Update libra/eval/run_libra.py

libra/eval/run_libra.py

@@ -28,6 +28,7 @@ def load_model(model_path, model_base=None):
     disable_torch_init()
     model_name = get_model_name_from_path(model_path)
     tokenizer, model, image_processor, context_len = load_pretrained_model(model_path, model_base, model_name)
+    model.to("cpu")
     return tokenizer, model, image_processor, context_len
 
 def load_images(image_file):
@@ -92,7 +93,7 @@ def load_images(image_file):
 
     return image
 
-def get_image_tensors(image_path, image_processor, model, device='cuda'):
+def get_image_tensors(image_path, image_processor, model, device='cpu'):
     # Load and preprocess the images
     if isinstance(image_path, str):
         image = []
@@ -144,7 +145,7 @@ def libra_eval(
 ):
     # Model
     disable_torch_init()
-
+    device = "cpu"  
     if libra_model is not None:
         tokenizer, model, image_processor, context_len = libra_model
         model_name = model.config._name_or_path
@@ -171,18 +172,18 @@ def libra_eval(
     conv.append_message(conv.roles[1], None)
     prompt = conv.get_prompt()
 
-    input_ids = tokenizer_image_token(prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors='pt').unsqueeze(0).cuda()
-    attention_mask = torch.ones(input_ids.shape, dtype=torch.long)
+    input_ids = tokenizer_image_token(prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors='pt').unsqueeze(0).to(device)
+    attention_mask = torch.ones(input_ids.shape, dtype=torch.long, device=device)
     pad_token_id = tokenizer.pad_token_id
 
-    image_tensor = get_image_tensors(image_file, image_processor, model)
+    image_tensor = get_image_tensors(image_file, image_processor, model, , device=device)
     
     stop_str = conv.sep if conv.sep_style != SeparatorStyle.TWO else conv.sep2
     keywords = [stop_str]
     stopping_criteria = KeywordsStoppingCriteria(keywords, tokenizer, input_ids)
 
     with torch.inference_mode():
-        torch.cuda.empty_cache()
+        
         if num_beams > 1:
             output_ids = model.generate(
                 input_ids=input_ids,