diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
index 21b7a1f9b..680888188 100644
--- a/ggml/src/ggml-vulkan/ggml-vulkan.cpp
+++ b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
@@ -252,6 +252,7 @@ struct vk_device_struct {
     vk_pipeline pipeline_rope_neox_f32, pipeline_rope_neox_f16;
     vk_pipeline pipeline_argsort_f32;
     vk_pipeline pipeline_sum_rows_f32;
+    vk_pipeline pipeline_argmax_f32;
     vk_pipeline pipeline_im2col_f32, pipeline_im2col_f32_f16;
     vk_pipeline pipeline_timestep_embedding_f32;
     vk_pipeline pipeline_pool2d_f32;
@@ -2149,6 +2150,8 @@ static void ggml_vk_load_shaders(vk_device& device) {
 
     ggml_vk_create_pipeline(device, device->pipeline_argsort_f32, "argsort_f32", argsort_f32_len, argsort_f32_data, "main", 2, sizeof(vk_op_argsort_push_constants), {1024, 1, 1}, {}, 1);
 
+    ggml_vk_create_pipeline(device, device->pipeline_argmax_f32, "argmax_f32", argmax_f32_len, argmax_f32_data, "main", 2, sizeof(vk_op_push_constants), {1, 1, 1}, { device->subgroup_size }, 1);
+
     ggml_vk_create_pipeline(device, device->pipeline_sum_rows_f32, "sum_rows_f32", sum_rows_f32_len, sum_rows_f32_data, "main", 2, sizeof(vk_op_push_constants), {1, 1, 1}, { device->subgroup_size }, 1);
 
     ggml_vk_create_pipeline(device, device->pipeline_im2col_f32, "im2col_f32", im2col_f32_len, im2col_f32_data, "main", 2, sizeof(vk_op_im2col_push_constants), {512, 1, 1}, { device->subgroup_size }, 1, true);
@@ -5282,6 +5285,11 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
             return ctx->device->pipeline_sum_rows_f32;
         }
         return nullptr;
+    case GGML_OP_ARGMAX:
+        if (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_I32) {
+            return ctx->device->pipeline_argmax_f32;
+        }
+        return nullptr;
     case GGML_OP_IM2COL:
         if (src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32) {
             return ctx->device->pipeline_im2col_f32;
@@ -5545,6 +5553,7 @@ static void ggml_vk_op_f32(ggml_backend_vk_context * ctx, vk_context& subctx, co
     case GGML_OP_RMS_NORM:
     case GGML_OP_SOFT_MAX:
     case GGML_OP_SUM_ROWS:
+    case GGML_OP_ARGMAX:
         {
             const uint32_t nr = ggml_nrows(src0);
             if (nr > 262144) {
@@ -6149,6 +6158,10 @@ static void ggml_vk_sum_rows(ggml_backend_vk_context * ctx, vk_context& subctx,
     ggml_vk_op_f32<vk_op_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_SUM_ROWS, { (uint32_t)src0->ne[0], 0, 0.0f, 0.0f }, dryrun);
 }
 
+static void ggml_vk_argmax(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
+    ggml_vk_op_f32<vk_op_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_ARGMAX, { (uint32_t)src0->ne[0], 0, 0.0f, 0.0f }, dryrun);
+}
+
 static void ggml_vk_im2col(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst, bool dryrun = false) {
     const int32_t s0 = dst->op_params[0];
     const int32_t s1 = dst->op_params[1];
@@ -7040,6 +7053,7 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_tensor * nod
     case GGML_OP_ARGSORT:
     case GGML_OP_SUM:
     case GGML_OP_SUM_ROWS:
+    case GGML_OP_ARGMAX:
     case GGML_OP_IM2COL:
     case GGML_OP_TIMESTEP_EMBEDDING:
     case GGML_OP_POOL_2D:
@@ -7092,6 +7106,7 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_tensor * nod
         case GGML_OP_ARGSORT:
         case GGML_OP_SUM:
         case GGML_OP_SUM_ROWS:
+        case GGML_OP_ARGMAX:
         case GGML_OP_IM2COL:
         case GGML_OP_TIMESTEP_EMBEDDING:
         case GGML_OP_POOL_2D:
@@ -7219,6 +7234,10 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_tensor * nod
     case GGML_OP_SUM_ROWS:
         ggml_vk_sum_rows(ctx, compute_ctx, src0, node, dryrun);
 
+        break;
+    case GGML_OP_ARGMAX:
+        ggml_vk_argmax(ctx, compute_ctx, src0, node, dryrun);
+
         break;
     case GGML_OP_IM2COL:
         ggml_vk_im2col(ctx, compute_ctx, src0, src1, node, dryrun);
@@ -7331,6 +7350,7 @@ static bool ggml_vk_compute_forward(ggml_backend_vk_context * ctx, ggml_tensor *
     case GGML_OP_ARGSORT:
     case GGML_OP_SUM:
     case GGML_OP_SUM_ROWS:
+    case GGML_OP_ARGMAX:
     case GGML_OP_IM2COL:
     case GGML_OP_TIMESTEP_EMBEDDING:
     case GGML_OP_POOL_2D:
@@ -8266,6 +8286,7 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
         case GGML_OP_ARGSORT:
         case GGML_OP_SUM:
         case GGML_OP_SUM_ROWS:
+        case GGML_OP_ARGMAX:
         case GGML_OP_IM2COL:
         case GGML_OP_TIMESTEP_EMBEDDING:
         case GGML_OP_POOL_2D:
@@ -8840,6 +8861,8 @@ static void ggml_vk_check_results_0(ggml_tensor * tensor) {
         tensor_clone = ggml_sum(ggml_ctx, src0_clone);
     } else if (tensor->op == GGML_OP_SUM_ROWS) {
         tensor_clone = ggml_sum_rows(ggml_ctx, src0_clone);
+    } else if (tensor->op == GGML_OP_ARGMAX) {
+        tensor_clone = ggml_argmax(ggml_ctx, src0_clone);
     } else if (tensor->op == GGML_OP_IM2COL) {
         const int32_t s0 = tensor->op_params[0];
         const int32_t s1 = tensor->op_params[1];
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/argmax.comp b/ggml/src/ggml-vulkan/vulkan-shaders/argmax.comp
new file mode 100644
index 000000000..eaf4da341
--- /dev/null
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/argmax.comp
@@ -0,0 +1,51 @@
+#version 450
+
+#include "generic_head.comp"
+#include "types.comp"
+
+#extension GL_EXT_control_flow_attributes : enable
+
+layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
+
+layout (binding = 0) readonly buffer A {A_TYPE data_a[];};
+layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
+
+layout (constant_id = 0) const uint BLOCK_SIZE = 32;
+
+shared FLOAT_TYPE tmpmax[BLOCK_SIZE];
+shared uint tmp[BLOCK_SIZE];
+
+void main() {
+    const uint row = gl_WorkGroupID.z * 262144 + gl_WorkGroupID.y * 512 + gl_WorkGroupID.x;
+    const uint col = gl_LocalInvocationID.x;
+
+    if (col >= p.KX) {
+        return;
+    }
+    A_TYPE amax = data_a[row*p.KX + col];
+    tmp[col] = col;
+
+    for (uint i = col + BLOCK_SIZE; i < p.KX; i += BLOCK_SIZE) {
+        A_TYPE val = data_a[row*p.KX + i];
+        if (val > amax) {
+            amax = val;
+            tmp[col] = i;
+        }
+    }
+    tmpmax[col] = amax;
+
+    barrier();
+    [[unroll]] for (int s = int(BLOCK_SIZE) / 2; s > 0; s >>= 1) {
+        if (col < s && col + s < p.KX) {
+            if (tmpmax[col] < tmpmax[col + s]) {
+                tmpmax[col] = tmpmax[col + s];
+                tmp[col] = tmp[col + s];
+            }
+        }
+        barrier();
+    }
+
+    if (col == 0) {
+        data_d[row] = D_TYPE(tmp[0]);
+    }
+}
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp b/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
index 77e7e1148..3594fef53 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
@@ -484,6 +484,7 @@ void process_shaders() {
 
     string_to_spv("argsort_f32", "argsort.comp", {{"A_TYPE", "float"}});
 
+    string_to_spv("argmax_f32", "argmax.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "int"}}));
     string_to_spv("sum_rows_f32", "sum_rows.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
 
     string_to_spv("im2col_f32", "im2col.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));