structure

LRM/
 ├─ main.py
 ├─ README.md
 └─ data/
     ├─ FB15k-237/
     │   └─ data/
     │       ├─ train.txt
     │       ├─ valid.txt
     │       └─ test.txt
     └─ WN18RR/
         ├─ train.csv
         ├─ valid.csv
         └─ test.csv