Add adapter config path within qwenvl2

node-hub/dora-qwenvl/dora_qwenvl/main.py

@@ -12,6 +12,7 @@ DEFAULT_QUESTION = os.getenv(
     "DEFAULT_QUESTION",
     "Describe this image",
 )
+ADAPTER_PATH = os.getenv("ADAPTER_PATH", "")
 
 # Check if flash_attn is installed
 try:
@@ -23,7 +24,7 @@ try:
         device_map="auto",
         attn_implementation="flash_attention_2",
     )
-except ImportError:
+except ImportError or ModuleNotFoundError:
     model = Qwen2VLForConditionalGeneration.from_pretrained(
         CUSTOM_MODEL_PATH,
         torch_dtype="auto",
@@ -31,8 +32,12 @@ except ImportError:
     )
 
 
+if ADAPTER_PATH != "":
+    model.load_adapter(ADAPTER_PATH, "dora")
+
+
 # default processor
-processor = AutoProcessor.from_pretrained(DEFAULT_PATH)
+processor = AutoProcessor.from_pretrained(CUSTOM_MODEL_PATH)
 
 
 def generate(frames: dict, question):