could you supply these two 5x64 models refered by the paper
l1t1 opened this issue · comments
l1t1 commented
5.3. AlphaGo Zero vs. AlphaZero
We hypothesize that AZ training vastly outperforms AGZ
training (given equivalent hardware) due to the former’s
asynchrony and consequent improved throughput. We train
64-filter, 5-block models from scratch with both AGZ and
AZ for 12 hours and compare their final performance. The
model trained with AZ wins 100:0 over the model trained
with AGZ,
Junyan Xu commented
They're not trained on games from a higher-block model, so I don't expect them to be very strong...