From ded881802c82190681a1f37caf1edefb16e67491 Mon Sep 17 00:00:00 2001
From: jmwang66 <wangjiaming.wjm@alibaba-inc.com>
Date: 星期二, 07 二月 2023 10:17:52 +0800
Subject: [PATCH] update data2vec pretrain
---
funasr/datasets/large_datasets/dataset.py | 7 ++++---
funasr/datasets/large_datasets/datapipes/batch.py | 1 +
2 files changed, 5 insertions(+), 3 deletions(-)
diff --git a/funasr/datasets/large_datasets/datapipes/batch.py b/funasr/datasets/large_datasets/datapipes/batch.py
index c980ae3..8ec43e9 100644
--- a/funasr/datasets/large_datasets/datapipes/batch.py
+++ b/funasr/datasets/large_datasets/datapipes/batch.py
@@ -46,6 +46,7 @@
batch = []
bucket = []
max_lengths = 0
+ min_lengths = 999999
batch_lengths = 0
if self.batch_mode == "clipping":
diff --git a/funasr/datasets/large_datasets/dataset.py b/funasr/datasets/large_datasets/dataset.py
index 2d3ffd5..55b0678 100644
--- a/funasr/datasets/large_datasets/dataset.py
+++ b/funasr/datasets/large_datasets/dataset.py
@@ -158,9 +158,10 @@
filter_fn = partial(filter, **filter_conf)
dataset = FilterIterDataPipe(dataset, fn=filter_fn)
- vocab = {'vocab': dict, 'seg_dict': seg_dict}
- tokenize_fn = partial(tokenize, **vocab)
- dataset = MapperIterDataPipe(dataset, fn=tokenize_fn)
+ if "text" in data_names:
+ vocab = {'vocab': dict, 'seg_dict': seg_dict}
+ tokenize_fn = partial(tokenize, **vocab)
+ dataset = MapperIterDataPipe(dataset, fn=tokenize_fn)
if shuffle:
buffer_conf = conf.get('shuffle_conf', {})
--
Gitblit v1.9.1