search

cmavro / GNN-RAG

GNN-RAG: Graph Neural Retrieval for Large Language Modeling Reasoning
252 stars 48 forks source link

About Pre-trained Checkpoints for ReaRev + LMSR #16

Open mufeili opened 1 week ago

mufeili commented 1 week ago

Thanks for the great work!

When I tried to train a GNN retriever myself following the instructions here, I encountered the issue in the title. Could you release the checkpoints required for training the GNN retrievers on WebQSP and CWQ? Thanks.

cmavro commented 1 week ago

Thanks for the interest in our work!

You can use ReaRev_CWQ.ckpt for --load_experiment, which is provided here, or you can train from scratch, omitting --load_experiment argument.

mufeili commented 1 week ago

Thank you for the kind response!

python main.py ReaRev --entity_dim 50 --num_epoch 200 --batch_size 8 --eval_every 2 --data_folder data/webqsp/ --lm sbert --num_iter 3 --num_ins 2 --num_gnn 3 --relation_word_emb True --load_experiment ReaRev_webqsp.ckpt --is_eval --name webqsp

It seems that the released checkpoints are only for ReaRev + SBERT, which I did manage to load and evaluate. Could you also share the checkpoints for ReaRev + LMSR and the instructions to load and evaluate them?