diff --git a/candle-core/src/metal_backend.rs b/candle-core/src/metal_backend.rs
index c1c4aa4b..5d72bd68 100644
--- a/candle-core/src/metal_backend.rs
+++ b/candle-core/src/metal_backend.rs
@@ -675,6 +675,7 @@ impl BackendStorage for MetalStorage {
                 ("uround", DType::F32) => contiguous::round::FLOAT,
                 ("urecip", DType::F32) => contiguous::recip::FLOAT,
                 ("utanh", DType::F32) => contiguous::tanh::FLOAT,
+                ("urelu", DType::F32) => contiguous::relu::FLOAT,
                 ("ucos", DType::F16) => contiguous::cos::HALF,
                 ("usin", DType::F16) => contiguous::sin::HALF,
                 ("usqr", DType::F16) => contiguous::sqr::HALF,
@@ -691,6 +692,7 @@ impl BackendStorage for MetalStorage {
                 ("uround", DType::F16) => contiguous::round::HALF,
                 ("urecip", DType::F16) => contiguous::recip::HALF,
                 ("utanh", DType::F16) => contiguous::tanh::HALF,
+                ("urelu", DType::F16) => contiguous::relu::HALF,
                 (name, dtype) => {
                     crate::bail!("Metal contiguous unary {name} {dtype:?} not implemented")
                 }
@@ -721,6 +723,7 @@ impl BackendStorage for MetalStorage {
                 ("uabs", DType::F32) => strided::abs::FLOAT,
                 ("uceil", DType::F32) => strided::ceil::FLOAT,
                 ("ufloor", DType::F32) => strided::floor::FLOAT,
+                ("urelu", DType::F32) => strided::relu::FLOAT,
                 ("uround", DType::F32) => strided::round::FLOAT,
                 ("ucos", DType::F16) => strided::cos::HALF,
                 ("usin", DType::F16) => strided::sin::HALF,
@@ -735,6 +738,7 @@ impl BackendStorage for MetalStorage {
                 ("uabs", DType::F16) => strided::abs::HALF,
                 ("uceil", DType::F16) => strided::ceil::HALF,
                 ("ufloor", DType::F16) => strided::floor::HALF,
+                ("urelu", DType::F16) => strided::relu::HALF,
                 ("uround", DType::F16) => strided::round::HALF,
                 (name, dtype) => {
                     crate::bail!("Metal strided unary {name} {dtype:?} not implemented")
diff --git a/candle-metal-kernels/src/lib.rs b/candle-metal-kernels/src/lib.rs
index 5d34f61a..c872dc60 100644
--- a/candle-metal-kernels/src/lib.rs
+++ b/candle-metal-kernels/src/lib.rs
@@ -174,8 +174,8 @@ macro_rules! ops{
 
 pub mod unary {
     ops!(
-        cos, sin, exp, sqr, sqrt, neg, log, gelu, abs, ceil, floor, round, erf, gelu_erf, tanh,
-        recip
+        cos, sin, exp, sqr, sqrt, neg, log, gelu, abs, ceil, floor, relu, round, erf, gelu_erf,
+        tanh, recip
     );
 }
 pub mod binary {
diff --git a/candle-metal-kernels/src/unary.metal b/candle-metal-kernels/src/unary.metal
index 7fbb613d..f95f6ba9 100644
--- a/candle-metal-kernels/src/unary.metal
+++ b/candle-metal-kernels/src/unary.metal
@@ -58,6 +58,12 @@ template <typename T> METAL_FUNC T gelu(T x) {
     T beta =  (static_cast<T>(M_2_SQRTPI_F * M_SQRT1_2_F) * alpha);
     return static_cast<T>(0.5) * x * (static_cast<T>(1.0) + T(tanh(beta)));
 }
+template <typename T> METAL_FUNC T relu(T in){
+    if (in < 0) {
+        return 0;
+    }
+    return in;
+}
 
 #define UNARY(FN, TYPENAME, FN_NAME, FN_NAME_STRIDED) \
 kernel void FN_NAME( \
@@ -110,6 +116,7 @@ UNARY_OP(gelu_erf)
 UNARY_OP(erf)
 UNARY_OP(tanh)
 UNARY_OP(recip)
+UNARY_OP(relu)
 
 UNARY(id, float, copy_f32, copy_f32_strided)
 UNARY(id, half, copy_f16, copy_f16_strided)
@@ -136,6 +143,7 @@ BFLOAT_UNARY_OP(gelu_erf)
 BFLOAT_UNARY_OP(erf)
 BFLOAT_UNARY_OP(tanh)
 BFLOAT_UNARY_OP(recip)
+BFLOAT_UNARY_OP(relu)
 
 UNARY(id, bfloat, copy_bf16, copy_bf16_strided)
 #endif