diff --git a/gemma.cc b/gemma.cc
index f2a2275..35a4a47 100644
--- a/gemma.cc
+++ b/gemma.cc
@@ -839,11 +839,6 @@ Gemma::Gemma(const Path& tokenizer_path, const Path& compressed_weights_path,
   }
 }
 
-Gemma::Gemma(const Path& tokenizer_path, const Path& compressed_weights_path,
-             Model model_type, hwy::ThreadPool& pool)
-    : Gemma(tokenizer_path, compressed_weights_path, Path{""}, model_type,
-            pool) {}
-
 Gemma::~Gemma() = default;  // after GemmaInterface is defined
 
 const sentencepiece::SentencePieceProcessor* Gemma::Tokenizer() const {
diff --git a/gemma.h b/gemma.h
index cdd4873..f5e88fa 100644
--- a/gemma.h
+++ b/gemma.h
@@ -76,8 +76,6 @@ struct GemmaInterface;
 struct Gemma {
   Gemma(const Path& tokenizer_path, const Path& compressed_weights_path,
         const Path& weights_path, Model model_type, hwy::ThreadPool& pool);
-  Gemma(const Path& tokenizer_path, const Path& compressed_weights_path,
-        Model model_type, hwy::ThreadPool& pool);
   ~Gemma();  // must be defined after GemmaInterface's dtor is defined.
   const sentencepiece::SentencePieceProcessor* Tokenizer() const;
   std::unique_ptr<GemmaInterface> impl_;
diff --git a/run.cc b/run.cc
index b08e4ca..fcf974b 100644
--- a/run.cc
+++ b/run.cc
@@ -234,7 +234,7 @@ void Run(LoaderArgs& loader, InferenceArgs& inference, AppArgs& app) {
              [](uint64_t /*task*/, size_t thread) { PinThreadToCore(thread); });
   }
 
-  gcpp::Gemma model(loader.tokenizer, loader.compressed_weights,
+  gcpp::Gemma model(loader.tokenizer, loader.compressed_weights, loader.weights,
                     loader.ModelType(), pool);
 
   auto kv_cache = CreateKVCache(loader.ModelType());