FiD model trained on WebQ

-- This is the model checkpoint of FiD [2], based on the T5 (with 3B parameters) and trained on the WebQ dataset [1].

-- Hyperparameters: 8 x 40GB A100 GPUs; batch size 8; AdamW; LR 3e-5; 30000 steps

References:

[1] Semantic parsing on freebase from question-answer pairs. EMNLP 2013.

[2] Leveraging Passage Retrieval with Generative Models for Open Domain Question Answering. EACL 2021.

Model performance

We evaluate it on the WebQ dataset, the EM score is 51.4 on the test set.