diff --git a/hnswlib/hnswalg.h b/hnswlib/hnswalg.h
index 8ee24d95..4ebcc2d0 100644
--- a/hnswlib/hnswalg.h
+++ b/hnswlib/hnswalg.h
@@ -20,7 +20,7 @@ namespace hnswlib {
     typedef unsigned int tableint;
     typedef unsigned int linklistsizeint;
 
-    template<typename dist_t, typename data_t>
+    template<typename dist_t>
     class HierarchicalNSW : public AlgorithmInterface<dist_t> {
     public:
 
@@ -600,6 +600,7 @@ namespace hnswlib {
             revSize_ = 1.0 / mult_;
             ef_ = 10;
             for (size_t i = 0; i < cur_element_count; i++) {
+                label_lookup_[getExternalLabel(i)]=i;
                 unsigned int linkListSize;
                 readBinaryPOD(input, linkListSize);
                 if (linkListSize == 0) {
@@ -617,6 +618,7 @@ namespace hnswlib {
             return;
         }
 
+        template<typename data_t>
         std::vector<data_t> getDataByLabel(labeltype label)
         {
           tableint label_c = label_lookup_[label];
@@ -645,6 +647,7 @@ namespace hnswlib {
                     throw std::runtime_error("The number of elements exceeds the specified limit");
                 };
                 cur_c = cur_element_count;
+                label_lookup_[label] = cur_c;  // expected unique, if not will overwrite
                 cur_element_count++;
             }
             std::unique_lock <std::mutex> lock_el(link_list_locks_[cur_c]);
@@ -667,7 +670,7 @@ namespace hnswlib {
             // Initialisation of the data and label
             memcpy(getExternalLabeLp(cur_c), &label, sizeof(labeltype));
             memcpy(getDataByInternalId(cur_c), data_point, data_size_);
-            label_lookup_[label] = cur_c;  // expected unique, if not will overwrite
+
 
             if (curlevel) {
                 linkLists_[cur_c] = (char *) malloc(size_links_per_element_ * curlevel + 1);
diff --git a/python_bindings/bindings.cpp b/python_bindings/bindings.cpp
index 06a049ff..bd782a55 100644
--- a/python_bindings/bindings.cpp
+++ b/python_bindings/bindings.cpp
@@ -97,7 +97,7 @@ class Index {
             throw new std::runtime_error("The index is already initiated.");
         }
         cur_l = 0;
-        appr_alg = new hnswlib::HierarchicalNSW<dist_t, data_t>(l2space, maxElements, M, efConstruction, random_seed);
+        appr_alg = new hnswlib::HierarchicalNSW<dist_t>(l2space, maxElements, M, efConstruction, random_seed);
         index_inited = true;
         ep_added = false;
     }
@@ -119,7 +119,7 @@ class Index {
             std::cerr<<"Warning: Calling load_index for an already inited index. Old index is being deallocated.";
             delete appr_alg;
         }
-        appr_alg = new hnswlib::HierarchicalNSW<dist_t, data_t>(l2space, path_to_index, false, max_elements);
+        appr_alg = new hnswlib::HierarchicalNSW<dist_t>(l2space, path_to_index, false, max_elements);
 		cur_l = appr_alg->cur_element_count;
     }
 	void normalize_vector(float *data, float *norm_array){
@@ -230,7 +230,7 @@ class Index {
 
         std::vector<std::vector<data_t>> data;
         for (auto id : ids) {
-            data.push_back(appr_alg->getDataByLabel(id));
+            data.push_back(appr_alg->template getDataByLabel<data_t>(id));
         }
         return data;
     }
@@ -341,7 +341,7 @@ class Index {
     bool normalize;
     int num_threads_default;
     hnswlib::labeltype cur_l;
-    hnswlib::HierarchicalNSW<dist_t, data_t> *appr_alg;
+    hnswlib::HierarchicalNSW<dist_t> *appr_alg;
     hnswlib::SpaceInterface<float> *l2space;
 
     ~Index() {