From 66d6f52670d32f95a25befa75170908e488330ed Mon Sep 17 00:00:00 2001
From: dvres <dvres@fri.uni-lj.si>
Date: Thu, 3 Oct 2024 13:30:31 +0200
Subject: [PATCH] Fixed load_weights for models with untied embeddings

---
 vllm/model_executor/models/opt.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/vllm/model_executor/models/opt.py b/vllm/model_executor/models/opt.py
index 88d2bcb9f0c9..47bc8adc3bc1 100644
--- a/vllm/model_executor/models/opt.py
+++ b/vllm/model_executor/models/opt.py
@@ -353,7 +353,7 @@ def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
         ]
         params_dict = dict(self.named_parameters(remove_duplicate=False))
         for name, loaded_weight in weights:
-            if "lm_head.weight" in name:
+            if "lm_head.weight" in name and self.config.tie_word_embeddings:
                 continue
             if name.startswith("decoder."):
                 name = "model." + name