diff --git a/tokenizers/src/models/bpe/serialization.rs b/tokenizers/src/models/bpe/serialization.rs
index 8f60d6ddb..ca2590f1d 100644
--- a/tokenizers/src/models/bpe/serialization.rs
+++ b/tokenizers/src/models/bpe/serialization.rs
@@ -21,6 +21,7 @@ impl Serialize for BPE {
         model.serialize_field("end_of_word_suffix", &self.end_of_word_suffix)?;
         model.serialize_field("fuse_unk", &self.fuse_unk)?;
         model.serialize_field("byte_fallback", &self.byte_fallback)?;
+        model.serialize_field("ignore_merges", &self.ignore_merges)?;
 
         // Then the large ones
         let mut merges: Vec<(&Pair, &u32)> = self
@@ -57,6 +58,7 @@ impl<'de> Deserialize<'de> for BPE {
                 "end_of_word_suffix",
                 "fuse_unk",
                 "byte_fallback",
+                "ignore_merges",
                 "vocab",
                 "merges",
             ],
@@ -112,6 +114,11 @@ impl<'de> Visitor<'de> for BPEVisitor {
                         builder = builder.byte_fallback(suffix);
                     }
                 }
+                "ignore_merges" => {
+                    if let Some(suffix) = map.next_value()? {
+                        builder = builder.ignore_merges(suffix);
+                    }
+                }
                 "vocab" => vocab = Some(map.next_value()?),
                 "merges" => merges = Some(map.next_value()?),
                 "type" => match map.next_value()? {
@@ -136,3 +143,49 @@ impl<'de> Visitor<'de> for BPEVisitor {
         }
     }
 }
+
+#[cfg(test)]
+mod test {
+    use super::*;
+    use crate::models::bpe::Vocab;
+
+    #[test]
+    fn test_serialization() {
+        let vocab: Vocab = [("<unk>".into(), 0), ("a".into(), 1), ("b".into(), 2)]
+            .iter()
+            .cloned()
+            .collect();
+        let bpe = BpeBuilder::default()
+            .vocab_and_merges(vocab, vec![])
+            .unk_token("<unk>".to_string())
+            .ignore_merges(true)
+            .build()
+            .unwrap();
+
+        let data = serde_json::to_string(&bpe).unwrap();
+        let reconstructed = serde_json::from_str(&data).unwrap();
+
+        assert_eq!(bpe, reconstructed);
+    }
+
+    #[test]
+    fn test_serialization_ignore_merges() {
+        let vocab: Vocab = [("<unk>".into(), 0), ("a".into(), 1), ("b".into(), 2)]
+            .iter()
+            .cloned()
+            .collect();
+        let mut bpe = BpeBuilder::default()
+            .vocab_and_merges(vocab, vec![])
+            .unk_token("<unk>".to_string())
+            .ignore_merges(true)
+            .build()
+            .unwrap();
+
+        let bpe_string = r#"{"type":"BPE","dropout":null,"unk_token":"<unk>","continuing_subword_prefix":null,"end_of_word_suffix":null,"fuse_unk":false,"byte_fallback":false,"ignore_merges":true,"vocab":{"<unk>":0,"a":1,"b":2},"merges":[]}"#;
+        assert_eq!(serde_json::from_str::<BPE>(bpe_string).unwrap(), bpe);
+
+        bpe.ignore_merges = false;
+        let bpe_string = r#"{"type":"BPE","dropout":null,"unk_token":"<unk>","continuing_subword_prefix":null,"end_of_word_suffix":null,"fuse_unk":false,"byte_fallback":false,"vocab":{"<unk>":0,"a":1,"b":2},"merges":[]}"#;
+        assert_eq!(serde_json::from_str::<BPE>(bpe_string).unwrap(), bpe);
+    }
+}