step 8_520_000 . Checkpoint from initial model taken and trained further at a lower learning rate

820

820

828

824

822

822

second

Initial commit