diff --git a/examples/demo_qwen.cpp b/examples/demo_qwen.cpp
index ba958662..422540ec 100644
--- a/examples/demo_qwen.cpp
+++ b/examples/demo_qwen.cpp
@@ -19,7 +19,7 @@ int main(int argc, char **argv) {
     cmdline::parser cmdParser;
     cmdParser.add<string>("vocab", 'v', "specify mllm tokenizer model path", false, "../vocab/qwen_vocab.mllm");
     cmdParser.add<string>("merge", 'e', "specify mllm merge file path", false, "../vocab/qwen_merges.txt");
-    cmdParser.add<string>("model", 'm', "specify mllm model path", false, "../models/qwen-1.5-0.5b-q4_k.mllm");
+    cmdParser.add<string>("model", 'm', "specify mllm model path", false, "../models/qwen-1.5-1.8b-q8_0.mllm");
     cmdParser.add<int>("limits", 'l', "max KV cache size", false, 400);
     cmdParser.add<int>("thread", 't', "num of threads", false, 4);
     cmdParser.parse_check(argc, argv);
@@ -31,7 +31,7 @@ int main(int argc, char **argv) {
     CPUBackend::cpu_threads = cmdParser.get<int>("thread");
 
     auto tokenizer = QWenTokenizer(vocab_path, merge_path);
-    QWenConfig config(tokens_limit, "0.5B", RoPEType::HFHUBROPE);
+    QWenConfig config(tokens_limit, "1.8B", RoPEType::HFHUBROPE);
     auto model = QWenForCausalLM(config);
     model.load(model_path);
 
diff --git a/src/backends/cpu/CPULinear.cpp b/src/backends/cpu/CPULinear.cpp
index 4a6680d8..319d1eed 100644
--- a/src/backends/cpu/CPULinear.cpp
+++ b/src/backends/cpu/CPULinear.cpp
@@ -1,5 +1,6 @@
 
 #include "CPULinear.hpp"
+#include <iostream>
 
 namespace mllm {
 
diff --git a/src/quantizer/QuantWriter.cpp b/src/quantizer/QuantWriter.cpp
index 752e9c62..2b01d09d 100644
--- a/src/quantizer/QuantWriter.cpp
+++ b/src/quantizer/QuantWriter.cpp
@@ -69,7 +69,7 @@ void QuantWriter::quantParams(DataType dataType) {
         }
         void *quant_ptr = nullptr;
         std::pair<void *, uint64_t> block_t;
-        if (find_names(name, q6_layers)) {
+        if (find_names(name, q6_layers) && (dataType== MLLM_TYPE_Q6_K ||dataType == MLLM_TYPE_Q4_K)) {
             if(tmp_hidden_dim>0 && (size/tmp_hidden_dim)%256!=0){
                 std::cout << "Quantize param " << name << " to " << DataTypeName(MLLM_TYPE_F32) << "\t";
                 const auto s = param_loader_->offsets_[name].second / sizeof(float);