From 407998267c8a8796d116cc363290a2d634ae322f Mon Sep 17 00:00:00 2001 From: Guanbo Wang Date: Mon, 11 Apr 2022 18:12:54 -0400 Subject: [PATCH] Remove lazy_load option --- egs/gigaspeech/ASR/conformer_ctc/asr_datamodule.py | 11 +---------- 1 file changed, 1 insertion(+), 10 deletions(-) diff --git a/egs/gigaspeech/ASR/conformer_ctc/asr_datamodule.py b/egs/gigaspeech/ASR/conformer_ctc/asr_datamodule.py index 9fc3d1328..ab958fa68 100644 --- a/egs/gigaspeech/ASR/conformer_ctc/asr_datamodule.py +++ b/egs/gigaspeech/ASR/conformer_ctc/asr_datamodule.py @@ -180,12 +180,6 @@ class GigaSpeechAsrDataModule: default="XL", help="Select the GigaSpeech subset (XS|S|M|L|XL)", ) - group.add_argument( - "--lazy-load", - type=str2bool, - default=True, - help="lazily open CutSets to avoid OOM (for L|XL subset)", - ) group.add_argument( "--small-dev", type=str2bool, @@ -361,10 +355,7 @@ class GigaSpeechAsrDataModule: def train_cuts(self) -> CutSet: logging.info(f"About to get train_{self.args.subset} cuts") path = self.args.manifest_dir / f"cuts_{self.args.subset}.jsonl.gz" - if self.args.subset in ["L", "XL"] and self.args.lazy_load: - cuts_train = CutSet.from_jsonl_lazy(path) - else: - cuts_train = CutSet.from_file(path) + cuts_train = CutSet.from_jsonl_lazy(path) return cuts_train @lru_cache()