From 75beba29b5c7316a9ebde9b0886f609dd5bf05bd Mon Sep 17 00:00:00 2001
From: Fazlul Shahriar <97305712+lpfhs@users.noreply.github.com>
Date: Sat, 8 Jul 2023 18:26:28 -0400
Subject: [PATCH] Don't try to load training_args.bin (#373)

---
 vllm/model_executor/weight_utils.py | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/vllm/model_executor/weight_utils.py b/vllm/model_executor/weight_utils.py
index 94e84a8b19b6..a9d899ad29e1 100644
--- a/vllm/model_executor/weight_utils.py
+++ b/vllm/model_executor/weight_utils.py
@@ -39,7 +39,10 @@ def hf_model_weights_iterator(
     else:
         hf_folder = model_name_or_path
 
-    hf_bin_files = glob.glob(os.path.join(hf_folder, "*.bin"))
+    hf_bin_files = [
+        x for x in glob.glob(os.path.join(hf_folder, "*.bin"))
+        if not x.endswith("training_args.bin")
+    ]
 
     if use_np_cache:
         # Convert the model weights from torch tensors to numpy arrays for