diff --git a/Cargo.lock b/Cargo.lock
index 8ba5c2d..263317a 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -1077,7 +1077,7 @@ dependencies = [
 
 [[package]]
 name = "fsrs"
-version = "1.2.2"
+version = "1.2.3"
 dependencies = [
  "burn",
  "chrono",
diff --git a/Cargo.toml b/Cargo.toml
index 393f6e2..d193930 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "fsrs"
-version = "1.2.2"
+version = "1.2.3"
 authors = ["Open Spaced Repetition"]
 categories = ["algorithms", "science"]
 edition = "2021"
diff --git a/src/batch_shuffle.rs b/src/batch_shuffle.rs
index 389ae68..52cbbf9 100644
--- a/src/batch_shuffle.rs
+++ b/src/batch_shuffle.rs
@@ -263,12 +263,41 @@ where
 
 #[cfg(test)]
 mod tests {
+    use burn::backend::{ndarray::NdArrayDevice, NdArray};
+
     use super::*;
-    use crate::{convertor_tests::anki21_sample_file_converted_to_fsrs, FSRSItem, FSRSReview};
+    use crate::{
+        convertor_tests::anki21_sample_file_converted_to_fsrs,
+        dataset::{prepare_training_data, FSRSBatcher, FSRSDataset},
+        FSRSItem, FSRSReview,
+    };
+
+    #[test]
+    fn batch_shuffle_dataloader() {
+        let train_set = anki21_sample_file_converted_to_fsrs();
+        let (_pre_train_set, train_set) = prepare_training_data(train_set);
+        let dataset = FSRSDataset::from(train_set);
+        let batch_size = 512;
+        let seed = 42;
+        let device = NdArrayDevice::Cpu;
+        type Backend = NdArray<f32>;
+        let batcher = FSRSBatcher::<Backend>::new(device);
+        let dataloader =
+            BatchShuffledDataLoaderBuilder::new(batcher).build(dataset, batch_size, seed);
+        let item = dataloader.iter().next().unwrap();
+        assert_eq!(
+            item.t_historys.shape(),
+            burn::tensor::Shape { dims: [6, 512] }
+        );
+        let item2 = dataloader.iter().next().unwrap();
+        assert_eq!(
+            item2.t_historys.shape(),
+            burn::tensor::Shape { dims: [4, 512] }
+        );
+    }
 
     #[test]
     fn batch_shuffle() {
-        use crate::dataset::FSRSDataset;
         let dataset = Arc::new(FSRSDataset::from(anki21_sample_file_converted_to_fsrs()));
         let batch_size = 10;
         let seed = 42;
@@ -484,7 +513,6 @@ mod tests {
 
     #[test]
     fn item_shuffle() {
-        use crate::dataset::FSRSDataset;
         use burn::data::dataset::transform::ShuffledDataset;
         let dataset = FSRSDataset::from(anki21_sample_file_converted_to_fsrs());
         let seed = 42;
diff --git a/src/dataset.rs b/src/dataset.rs
index 41d863f..08c49d3 100644
--- a/src/dataset.rs
+++ b/src/dataset.rs
@@ -6,6 +6,7 @@ use burn::{
     tensor::{backend::Backend, Data, ElementConversion, Float, Int, Shape, Tensor},
 };
 
+use itertools::Itertools;
 use serde::{Deserialize, Serialize};
 
 /// Stores a list of reviews for a card, in chronological order. Each FSRSItem corresponds
@@ -190,7 +191,7 @@ pub fn filter_outlier(
     let mut filtered_items = vec![];
     let mut removed_pairs: [HashSet<_>; 5] = Default::default();
 
-    for (rating, delta_t_groups) in groups.into_iter() {
+    for (rating, delta_t_groups) in groups.into_iter().sorted_by_key(|&(k, _)| k) {
         let mut sub_groups = delta_t_groups.into_iter().collect::<Vec<_>>();
 
         // order by size of sub group ascending and delta_t descending