llama : resolve rwkv conflict

ggml-ci

Author: ggerganov

src/llama.cpp

@@ -7076,19 +7076,13 @@ struct llm_build_context {
     //            1
     //        );
 
+    //        struct ggml_tensor * last_norm_att = ggml_view_3d(ctx0, x_norm_att, n_embd, 1, n_seqs, x_norm_att->nb[1], x_norm_att->nb[2], (n_seq_tokens-1)*n_embd*ggml_element_size(x_norm_att));
     //        ggml_build_forward_expand(
     //            gf,
     //            ggml_cpy(
     //                ctx0,
-    //                wkv_states,
-    //                ggml_view_1d(
-    //                    ctx0,
-    //                    kv_self.v_l[il],
-    //                    hparams.n_embd_v_s() * n_seqs,
-    //                    hparams.n_embd_v_s() * kv_head * ggml_element_size(kv_self.v_l[il])
-    //                )
-    //            )
-    //        );
+    //                ggml_view_1d(ctx0, last_norm_att, n_embd * n_seqs, 0),
+    //                ggml_view_1d(ctx0, kv_self.k_l[il], hparams.n_embd_k_s() * n_seqs, hparams.n_embd_k_s() * kv_head * ggml_element_size(kv_self.k_l[il]))
 
     //        struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, build_rwkv6_time_mix(layer, x_norm_att, x_prev, &wkv_states, hparams.wkv_head_size, hparams.n_head_kv()));
     //        ggml_build_forward_expand(gf, ffn_inp);