diff --git a/rlmolecule/actors.py b/rlmolecule/actors.py
index 55cfd8c..abc1126 100644
--- a/rlmolecule/actors.py
+++ b/rlmolecule/actors.py
@@ -4,7 +4,7 @@
 from typing import Any, List
 
 import ray
-from lru import LRU
+from pylru import lrucache
 
 logger = logging.getLogger(__name__)
 
@@ -32,7 +32,7 @@ class RayDictCache(DictCache):
 @ray.remote
 class RayLRUCache(DictCache):
     def __init__(self, max_size: int = int(1e5)):
-        self._dict = LRU(max_size)
+        self._dict = lrucache(max_size)
 
 
 @ray.remote
diff --git a/rlmolecule/builder.py b/rlmolecule/builder.py
index 740b615..b9491cb 100644
--- a/rlmolecule/builder.py
+++ b/rlmolecule/builder.py
@@ -53,7 +53,7 @@ def __init__(
 
         Args:
             max_atoms (int, optional): Maximum number of heavy atoms. Defaults to 10.
-            min_atoms (int, optional): minimum number of heavy atoms. Defaults to 4.
+            min_atoms (int, optional): Minimum number of heavy atoms. Defaults to 4.
             atom_additions (Optional[List], optional): potential atom types to consider.
                 Defaults to ('C', 'N', 'O')
             stereoisomers (bool, optional): whether to consider stereoisomers different
diff --git a/rlmolecule/molecule_state.py b/rlmolecule/molecule_state.py
index 6e80ebe..a4c1e46 100644
--- a/rlmolecule/molecule_state.py
+++ b/rlmolecule/molecule_state.py
@@ -118,18 +118,24 @@ def _get_children(self) -> Sequence[V]:
             return []
 
         next_actions = [self.new(molecule) for molecule in self.builder(self.molecule)]
-        next_actions.extend(self._get_terminal_actions())
+            
+        # Only add the terminal state as an action if there are enough atoms
+        terminal_actions = [] 
+        if self.num_atoms >= self.min_num_atoms:
+            terminal_actions = self._get_terminal_actions()
 
-        if self.data.prune_terminal_states:
-            next_actions = self._prune_next_actions(next_actions)
+            if self.data.prune_terminal_states:
+                terminal_actions = self._prune_next_actions(terminal_actions)
 
-        if len(next_actions) >= self.max_num_actions:
+        if len(next_actions) + len(terminal_actions) >= self.max_num_actions:
             logger.info(
-                f"{self} has {len(next_actions) + 1} next actions when the "
+                f"{self} has {len(next_actions) + 1 + len(terminal_actions)} next actions when the "
                 f"maximum is {self.max_num_actions}"
             )
-            next_actions = random.sample(next_actions, self.max_num_actions)
+            next_actions = random.sample(next_actions, self.max_num_actions - len(terminal_actions))
 
+        next_actions.extend(terminal_actions)
+            
         return next_actions
 
     def _get_terminal_actions(self) -> Sequence[V]:
@@ -145,6 +151,7 @@ def _get_terminal_actions(self) -> Sequence[V]:
     def _prune_next_actions(self, next_actions: Sequence[V]) -> Sequence[V]:
         """Use the ray actor handle in self.data (or a simple set) to find terminal
         states that have already been evaluated and remove them from the search tree.
+        Useful to force the builder to continually make new molecules.
 
         Args:
             next_actions (Sequence[V]): A list of MoleculeStates to be pruned
@@ -242,6 +249,10 @@ def num_atoms(self) -> int:
     def max_num_actions(self) -> int:
         return self.data.max_num_actions
 
+    @property
+    def min_num_atoms(self) -> int:
+        return self.builder.min_atoms
+
     def __repr__(self) -> str:
         """
         delegates to the SMILES string
diff --git a/setup.cfg b/setup.cfg
index 8969cb5..0893ed8 100644
--- a/setup.cfg
+++ b/setup.cfg
@@ -15,7 +15,7 @@ include_package_data = true
 install_requires =
   graphenv
   tensorflow
-  lru-dict
+  pylru
   rdkit
   nfp
   
diff --git a/tests/conftest.py b/tests/conftest.py
index 0a6fd94..47ba9c5 100644
--- a/tests/conftest.py
+++ b/tests/conftest.py
@@ -23,7 +23,7 @@ def ray_init():
 
 @pytest.fixture
 def builder() -> MoleculeBuilder:
-    return MoleculeBuilder(max_atoms=5)
+    return MoleculeBuilder(max_atoms=5, min_atoms=1)
 
 
 @pytest.fixture(scope="class")
diff --git a/tests/test_molecule_state.py b/tests/test_molecule_state.py
index a50a036..d53f64f 100644
--- a/tests/test_molecule_state.py
+++ b/tests/test_molecule_state.py
@@ -61,7 +61,7 @@ def test_prune_terminal_ray(ray_init):
     qed_root = QEDState(
         rdkit.Chem.MolFromSmiles("C"),
         MoleculeData(
-            MoleculeBuilder(max_atoms=5, cache=True),
+            MoleculeBuilder(max_atoms=5, min_atoms=1, cache=True),
             max_num_actions=20,
             prune_terminal_states=True,
         ),
@@ -95,7 +95,7 @@ def test_csv_writer(ray_init, caplog):
     with TemporaryDirectory() as tempdir:
 
         data = MoleculeData(
-            MoleculeBuilder(max_atoms=5, cache=True),
+            MoleculeBuilder(max_atoms=5, min_atoms=1, cache=True),
             max_num_actions=20,
             prune_terminal_states=True,
             log_reward_filepath=Path(tempdir, "test.csv"),