Closed nmiculinic closed 6 years ago
Closing since this one has a lot going on
yes, it should change to 100, thanks.
On Mon, Apr 30, 2018, 5:40 PM Neven Miculinic notifications@github.com wrote:
@nmiculinic commented on this pull request.
In chiron/chiron_eval.py https://github.com/haotianteng/Chiron/pull/50#discussion_r184925251:
logits_queue = tf.FIFOQueue(
- capacity=FLAGS.batch_size * 100,
- dtypes=[tf.float32, tf.int32, tf.int32],
- shapes = [logits.shape,logits_index.shape,seq_length.shape]
- capacity=100 * FLAGS.batch_size,
Actually, this should be waaaay smaller. I'd say 16 is okish. Each queue element is whole batch, thus multiplying by batch size is wrong thing to do
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/haotianteng/Chiron/pull/50#pullrequestreview-116213388, or mute the thread https://github.com/notifications/unsubscribe-auth/AKo3X5d-sk-43QROF4WeUZY9amdhCLC6ks5ttsAJgaJpZM4Tr8Gm .
This PR merges following PR and fixes merge conflicts:
48
49
47