Update serve/builder.py

serve/builder.py

@@ -37,8 +37,8 @@ def load_pretrained_model(model_path, model_base, model_name, model_type, load_8
 
         print('Loading nanoLLaVA from base model...')
         if model_type == 'qwen1.5-1.8b' or model_type == 'qwen1.5-0.5b':
-            tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True)
-            model = AutoModelForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=lora_cfg_pretrained,
+            tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True, trust_remote_code=True)
+            model = AutoModelForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=lora_cfg_pretrained, trust_remote_code=True,
                                                          **kwargs)
 
         token_num, tokem_dim = model.lm_head.out_features, model.lm_head.in_features
@@ -82,8 +82,8 @@ def load_pretrained_model(model_path, model_base, model_name, model_type, load_8
 
         cfg_pretrained = AutoConfig.from_pretrained(model_path)
         if model_type == 'qwen1.5-1.8b' or model_type == 'qwen1.5-0.5b':
-            tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True)
-            model = AutoModelForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=cfg_pretrained,
+            tokenizer = AutoTokenizer.from_pretrained(model_base, use_fast=True, trust_remote_code=True)
+            model = AutoModelForCausalLM.from_pretrained(model_base, low_cpu_mem_usage=True, config=cfg_pretrained, trust_remote_code=True,
                                                          **kwargs)
 
         mm_projector_weights = torch.load(os.path.join(model_path, 'mm_projector.bin'), map_location='cpu')
@@ -91,8 +91,8 @@ def load_pretrained_model(model_path, model_base, model_name, model_type, load_8
         model.load_state_dict(mm_projector_weights, strict=False)
     else:
         if model_type == 'qwen1.5-1.8b' or model_type == 'qwen1.5-0.5b':
-            tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=True)
-            model = AutoModelForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs)
+            tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=True, trust_remote_code=True)
+            model = AutoModelForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, trust_remote_code=True, **kwargs)
 
     model.resize_token_embeddings(len(tokenizer))