diff --git a/backends/qualcomm/utils/utils.py b/backends/qualcomm/utils/utils.py
index e13705b3a8f..a53f8e39071 100644
--- a/backends/qualcomm/utils/utils.py
+++ b/backends/qualcomm/utils/utils.py
@@ -823,6 +823,7 @@ def generate_multi_graph_program(
     )
     assert qnn_mgr.Init().value == 0, "failed to load processed bytes"
     binary_info = bytes(qnn_mgr.Compile())
+    print("Checking the size of QNN binary info: ", len(binary_info))
     assert len(binary_info) != 0, "failed to generate QNN context binary"
     graph_names = qnn_mgr.GetGraphNames()
     for graph_name in graph_names:
diff --git a/examples/qualcomm/oss_scripts/llama/llama.py b/examples/qualcomm/oss_scripts/llama/llama.py
index 0af0f55b88f..cbcb3b0c04d 100755
--- a/examples/qualcomm/oss_scripts/llama/llama.py
+++ b/examples/qualcomm/oss_scripts/llama/llama.py
@@ -494,7 +494,8 @@ def compile(args, pte_filename, tokenizer):
                 annotate_linear_16a8w_in_affine_layer,
             )
         if args.ptq != None:
-            kv_quant_attrs = {}
+            import hashlib
+            kv_quant_attrs, parameter_hash = {}, []
             for i, llama_instance in enumerate(llama_instance_list):
                 llama_instance.quantize(
                     quant_dtype=quant_dtype,
@@ -517,6 +518,31 @@ def compile(args, pte_filename, tokenizer):
                             kv_quant_attrs=kv_quant_attrs,
                         ),
                     )
+                    
+                tensor_to_md5 = {}
+                for name, buffer in llama_instance.llama_model.named_buffers():
+                    md5_buffer = hashlib.md5(buffer.numpy().tobytes()).hexdigest()
+                    if md5_buffer in tensor_to_md5:
+                        tensor_to_md5[md5_buffer].append(name)
+                    else:
+                        tensor_to_md5[md5_buffer] = [name]
+                parameter_hash.append(tensor_to_md5)
+
+        # check tensors in prefill & decode are exactly the same
+        assert len(parameter_hash[0]) == len(parameter_hash[1])
+        num_keys = len(parameter_hash[0])
+        # Remove common keys from both dictionaries
+        for key in set(parameter_hash[0]).intersection(set(parameter_hash[1])):
+            del parameter_hash[0][key]
+            del parameter_hash[1][key]
+        print(f"{num_keys - len(parameter_hash[0])} / {num_keys}  tensors are matched")
+            
+        for buf, name in parameter_hash[0].items(): # kv
+            print(f"KV buffers: {name} cannot find a match")
+        for buf, name in parameter_hash[1].items(): # prefill
+            print(f"Prefill buffers: {name} cannot find a match")
+      
+
         end_quantize_ts = time.time()
         logging.info(f"Time for quantizing: {end_quantize_ts - start_quantize_ts}")
 
diff --git a/examples/qualcomm/oss_scripts/llama/runner/runner.cpp b/examples/qualcomm/oss_scripts/llama/runner/runner.cpp
index e06d52fbb37..99d6d715db7 100644
--- a/examples/qualcomm/oss_scripts/llama/runner/runner.cpp
+++ b/examples/qualcomm/oss_scripts/llama/runner/runner.cpp
@@ -404,10 +404,10 @@ Error Runner::generate(
         token_callback(piece_res.get().c_str());
       }
 
-      if (pos >= num_prompt_tokens && eos_id_.count(cur_token) > 0) {
-        ET_LOG(Info, "\nReached to the end of generation");
-        break;
-      }
+      // if (pos >= num_prompt_tokens && eos_id_.count(cur_token) > 0) {
+      //   ET_LOG(Info, "\nReached to the end of generation");
+      //   break;
+      // }
     }
   };
 
diff --git a/examples/qualcomm/utils.py b/examples/qualcomm/utils.py
index c2d2f002aa8..a1890022be4 100755
--- a/examples/qualcomm/utils.py
+++ b/examples/qualcomm/utils.py
@@ -167,6 +167,8 @@ def execute(self, custom_runner_cmd=None, method_index=0):
             )
         else:
             qnn_executor_runner_cmds = custom_runner_cmd
+            
+        print("Execution command is: ", qnn_executor_runner_cmds)
 
         self._adb(["shell", f"{qnn_executor_runner_cmds}"])