diff --git a/crates/cubecl-core/src/frontend/container/registry/base.rs b/crates/cubecl-core/src/frontend/container/registry/base.rs
index 63a798a79..889b9ae4f 100644
--- a/crates/cubecl-core/src/frontend/container/registry/base.rs
+++ b/crates/cubecl-core/src/frontend/container/registry/base.rs
@@ -29,7 +29,7 @@ impl From<ExpandElementTyped<u32>> for u32 {
     }
 }
 
-impl<K: PartialOrd + Ord, V: CubeType + Clone> Registry<K, V> {
+impl<K: PartialOrd + Ord + core::fmt::Debug, V: CubeType + Clone> Registry<K, V> {
     /// Create a new registry.
     pub fn new() -> Self {
         Self::default()
@@ -51,7 +51,10 @@ impl<K: PartialOrd + Ord, V: CubeType + Clone> Registry<K, V> {
         let key = query.into();
         let map = self.map.as_ref().borrow();
 
-        map.get(&key).unwrap().clone()
+        match map.get(&key) {
+            Some(val) => val.clone(),
+            None => panic!("No value found for key {key:?}"),
+        }
     }
 
     /// Insert an item in the registry.
@@ -88,12 +91,15 @@ impl<K: PartialOrd + Ord, V: CubeType + Clone> Registry<K, V> {
     }
 }
 
-impl<K: PartialOrd + Ord, V: Clone> Registry<K, V> {
+impl<K: PartialOrd + Ord + core::fmt::Debug, V: Clone> Registry<K, V> {
     /// Expand method of [Self::find].
     pub fn __expand_find_method(&self, _scope: &mut Scope, key: K) -> V {
         let map = self.map.as_ref().borrow();
 
-        map.get(&key).unwrap().clone()
+        match map.get(&key) {
+            Some(val) => val.clone(),
+            None => panic!("No value found for key {key:?}"),
+        }
     }
 
     /// Expand method of [Self::insert].
diff --git a/crates/cubecl-core/src/frontend/container/sequence/base.rs b/crates/cubecl-core/src/frontend/container/sequence/base.rs
index fa41549e1..06fbf147b 100644
--- a/crates/cubecl-core/src/frontend/container/sequence/base.rs
+++ b/crates/cubecl-core/src/frontend/container/sequence/base.rs
@@ -13,7 +13,7 @@ use std::{cell::RefCell, rc::Rc};
 /// All methods [push](Sequence::push), [index](Sequence::index) and
 /// [into_iter](Sequence::into_iter) are executed _during_ compilation and don't add any overhead
 /// on the generated kernel.
-#[derive(Debug, Clone, Hash, PartialEq, Eq, Serialize, Deserialize)]
+#[derive(Debug, Clone, Hash, PartialEq, Eq, Serialize, Deserialize, PartialOrd, Ord)]
 pub struct Sequence<T: CubeType> {
     values: Vec<T>,
 }
@@ -30,6 +30,14 @@ impl<T: CubeType> Init for Sequence<T> {
     }
 }
 
+impl<T: CubeType + Clone> Sequence<T> {
+    pub fn rev(&self) -> Self {
+        Self {
+            values: self.values.iter().rev().cloned().collect(),
+        }
+    }
+}
+
 impl<T: CubeType> Sequence<T> {
     /// Create a new empty sequence.
     pub fn new() -> Self {
@@ -170,6 +178,10 @@ impl<T: CubeType> CubeType for Sequence<T> {
 }
 
 impl<T: CubeType> SequenceExpand<T> {
+    #[allow(clippy::len_without_is_empty)]
+    pub fn len(&self) -> u32 {
+        self.values.borrow().len() as u32
+    }
     /// Expand method of [push](Sequence::push).
     pub fn __expand_push_method(&mut self, _scope: &mut Scope, value: T::ExpandType) {
         self.values.borrow_mut().push(value);
diff --git a/crates/cubecl-core/src/frontend/element/uint.rs b/crates/cubecl-core/src/frontend/element/uint.rs
index 8243025ec..e716cd3d2 100644
--- a/crates/cubecl-core/src/frontend/element/uint.rs
+++ b/crates/cubecl-core/src/frontend/element/uint.rs
@@ -28,6 +28,12 @@ macro_rules! declare_uint {
             }
         }
 
+        impl Init for $primitive {
+            fn init(self, _scope: &mut Scope) -> Self {
+                self
+            }
+        }
+
         impl IntoRuntime for $primitive {
             fn __expand_runtime_method(self, scope: &mut Scope) -> ExpandElementTyped<Self> {
                 let expand: ExpandElementTyped<Self> = self.into();
diff --git a/crates/cubecl-linalg/src/matmul/components/batch/one_to_one.rs b/crates/cubecl-linalg/src/matmul/components/batch/one_to_one.rs
index e3ba06ae6..221d7f62d 100644
--- a/crates/cubecl-linalg/src/matmul/components/batch/one_to_one.rs
+++ b/crates/cubecl-linalg/src/matmul/components/batch/one_to_one.rs
@@ -108,6 +108,7 @@ impl<MP: MatmulPrecision, GMM: GlobalMatmul<MP>, C: CubeDispatch> BatchMatmul<MP
         let k_range = (0, lhs.shape(rank - 1));
 
         let gmm_config = config.to_gmm_config();
+
         gmm_execute::<MP, GMM>(
             lhs,
             rhs,