fix: tokenizer ref

handler.py

@@ -55,7 +55,7 @@ class EndpointHandler:
         ).convert("RGB")
 
         inputs = self.model.build_conversation_input_ids(
-            tokenizer, query=query, images=[image]
+            self.tokenizer, query=query, images=[image]
         )
         inputs = {
             "input_ids": inputs["input_ids"].unsqueeze(0).to("cuda"),
@@ -68,7 +68,7 @@ class EndpointHandler:
         with torch.no_grad():
             outputs = self.model.generate(**inputs, **gen_kwargs)
             outputs = outputs[:, inputs["input_ids"].shape[1] :]
-            result = tokenizer.decode(outputs[0])
+            result = self.tokenizer.decode(outputs[0])
             return result