diff --git a/kompute-shaders/op_rope_f16.comp b/kompute-shaders/op_rope_f16.comp
index 3abe3ed33..b44622584 100644
--- a/kompute-shaders/op_rope_f16.comp
+++ b/kompute-shaders/op_rope_f16.comp
@@ -40,26 +40,34 @@ void main() {
         }
     } else {
         const float inv_ndims = -1.f/pcs.n_dims;
-        for (uint ib = 0; ib < pcs.ne0/pcs.n_dims; ++ib) {
-            for (uint ic = 0; ic < pcs.n_dims; ic += 2) {
-                const uint cur_rot = ib * pcs.n_dims + ic;
+        for (uint ic = 0; ic < pcs.n_dims; ic += 2) {
+            const uint cur_rot = ic;
 
-                float cos_theta, sin_theta;
-                rope_yarn(theta, pcs.freq_scale, corr_dims, cur_rot, pcs.ext_factor, pcs.attn_factor, cos_theta, sin_theta);
+            float cos_theta, sin_theta;
+            rope_yarn(theta, pcs.freq_scale, corr_dims, cur_rot, pcs.ext_factor, pcs.attn_factor, cos_theta, sin_theta);
 
-                theta *= theta_scale;
+            theta *= theta_scale;
 
-                const uint i0 = ib*pcs.n_dims + ic/2;
+            const uint i0 = ic/2;
 
-                const uint src      = uint((i3*pcs.nb03 + i2*pcs.nb02 + i1*pcs.nb01 + i0*pcs.nb00) / 2) + pcs.inAOff; // Based from in
-                const uint dst_data = uint((i3*pcs.nb3  + i2*pcs.nb2  + i1*pcs.nb1  + i0*pcs.nb0)  / 2) + pcs.outOff; // Based from out_
+            const uint src      = uint((i3*pcs.nb03 + i2*pcs.nb02 + i1*pcs.nb01 + i0*pcs.nb00) / 2) + pcs.inAOff; // Based from in
+            const uint dst_data = uint((i3*pcs.nb3  + i2*pcs.nb2  + i1*pcs.nb1  + i0*pcs.nb0)  / 2) + pcs.outOff; // Based from out_
 
-                const float x0 = float(inA[src]);
-                const float x1 = float(inA[src+pcs.n_dims/2]);
+            const float x0 = float(inA[src]);
+            const float x1 = float(inA[src+pcs.n_dims/2]);
 
-                out_[dst_data]              = float16_t(x0*cos_theta - x1*sin_theta);
-                out_[dst_data+pcs.n_dims/2] = float16_t(x0*sin_theta + x1*cos_theta);
-            }
+            out_[dst_data]              = float16_t(x0*cos_theta - x1*sin_theta);
+            out_[dst_data+pcs.n_dims/2] = float16_t(x0*sin_theta + x1*cos_theta);
+        }
+
+        for (uint ic = pcs.n_dims; ic < pcs.ne0; ic += 2) {
+            const uint i0 = ic;
+
+            const uint src      = uint((i3*pcs.nb03 + i2*pcs.nb02 + i1*pcs.nb01 + i0*pcs.nb00) / 2) + pcs.inAOff; // Based from in
+            const uint dst_data = uint((i3*pcs.nb3  + i2*pcs.nb2  + i1*pcs.nb1  + i0*pcs.nb0)  / 2) + pcs.outOff; // Based from out_
+
+            out_[dst_data + 0] = inA[src + 0];
+            out_[dst_data + 1] = inA[src + 1];
         }
     }
 }
diff --git a/kompute-shaders/op_rope_f32.comp b/kompute-shaders/op_rope_f32.comp
index 2adf5eb4e..2c0235d75 100644
--- a/kompute-shaders/op_rope_f32.comp
+++ b/kompute-shaders/op_rope_f32.comp
@@ -29,8 +29,8 @@ void main() {
 
             theta *= theta_scale;
 
-            const uint src = uint((i3*pcs.nb03 + i2*pcs.nb02 + i1*pcs.nb01 + i0*pcs.nb00) / 4) + pcs.inAOff; // Based from in
-            const uint dst_data = uint((i3*pcs.nb3  + i2*pcs.nb2  + i1*pcs.nb1  + i0*pcs.nb0) / 4) + pcs.outOff; // Based from out_
+            const uint src      = uint((i3*pcs.nb03 + i2*pcs.nb02 + i1*pcs.nb01 + i0*pcs.nb00) / 4) + pcs.inAOff; // Based from in
+            const uint dst_data = uint((i3*pcs.nb3  + i2*pcs.nb2  + i1*pcs.nb1  + i0*pcs.nb0)  / 4) + pcs.outOff; // Based from out_
 
             const float x0 = inA[src];
             const float x1 = inA[src+1];
@@ -50,8 +50,8 @@ void main() {
 
             const uint i0 = ic/2;
 
-            const uint src = uint((i3*pcs.nb03 + i2*pcs.nb02 + i1*pcs.nb01 + i0*pcs.nb00) / 4) + pcs.inAOff; // Based from in
-            const uint dst_data = uint((i3*pcs.nb3  + i2*pcs.nb2  + i1*pcs.nb1  + i0*pcs.nb0) / 4) + pcs.outOff; // Based from out_
+            const uint src      = uint((i3*pcs.nb03 + i2*pcs.nb02 + i1*pcs.nb01 + i0*pcs.nb00) / 4) + pcs.inAOff; // Based from in
+            const uint dst_data = uint((i3*pcs.nb3  + i2*pcs.nb2  + i1*pcs.nb1  + i0*pcs.nb0)  / 4) + pcs.outOff; // Based from out_
 
             const float x0 = inA[src];
             const float x1 = inA[src+pcs.n_dims/2];