diff --git a/sw/apps/transformer/src/transformer.h b/sw/apps/transformer/src/transformer.h
index 0b00796c4..ddbf70038 100644
--- a/sw/apps/transformer/src/transformer.h
+++ b/sw/apps/transformer/src/transformer.h
@@ -838,13 +838,14 @@ static inline void transformer_layer_fp64(transformer_layer_fp64_t *const l) {
         // double used_memory_kB =  (double)((uint64_t)tcdm_ptr - (uint64_t)snrt_l1_next()) / 1024.0f;
         // dump_debug(used_memory_kB);
 
-        // determine the column offset for the current cluster
+        // determine the column offset of the ifmap for the current cluster
         uint32_t cluster_ifmap_offset = cluster_id * l->seq_len * l->positional_embeddings_fa;
-        if (core_id == 0) {
-            dump_idx(cluster_id);
-            dump_idx(cluster_ifmap_offset);
-        }
-        uint32_t cluster_weights_offset = cluster_id * l->positional_embeddings_fa * B_c_lin2;
+        // if (core_id == 0) {
+        //     dump_idx(cluster_id);
+        //     dump_idx(cluster_ifmap_offset);
+        // }
+        // determine the column offset of the weights for the current cluster
+        uint32_t cluster_weights_offset = cluster_id * l->positional_embeddings_fa * l->embeddings_lin2;
 
         uint32_t start_loop_outer = snrt_mcycle();
         for (int t_r = 0; t_r < T_r_lin2; t_r++) {
@@ -867,12 +868,44 @@ static inline void transformer_layer_fp64(transformer_layer_fp64_t *const l) {
 
                 snrt_dma_wait_all();
 
-                for (int i = 0; i < B_r_lin2 * l->positional_embeddings_fa; i++) {
-                    dump_idx(i + ifmap_offset);
-                    dump_debug(ifmap_lin2[i]);
-                    // printf("ifmap[%d] = %f\n", i, ifmap_lin2[i]);
+                // for (int i = 0; i < B_r_lin2 * l->positional_embeddings_fa; i++) {
+                //     dump_idx(i + ifmap_offset);
+                //     dump_debug(ifmap_lin2[i]);
+                //     // printf("ifmap[%d] = %f\n", i + ifmap_offset, ifmap_lin2[i]);
+                // }
+
+            }
+            uint32_t end_dma = snrt_mcycle();
+
+            snrt_cluster_hw_barrier();
+
+            uint32_t start_loop_inner = snrt_mcycle();
+            for (int t_c = 0; t_c < T_c_lin2; t_c++) {
+                // weights: P x B_c 
+                uint32_t weights_offset = t_c * B_c_lin2 * l->positional_embeddings_fa + cluster_weights_offset;
+                uint32_t start_dma = snrt_mcycle();
+                if (!snrt_is_compute_core()) {
+                    // load the weights tile
+                    snrt_dma_txid_t txid_weights = 
+                        snrt_dma_start_2d(
+                            weights_lin2,                                   /* dst */
+                            l->weights_lin2 + weights_offset,               /* src */
+                            B_c_lin2 * sizeof(double),                      /* size */
+                            B_c_lin2 * sizeof(double),                      /* dst_stride */
+                            l->positional_embeddings_fa * sizeof(double),   /* src_stride */
+                            l->positional_embeddings_fa);                   /* repetitions */
+
+                    snrt_dma_wait_all();
+
+                    for (int i = 0; i < B_c_lin2 * l->positional_embeddings_fa; i++) {
+                        dump_idx(i + weights_offset);
+                        dump_debug(weights_lin2[i]);
+                        // printf("weights[%d] = %f\n", i + weights_offset, weights_lin2[i]);
+                    }
                 }
+                uint32_t end_dma = snrt_mcycle();
 
+                snrt_cluster_hw_barrier();
             }
         }