Upload builder.py

llava/model/builder.py

@@ -25,6 +25,7 @@ from llava.constants import DEFAULT_IMAGE_PATCH_TOKEN, DEFAULT_IM_START_TOKEN, D
 
 def load_pretrained_model(model_path, model_base, model_name, load_8bit=False, load_4bit=False, device_map="auto", device="cuda"):
     kwargs = {"device_map": device_map}
+    os.makedirs("offload_folder", exist_ok=True)
 
     if load_8bit:
         kwargs['load_in_8bit'] = True
@@ -100,7 +101,7 @@ def load_pretrained_model(model_path, model_base, model_name, load_8bit=False, l
                 model = LlavaMPTForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs)
             else:
                 tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False)
-                model = LlavaLlamaForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs)
+                model = LlavaLlamaForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, offload_folder="offload_folder",**kwargs)
     else:
         # Load language model
         if model_base is not None: