add trust_remote_code = True

builder.py

@@ -71,10 +71,10 @@ def load_pretrained_model(model_path, model_base, model_name, load_8bit=False, l
     if 'llava' in model_name.lower() or 'ferret' in model_name.lower():
         # Load LLaVA/FERRET model
         if 'lora' in model_name.lower() and model_base is not None:
-            lora_cfg_pretrained = AutoConfig.from_pretrained(model_path)
-            tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=False)
+            lora_cfg_pretrained = AutoConfig.from_pretrained(model_path, trust_remote_code=True)
+            tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=False, trust_remote_code=True)
             print('Loading LLaVA/FERRET from base model...')
-            model = AutoModelForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=lora_cfg_pretrained, **kwargs)
+            model = AutoModelForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=lora_cfg_pretrained, **kwargs, trust_remote_code=True)
             token_num, tokem_dim = model.lm_head.out_features, model.lm_head.in_features
             if model.lm_head.weight.shape[0] != token_num:
                 model.lm_head.weight = torch.nn.Parameter(torch.empty(token_num, tokem_dim, device=model.device, dtype=model.dtype))
@@ -100,7 +100,7 @@ def load_pretrained_model(model_path, model_base, model_name, load_8bit=False, l
 
             from peft import PeftModel
             print('Loading LoRA weights...')
-            model = PeftModel.from_pretrained(model, model_path)
+            model = PeftModel.from_pretrained(model, model_path, trust_remote_code=True)
             print('Merging LoRA weights...')
             model = model.merge_and_unload()
             print('Model is loaded...')
@@ -109,31 +109,31 @@ def load_pretrained_model(model_path, model_base, model_name, load_8bit=False, l
             print('Loading LLaVA/FERRET from base model...')
             tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=False)
             cfg_pretrained = AutoConfig.from_pretrained(model_path)
-            model = AutoModelForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=cfg_pretrained, **kwargs)
+            model = AutoModelForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=cfg_pretrained, **kwargs, trust_remote_code=True)
 
             mm_projector_weights = torch.load(os.path.join(model_path, 'mm_projector.bin'), map_location='cpu')
             mm_projector_weights = {k: v.to(torch.float16) for k, v in mm_projector_weights.items()}
             model.load_state_dict(mm_projector_weights, strict=False)
         else:
-            tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False)
-            model = AutoModelForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs)
+            tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False, trust_remote_code=True)
+            model = AutoModelForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs, trust_remote_code=True)
     else:
         # Load language model
         if model_base is not None:
             # PEFT model
             from peft import PeftModel
             tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=False)
-            model = AutoModelForCausalLM.from_pretrained(model_base, torch_dtype=torch.float16, low_cpu_mem_usage=True, device_map="auto")
+            model = AutoModelForCausalLM.from_pretrained(model_base, torch_dtype=torch.float16, low_cpu_mem_usage=True, device_map="auto", trust_remote_code=True)
             print(f"Loading LoRA weights from {model_path}")
-            model = PeftModel.from_pretrained(model, model_path)
+            model = PeftModel.from_pretrained(model, model_path, trust_remote_code=True)
             print(f"Merging weights")
             model = model.merge_and_unload()
             print('Convert to FP16...')
             model.to(torch.float16)
         else:
             use_fast = False
-            tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False)
-            model = AutoModelForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs)
+            tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False, trust_remote_code=True)
+            model = AutoModelForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs, trust_remote_code=True)
 
     image_processor = None