Strong baseline for visual question answering
This is the log file containing the weights for a model that has been trained for 50 epochs (1890 iterations each) on the training set of VQA v1. To load a model with these weights, you can run:
import torch
import model
log = torch.load('logs/2017-08-04_00:55:19.pth')
tokens = len(log['vocab']['question']) + 1
net = torch.nn.DataParallel(model.Net(tokens))
net.load_state_dict(log['weights'])