From 75c8e57823d74a6b6baf9e2fdf7910ffa96a1af8 Mon Sep 17 00:00:00 2001 From: Ilya Shutov Date: Sun, 5 Jul 2020 02:07:57 +0700 Subject: [PATCH] README.md typo fix --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index bd87d6d075..5219ea9477 100644 --- a/README.md +++ b/README.md @@ -117,7 +117,7 @@ mpiexec -n {ngpus} python jukebox/train.py --hps=small_vqvae,small_upsampler,all --restore_vqvae=logs/small_vqvae/checkpoint_latest.pth.tar --prior --levels=2 --level=0 --weight_decay=0.01 --save_iters=1000 ``` We pass `sample_length = n_ctx * downsample_of_level` so that after downsampling the tokens match the n_ctx of the prior hps. -Here, `n_ctx = 8192` and `downsamples = (32, 256)`, giving `sample_lengths = (8192 * 32, 8192 * 256) = (65536, 2097152)` respectively for the bottom and top level. +Here, `n_ctx = 8192` and `downsamples = (32, 256)`, giving `sample_lengths = (8192 * 32, 8192 * 256) = (262144, 2097152)` respectively for the bottom and top level. ### Learning rate annealing To get the best sample quality anneal the learning rate to 0 near the end of training. To do so, continue training from the latest