From a7cf41d3acc95c6b7c9233bc9b97e36d6ce55cde Mon Sep 17 00:00:00 2001 From: Fangjun Kuang Date: Wed, 27 Oct 2021 16:59:47 +0800 Subject: [PATCH] Add a note about the CUDA OOM error. Some users consider this kind of OOM as an error during decoding, but actually it is not. This pull request clarifies that. --- icefall/decode.py | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/icefall/decode.py b/icefall/decode.py index 62d27dd68..d799e9b24 100644 --- a/icefall/decode.py +++ b/icefall/decode.py @@ -732,6 +732,11 @@ def rescore_with_whole_lattice( logging.info( f"num_arcs before pruning: {inv_lattice.arcs.num_elements()}" ) + logging.info("This OOM is not an error. You can ignore it. " + "If your model does not converge well, or --max-duration " + "is too large, or the input sound file is difficult to " + "decode, you will meet this exception." + ) # NOTE(fangjun): The choice of the threshold 1e-9 is arbitrary here # to avoid OOM. You may need to fine tune it.