diff --git a/ggml/src/ggml-cuda/solve_tri.cu b/ggml/src/ggml-cuda/solve_tri.cu
index 726b08b99..a955d535c 100644
--- a/ggml/src/ggml-cuda/solve_tri.cu
+++ b/ggml/src/ggml-cuda/solve_tri.cu
@@ -44,6 +44,7 @@ static __global__ void setup_trsm_batch_pointers(
 }
 
 // Latency-optimized kernel for n=64, k=64 (single-token generation)
+#if 0
 static __global__ void solve_tri_f32_64x64_latency(
     const float * __restrict__ A,
     const float * __restrict__ B,
@@ -54,8 +55,7 @@ static __global__ void solve_tri_f32_64x64_latency(
     const size_t nb12,
     const size_t nb13,
     const size_t nb2,
-    const size_t nb3)
-{
+    const size_t nb3) {
     const int batch_idx = blockIdx.x;
     const int lane      = threadIdx.x;
     const int warp_id   = threadIdx.y;
@@ -146,6 +146,7 @@ static __global__ void solve_tri_f32_64x64_latency(
         X_batch[i] = sX[row * 65 + col];
     }
 }
+#endif
 
 static __global__ void solve_tri_f32_64x64_opt(const float * __restrict__ A,
                                                const float * __restrict__ B,
@@ -248,6 +249,7 @@ static __global__ void solve_tri_f32_64x64_opt(const float * __restrict__ A,
     }
 }
 
+#if 0
 static __global__ void solve_tri_f32_128x128_opt(const float * __restrict__ A,
                                                   const float * __restrict__ B,
                                                   float * __restrict__ X,
@@ -360,6 +362,7 @@ static __global__ void solve_tri_f32_128x128_opt(const float * __restrict__ A,
         X_batch[row * k + col] = sXt[col * 129 + row];
     }
 }
+#endif
 
 static __global__ void solve_tri_f32_256x256_tiled(const float * __restrict__ A,
                                                     const float * __restrict__ B,