diff --git a/vaporetto/Cargo.toml b/vaporetto/Cargo.toml
index 816a219..08ebba4 100644
--- a/vaporetto/Cargo.toml
+++ b/vaporetto/Cargo.toml
@@ -13,7 +13,7 @@ keywords = ["japanese", "analyzer", "tokenizer", "morphological"]
 categories = ["text-processing", "no-std"]
 
 [dependencies]
-bincode = { version = "2.0.0-rc.2", default-features = false, features = ["alloc", "derive"] }  # MIT
+bincode = { version = "2.0.0-rc.3", default-features = false, features = ["alloc", "derive"] }  # MIT
 daachorse = "1.0.0"  # MIT or Apache-2.0
 hashbrown = "0.13.2"  # MIT or Apache-2.0
 
diff --git a/vaporetto/src/char_scorer/boundary_tag_scorer.rs b/vaporetto/src/char_scorer/boundary_tag_scorer.rs
index 362eb92..995caea 100644
--- a/vaporetto/src/char_scorer/boundary_tag_scorer.rs
+++ b/vaporetto/src/char_scorer/boundary_tag_scorer.rs
@@ -11,7 +11,6 @@ use bincode::{
 use daachorse::charwise::CharwiseDoubleArrayAhoCorasick;
 #[cfg(not(feature = "charwise-pma"))]
 use daachorse::DoubleArrayAhoCorasick;
-use hashbrown::HashMap;
 
 use crate::char_scorer::CharWeightMerger;
 use crate::dict_model::DictModel;
@@ -19,7 +18,7 @@ use crate::errors::{Result, VaporettoError};
 use crate::ngram_model::{NgramModel, TagNgramModel};
 use crate::predictor::{PositionalWeight, PositionalWeightWithTag, WeightVector};
 use crate::sentence::Sentence;
-use crate::utils::SplitMix64Builder;
+use crate::utils::{SerializableHashMap, SplitMix64Builder};
 
 pub struct CharScorerBoundaryTag {
     #[cfg(not(feature = "charwise-pma"))]
@@ -27,7 +26,7 @@ pub struct CharScorerBoundaryTag {
     #[cfg(feature = "charwise-pma")]
     pma: CharwiseDoubleArrayAhoCorasick<u32>,
     weights: Vec<Option<PositionalWeight<WeightVector>>>,
-    tag_weight: Vec<Vec<HashMap<u32, WeightVector, SplitMix64Builder>>>,
+    tag_weight: Vec<Vec<SerializableHashMap<u32, WeightVector, SplitMix64Builder>>>,
 }
 
 impl<'de> BorrowDecode<'de> for CharScorerBoundaryTag {
@@ -40,11 +39,7 @@ impl<'de> BorrowDecode<'de> for CharScorerBoundaryTag {
         #[cfg(feature = "charwise-pma")]
         let (pma, _) = unsafe { CharwiseDoubleArrayAhoCorasick::deserialize_unchecked(pma_data) };
         let weights = Decode::decode(decoder)?;
-        let tag_weight: Vec<Vec<Vec<(u32, WeightVector)>>> = Decode::decode(decoder)?;
-        let tag_weight = tag_weight
-            .into_iter()
-            .map(|x| x.into_iter().map(|x| x.into_iter().collect()).collect())
-            .collect();
+        let tag_weight = Decode::decode(decoder)?;
         Ok(Self {
             pma,
             weights,
@@ -58,12 +53,7 @@ impl Encode for CharScorerBoundaryTag {
         let pma_data = self.pma.serialize();
         Encode::encode(&pma_data, encoder)?;
         Encode::encode(&self.weights, encoder)?;
-        let tag_weight: Vec<Vec<Vec<_>>> = self
-            .tag_weight
-            .iter()
-            .map(|x| x.iter().map(|x| x.iter().collect()).collect())
-            .collect();
-        Encode::encode(&tag_weight, encoder)?;
+        Encode::encode(&self.tag_weight, encoder)?;
         Ok(())
     }
 }
@@ -90,11 +80,10 @@ impl CharScorerBoundaryTag {
             let weight = PositionalWeightWithTag::with_boundary(-word_len, d.weights);
             merger.add(d.word, weight);
         }
-        let mut tag_weight =
-            vec![
-                vec![HashMap::with_hasher(SplitMix64Builder); usize::from(window_size) + 1];
-                tag_ngram_model.len()
-            ];
+        let mut tag_weight = vec![
+            vec![SerializableHashMap::default(); usize::from(window_size) + 1];
+            tag_ngram_model.len()
+        ];
         for (i, tag_model) in tag_ngram_model.into_iter().enumerate() {
             for d in tag_model.0 {
                 for w in d.weights {
diff --git a/vaporetto/src/ngram_model.rs b/vaporetto/src/ngram_model.rs
index 6a00e34..1d141ec 100644
--- a/vaporetto/src/ngram_model.rs
+++ b/vaporetto/src/ngram_model.rs
@@ -9,7 +9,7 @@ pub struct NgramData<T> {
 }
 
 #[derive(Default, Debug, Decode, Encode)]
-pub struct NgramModel<T>(pub Vec<NgramData<T>>);
+pub struct NgramModel<T: 'static>(pub Vec<NgramData<T>>);
 
 #[derive(Clone, Debug, Decode, Encode)]
 pub struct TagWeight {
@@ -24,4 +24,4 @@ pub struct TagNgramData<T> {
 }
 
 #[derive(Default, Debug, Decode, Encode)]
-pub struct TagNgramModel<T>(pub Vec<TagNgramData<T>>);
+pub struct TagNgramModel<T: 'static>(pub Vec<TagNgramData<T>>);
diff --git a/vaporetto/src/predictor.rs b/vaporetto/src/predictor.rs
index 1f447cf..bb2fa66 100644
--- a/vaporetto/src/predictor.rs
+++ b/vaporetto/src/predictor.rs
@@ -44,6 +44,12 @@ pub enum WeightVector {
     Fixed(I32Simd),
 }
 
+impl Default for WeightVector {
+    fn default() -> Self {
+        Self::Variable(vec![])
+    }
+}
+
 impl Decode for WeightVector {
     fn decode<D: Decoder>(decoder: &mut D) -> Result<Self, DecodeError> {
         let weight: Vec<i32> = Decode::decode(decoder)?;
diff --git a/vaporetto/src/type_scorer/boundary_tag_scorer.rs b/vaporetto/src/type_scorer/boundary_tag_scorer.rs
index 4509c27..d0c3eb1 100644
--- a/vaporetto/src/type_scorer/boundary_tag_scorer.rs
+++ b/vaporetto/src/type_scorer/boundary_tag_scorer.rs
@@ -7,19 +7,18 @@ use bincode::{
     BorrowDecode, Decode, Encode,
 };
 use daachorse::DoubleArrayAhoCorasick;
-use hashbrown::HashMap;
 
 use crate::errors::{Result, VaporettoError};
 use crate::ngram_model::{NgramModel, TagNgramModel};
 use crate::predictor::{PositionalWeight, PositionalWeightWithTag, WeightVector};
 use crate::sentence::Sentence;
 use crate::type_scorer::TypeWeightMerger;
-use crate::utils::SplitMix64Builder;
+use crate::utils::{SerializableHashMap, SplitMix64Builder};
 
 pub struct TypeScorerBoundaryTag {
     pma: DoubleArrayAhoCorasick<u32>,
     weights: Vec<Option<PositionalWeight<WeightVector>>>,
-    tag_weight: Vec<Vec<HashMap<u32, WeightVector, SplitMix64Builder>>>,
+    tag_weight: Vec<Vec<SerializableHashMap<u32, WeightVector, SplitMix64Builder>>>,
 }
 
 impl<'de> BorrowDecode<'de> for TypeScorerBoundaryTag {
@@ -29,11 +28,7 @@ impl<'de> BorrowDecode<'de> for TypeScorerBoundaryTag {
         let pma_data: &[u8] = BorrowDecode::borrow_decode(decoder)?;
         let (pma, _) = unsafe { DoubleArrayAhoCorasick::deserialize_unchecked(pma_data) };
         let weights = Decode::decode(decoder)?;
-        let tag_weight: Vec<Vec<Vec<(u32, WeightVector)>>> = Decode::decode(decoder)?;
-        let tag_weight = tag_weight
-            .into_iter()
-            .map(|x| x.into_iter().map(|x| x.into_iter().collect()).collect())
-            .collect();
+        let tag_weight = Decode::decode(decoder)?;
         Ok(Self {
             pma,
             weights,
@@ -47,12 +42,7 @@ impl Encode for TypeScorerBoundaryTag {
         let pma_data = self.pma.serialize();
         Encode::encode(&pma_data, encoder)?;
         Encode::encode(&self.weights, encoder)?;
-        let tag_weight: Vec<Vec<Vec<_>>> = self
-            .tag_weight
-            .iter()
-            .map(|x| x.iter().map(|x| x.iter().collect()).collect())
-            .collect();
-        Encode::encode(&tag_weight, encoder)?;
+        Encode::encode(&self.tag_weight, encoder)?;
         Ok(())
     }
 }
@@ -68,11 +58,10 @@ impl TypeScorerBoundaryTag {
             let weight = PositionalWeightWithTag::with_boundary(-i16::from(window_size), d.weights);
             merger.add(d.ngram, weight);
         }
-        let mut tag_weight =
-            vec![
-                vec![HashMap::with_hasher(SplitMix64Builder); usize::from(window_size) + 1];
-                tag_ngram_model.len()
-            ];
+        let mut tag_weight = vec![
+            vec![SerializableHashMap::default(); usize::from(window_size) + 1];
+            tag_ngram_model.len()
+        ];
         for (i, tag_model) in tag_ngram_model.into_iter().enumerate() {
             for d in tag_model.0 {
                 for w in d.weights {
diff --git a/vaporetto/src/utils.rs b/vaporetto/src/utils.rs
index 2d55b5f..51e3f11 100644
--- a/vaporetto/src/utils.rs
+++ b/vaporetto/src/utils.rs
@@ -12,7 +12,7 @@ use bincode::{
     error::{DecodeError, EncodeError},
     Decode, Encode,
 };
-use hashbrown::HashMap;
+use hashbrown::{hash_map::DefaultHashBuilder, HashMap};
 
 #[cfg(feature = "fix-weight-length")]
 #[inline(always)]
@@ -35,42 +35,53 @@ impl Writer for VecWriter {
     }
 }
 
-#[derive(Debug)]
-pub struct SerializableHashMap<K, V>(pub HashMap<K, V>);
+#[derive(Clone, Debug, Default)]
+pub struct SerializableHashMap<K, V, S = DefaultHashBuilder>(pub HashMap<K, V, S>);
 
-impl<K, V> Deref for SerializableHashMap<K, V> {
-    type Target = HashMap<K, V>;
+impl<K, V, S> Deref for SerializableHashMap<K, V, S> {
+    type Target = HashMap<K, V, S>;
 
     fn deref(&self) -> &Self::Target {
         &self.0
     }
 }
 
-impl<K, V> DerefMut for SerializableHashMap<K, V> {
+impl<K, V, S> DerefMut for SerializableHashMap<K, V, S> {
     fn deref_mut(&mut self) -> &mut Self::Target {
         &mut self.0
     }
 }
 
-impl<K, V> Decode for SerializableHashMap<K, V>
+impl<K, V, S> Decode for SerializableHashMap<K, V, S>
 where
-    K: Encode + Decode + Eq + Hash,
-    V: Encode + Decode,
+    K: Decode + Eq + Hash,
+    V: Decode,
+    S: BuildHasher + Default,
 {
     fn decode<D: Decoder>(decoder: &mut D) -> Result<Self, DecodeError> {
-        let raw: Vec<(K, V)> = Decode::decode(decoder)?;
-        Ok(Self(raw.into_iter().collect()))
+        let mut result = HashMap::with_hasher(S::default());
+        let size: u64 = Decode::decode(decoder)?;
+        for _ in 0..size {
+            let k = Decode::decode(decoder)?;
+            let v = Decode::decode(decoder)?;
+            result.insert(k, v);
+        }
+        Ok(Self(result))
     }
 }
 
-impl<K, V> Encode for SerializableHashMap<K, V>
+impl<K, V, S> Encode for SerializableHashMap<K, V, S>
 where
-    K: Encode + Decode,
-    V: Encode + Decode,
+    K: Encode,
+    V: Encode,
 {
     fn encode<E: Encoder>(&self, encoder: &mut E) -> Result<(), EncodeError> {
-        let raw: Vec<(&K, &V)> = self.0.iter().collect();
-        Encode::encode(&raw, encoder)?;
+        let size = u64::try_from(self.0.len()).unwrap();
+        Encode::encode(&size, encoder)?;
+        for (k, v) in &self.0 {
+            Encode::encode(k, encoder)?;
+            Encode::encode(v, encoder)?;
+        }
         Ok(())
     }
 }