diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c
index 55a76f8248..2505489b1e 100644
--- a/ggml/src/ggml.c
+++ b/ggml/src/ggml.c
@@ -3015,6 +3015,10 @@ struct ggml_tensor * ggml_mul_mat(
         struct ggml_context * ctx,
         struct ggml_tensor  * a,
         struct ggml_tensor  * b) {
+
+    printf("Up: {%lld, %lld}\n", a->ne[0], a->ne[1]);
+    printf("Cur: {%lld, %lld}\n", b->ne[0], b->ne[1]);
+    
     GGML_ASSERT(ggml_can_mul_mat(a, b));
     GGML_ASSERT(!ggml_is_transposed(a));
 
diff --git a/src/llama-graph.cpp b/src/llama-graph.cpp
index a05de6e585..ae8b150d28 100644
--- a/src/llama-graph.cpp
+++ b/src/llama-graph.cpp
@@ -644,6 +644,8 @@ ggml_tensor * llm_graph_context::build_ffn(
      llm_ffn_op_type   type_op,
    llm_ffn_gate_type   type_gate,
                  int   il) const {
+
+    
     ggml_tensor * tmp = up ? build_lora_mm(up, cur) : cur;
     cb(tmp, "ffn_up", il);
 
@@ -1377,7 +1379,7 @@ ggml_tensor * llm_graph_context::build_attn(
     // TODO: if ubatch.equal_seqs() == true, we can split the three tensors below into ubatch.n_seqs_unq streams
     LLAMA_LOG_INFO("ubatch.equal_seqs() = %d, n_seqs = %d\n", ubatch.equal_seqs(), ubatch.n_seqs);
 
-    assert(!ubatch.equal_seqs());
+    //  sassert(!ubatch.equal_seqs());
 
     ggml_tensor * q = q_cur;
     ggml_tensor * k = k_cur;