diff --git a/src/lib.rs b/src/lib.rs
index 0203acf..047022d 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -1,7 +1,6 @@
 // This check is new and seems buggy (possibly with PyO3 interaction)
 #![allow(clippy::borrow_deref_ref)]
 
-use std::collections::HashSet;
 use std::num::NonZeroU64;
 use std::thread;
 
@@ -12,6 +11,7 @@ use pyo3::pybacked::PyBackedStr;
 use pyo3::types::{PyBytes, PyList, PyTuple};
 use pyo3::PyResult;
 use rustc_hash::FxHashMap as HashMap;
+use rustc_hash::FxHashSet as HashSet;
 
 type Rank = u32;
 
@@ -317,7 +317,7 @@ impl CoreBPE {
         if last_piece_token_len == 0 {
             // If last_piece_token_len is zero, the last token was a special token and we have
             // no unstable bytes
-            return (tokens, HashSet::new());
+            return (tokens, HashSet::default());
         }
         let (mut tokens, last_piece_token_len) =
             self._increase_last_piece_token_len(tokens, last_piece_token_len);
@@ -331,7 +331,7 @@ impl CoreBPE {
         // This would reduce the amount of retokenising when determining completions
         // Refer to the logic in an older version of this file
 
-        let mut completions = HashSet::new();
+        let mut completions = HashSet::default();
         if unstable_bytes.is_empty() {
             return (tokens, completions);
         }
@@ -514,7 +514,7 @@ impl CoreBPE {
                 Ok(text) => self._encode_ordinary_native(text),
                 Err(e) => {
                     let text = unsafe { std::str::from_utf8_unchecked(&bytes[..e.valid_up_to()]) };
-                    let (tokens, last_piece_token_len) = self._encode_native(text, &HashSet::new());
+                    let (tokens, last_piece_token_len) = self._encode_native(text, &HashSet::default());
                     let (mut tokens, last_piece_token_len) =
                         self._increase_last_piece_token_len(tokens, last_piece_token_len);
                     if !tokens.is_empty() && last_piece_token_len > 0 {